{"id":15093,"date":"2024-01-08T14:01:55","date_gmt":"2024-01-08T10:31:55","guid":{"rendered":"https:\/\/rasanegar.com\/blog\/%db%8c%d8%a7%d8%af%da%af%db%8c%d8%b1%db%8c-%d9%85%d8%a7%d8%b4%db%8c%d9%86%db%8c-%d8%aa%d8%b7%d8%a8%db%8c%d9%82-%d8%a8%db%8c%d8%b4-%d8%a7%d8%b2-%d8%ad%d8%af-%d8%af%d9%88%d8%b3%d8%aa-%d8%b4%d9%85%d8%a7\/"},"modified":"2024-01-08T14:01:55","modified_gmt":"2024-01-08T10:31:55","slug":"%db%8c%d8%a7%d8%af%da%af%db%8c%d8%b1%db%8c-%d9%85%d8%a7%d8%b4%db%8c%d9%86%db%8c-%d8%aa%d8%b7%d8%a8%db%8c%d9%82-%d8%a8%db%8c%d8%b4-%d8%a7%d8%b2-%d8%ad%d8%af-%d8%af%d9%88%d8%b3%d8%aa-%d8%b4%d9%85%d8%a7","status":"publish","type":"post","link":"https:\/\/rasanegaar.com\/blog\/%db%8c%d8%a7%d8%af%da%af%db%8c%d8%b1%db%8c-%d9%85%d8%a7%d8%b4%db%8c%d9%86%db%8c-%d8%aa%d8%b7%d8%a8%db%8c%d9%82-%d8%a8%db%8c%d8%b4-%d8%a7%d8%b2-%d8%ad%d8%af-%d8%af%d9%88%d8%b3%d8%aa-%d8%b4%d9%85%d8%a7\/","title":{"rendered":"\u06cc\u0627\u062f\u06af\u06cc\u0631\u06cc \u0645\u0627\u0634\u06cc\u0646\u06cc: \u062a\u0637\u0628\u06cc\u0642 \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u062f\u0648\u0633\u062a \u0634\u0645\u0627\u0633\u062a\u060c \u0646\u0647 \u062f\u0634\u0645\u0646 \u0634\u0645\u0627"},"content":{"rendered":"<div id=\"ez-toc-container\" class=\"ez-toc-v2_0_85 counter-hierarchy ez-toc-counter ez-toc-custom ez-toc-container-direction\">\n<div class=\"ez-toc-title-container\"><p class=\"ez-toc-title\" style=\"cursor:inherit\">\u0633\u0631\u0641\u0635\u0644\u0647\u0627\u06cc \u0645\u0637\u0644\u0628<\/p>\n<\/div><nav><ul class='ez-toc-list ez-toc-list-level-1 ' ><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-1\" href=\"https:\/\/rasanegaar.com\/blog\/%db%8c%d8%a7%d8%af%da%af%db%8c%d8%b1%db%8c-%d9%85%d8%a7%d8%b4%db%8c%d9%86%db%8c-%d8%aa%d8%b7%d8%a8%db%8c%d9%82-%d8%a8%db%8c%d8%b4-%d8%a7%d8%b2-%d8%ad%d8%af-%d8%af%d9%88%d8%b3%d8%aa-%d8%b4%d9%85%d8%a7\/#%d8%aa%d8%b7%d8%a8%db%8c%d9%82_%d8%a8%db%8c%d8%b4_%d8%a7%d8%b2_%d8%ad%d8%af_%d8%af%d8%b1_%d9%85%d9%82%d8%a7%d8%a8%d9%84_%d8%af%d8%a7%d9%86%d8%b4%d9%85%d9%86%d8%af%d8%a7%d9%86_%d8%af%d8%a7%d8%af%d9%87\" >\u062a\u0637\u0628\u06cc\u0642 \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u062f\u0631 \u0645\u0642\u0627\u0628\u0644 \u062f\u0627\u0646\u0634\u0645\u0646\u062f\u0627\u0646 \u062f\u0627\u062f\u0647<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-2\" href=\"https:\/\/rasanegaar.com\/blog\/%db%8c%d8%a7%d8%af%da%af%db%8c%d8%b1%db%8c-%d9%85%d8%a7%d8%b4%db%8c%d9%86%db%8c-%d8%aa%d8%b7%d8%a8%db%8c%d9%82-%d8%a8%db%8c%d8%b4-%d8%a7%d8%b2-%d8%ad%d8%af-%d8%af%d9%88%d8%b3%d8%aa-%d8%b4%d9%85%d8%a7\/#%da%86%d9%82%d8%af%d8%b1_%d8%a8%db%8c%d8%b4_%d8%a7%d8%b2_%d8%ad%d8%af_%d8%a8%d8%af_%d9%86%db%8c%d8%b3%d8%aa_%d8%a8%d9%87_%d9%87%d9%85%d8%a7%d9%86_%d8%a7%d9%86%d8%af%d8%a7%d8%b2%d9%87_%da%a9%d9%87_%d8%b3%d8%a7%d8%ae%d8%aa%d9%87_%d8%b4%d8%af%d9%87_%d8%a7%d8%b3%d8%aa\" >\u0686\u0642\u062f\u0631 \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0628\u062f \u0646\u06cc\u0633\u062a \u0628\u0647 \u0647\u0645\u0627\u0646 \u0627\u0646\u062f\u0627\u0632\u0647 \u06a9\u0647 \u0633\u0627\u062e\u062a\u0647 \u0634\u062f\u0647 \u0627\u0633\u062a<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-3\" href=\"https:\/\/rasanegaar.com\/blog\/%db%8c%d8%a7%d8%af%da%af%db%8c%d8%b1%db%8c-%d9%85%d8%a7%d8%b4%db%8c%d9%86%db%8c-%d8%aa%d8%b7%d8%a8%db%8c%d9%82-%d8%a8%db%8c%d8%b4-%d8%a7%d8%b2-%d8%ad%d8%af-%d8%af%d9%88%d8%b3%d8%aa-%d8%b4%d9%85%d8%a7\/#%d9%85%d8%b7%d8%a7%d9%84%d8%b9%d9%87_%d9%85%d9%88%d8%b1%d8%af%db%8c_%e2%80%93_%d8%a7%d8%b3%d8%aa%d8%af%d9%84%d8%a7%d9%84_%d8%a8%db%8c%d8%b4_%d8%a7%d8%b2_%d8%ad%d8%af_%d8%af%d9%88%d8%b3%d8%aa%d8%a7%d9%86%d9%87\" >\u0645\u0637\u0627\u0644\u0639\u0647 \u0645\u0648\u0631\u062f\u06cc &#8211; \u0627\u0633\u062a\u062f\u0644\u0627\u0644 \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u062f\u0648\u0633\u062a\u0627\u0646\u0647<\/a><ul class='ez-toc-list-level-3' ><li class='ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-4\" href=\"https:\/\/rasanegaar.com\/blog\/%db%8c%d8%a7%d8%af%da%af%db%8c%d8%b1%db%8c-%d9%85%d8%a7%d8%b4%db%8c%d9%86%db%8c-%d8%aa%d8%b7%d8%a8%db%8c%d9%82-%d8%a8%db%8c%d8%b4-%d8%a7%d8%b2-%d8%ad%d8%af-%d8%af%d9%88%d8%b3%d8%aa-%d8%b4%d9%85%d8%a7\/#%d9%85%d8%ac%d9%85%d9%88%d8%b9%d9%87_%d8%af%d8%a7%d8%af%d9%87_%d9%87%d8%a7%db%8c_cifar10_%d9%88_cifar100\" >\u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u0647\u0627\u06cc CIFAR10 \u0648 CIFAR100<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-5\" href=\"https:\/\/rasanegaar.com\/blog\/%db%8c%d8%a7%d8%af%da%af%db%8c%d8%b1%db%8c-%d9%85%d8%a7%d8%b4%db%8c%d9%86%db%8c-%d8%aa%d8%b7%d8%a8%db%8c%d9%82-%d8%a8%db%8c%d8%b4-%d8%a7%d8%b2-%d8%ad%d8%af-%d8%af%d9%88%d8%b3%d8%aa-%d8%b4%d9%85%d8%a7\/#%d9%88%d8%a7%d8%b1%d8%af_%da%a9%d8%b1%d8%af%d9%86_%d8%af%d8%a7%d8%af%d9%87_%d9%87%d8%a7\" >\u0648\u0627\u0631\u062f \u06a9\u0631\u062f\u0646 \u062f\u0627\u062f\u0647 \u0647\u0627<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-6\" href=\"https:\/\/rasanegaar.com\/blog\/%db%8c%d8%a7%d8%af%da%af%db%8c%d8%b1%db%8c-%d9%85%d8%a7%d8%b4%db%8c%d9%86%db%8c-%d8%aa%d8%b7%d8%a8%db%8c%d9%82-%d8%a8%db%8c%d8%b4-%d8%a7%d8%b2-%d8%ad%d8%af-%d8%af%d9%88%d8%b3%d8%aa-%d8%b4%d9%85%d8%a7\/#%d8%b9%d8%af%d9%85_%d8%aa%d9%86%d8%a7%d8%b3%d8%a8_%d9%be%d8%b1%d8%b3%d9%be%d8%aa%d8%b1%d9%88%d9%86_%da%86%d9%86%d8%af%d9%84%d8%a7%db%8c%d9%87\" >\u0639\u062f\u0645 \u062a\u0646\u0627\u0633\u0628 \u067e\u0631\u0633\u067e\u062a\u0631\u0648\u0646 \u0686\u0646\u062f\u0644\u0627\u06cc\u0647<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-7\" href=\"https:\/\/rasanegaar.com\/blog\/%db%8c%d8%a7%d8%af%da%af%db%8c%d8%b1%db%8c-%d9%85%d8%a7%d8%b4%db%8c%d9%86%db%8c-%d8%aa%d8%b7%d8%a8%db%8c%d9%82-%d8%a8%db%8c%d8%b4-%d8%a7%d8%b2-%d8%ad%d8%af-%d8%af%d9%88%d8%b3%d8%aa-%d8%b4%d9%85%d8%a7\/#%d8%a8%db%8c%d8%b4_%d8%a7%d8%b2_%d8%ad%d8%af_%d8%a8%d8%b1%d8%a7%d8%b2%d8%b4_%d8%b4%d8%a8%da%a9%d9%87_%d8%b9%d8%b5%d8%a8%db%8c_%da%a9%d8%a7%d9%86%d9%88%d9%84%d9%88%d8%b4%d9%86%d8%a7%d9%84_%d8%b1%d9%88%db%8c_cifar10\" >\u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0628\u0631\u0627\u0632\u0634 \u0634\u0628\u06a9\u0647 \u0639\u0635\u0628\u06cc \u06a9\u0627\u0646\u0648\u0644\u0648\u0634\u0646\u0627\u0644 \u0631\u0648\u06cc CIFAR10<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-8\" href=\"https:\/\/rasanegaar.com\/blog\/%db%8c%d8%a7%d8%af%da%af%db%8c%d8%b1%db%8c-%d9%85%d8%a7%d8%b4%db%8c%d9%86%db%8c-%d8%aa%d8%b7%d8%a8%db%8c%d9%82-%d8%a8%db%8c%d8%b4-%d8%a7%d8%b2-%d8%ad%d8%af-%d8%af%d9%88%d8%b3%d8%aa-%d8%b4%d9%85%d8%a7\/#%d8%b3%d8%a7%d8%af%d9%87_%d8%b3%d8%a7%d8%b2%db%8c_%d8%b4%d8%a8%da%a9%d9%87_%d8%b9%d8%b5%d8%a8%db%8c_%da%a9%d8%a7%d9%86%d9%88%d9%84%d9%88%d8%b4%d9%86%d8%a7%d9%84_%d8%b1%d9%88%db%8c_cifar10\" >\u0633\u0627\u062f\u0647 \u0633\u0627\u0632\u06cc \u0634\u0628\u06a9\u0647 \u0639\u0635\u0628\u06cc \u06a9\u0627\u0646\u0648\u0644\u0648\u0634\u0646\u0627\u0644 \u0631\u0648\u06cc CIFAR10<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-9\" href=\"https:\/\/rasanegaar.com\/blog\/%db%8c%d8%a7%d8%af%da%af%db%8c%d8%b1%db%8c-%d9%85%d8%a7%d8%b4%db%8c%d9%86%db%8c-%d8%aa%d8%b7%d8%a8%db%8c%d9%82-%d8%a8%db%8c%d8%b4-%d8%a7%d8%b2-%d8%ad%d8%af-%d8%af%d9%88%d8%b3%d8%aa-%d8%b4%d9%85%d8%a7\/#%d8%a8%db%8c%d8%b4_%d8%a7%d8%b2_%d8%ad%d8%af_%d8%a8%d8%b1%d8%a7%d8%b2%d8%b4_%d8%b4%d8%a8%da%a9%d9%87_%d8%b9%d8%b5%d8%a8%db%8c_%da%a9%d8%a7%d9%86%d9%88%d9%84%d9%88%d8%b4%d9%86_%d8%b1%d9%88%db%8c_cifar100\" >\u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0628\u0631\u0627\u0632\u0634 \u0634\u0628\u06a9\u0647 \u0639\u0635\u0628\u06cc \u06a9\u0627\u0646\u0648\u0644\u0648\u0634\u0646 \u0631\u0648\u06cc CIFAR100<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-10\" href=\"https:\/\/rasanegaar.com\/blog\/%db%8c%d8%a7%d8%af%da%af%db%8c%d8%b1%db%8c-%d9%85%d8%a7%d8%b4%db%8c%d9%86%db%8c-%d8%aa%d8%b7%d8%a8%db%8c%d9%82-%d8%a8%db%8c%d8%b4-%d8%a7%d8%b2-%d8%ad%d8%af-%d8%af%d9%88%d8%b3%d8%aa-%d8%b4%d9%85%d8%a7\/#%d8%b9%d8%af%d9%85_%d8%aa%d8%b9%d9%85%db%8c%d9%85_%d8%a8%d8%b9%d8%af_%d8%a7%d8%b2_%d8%b3%d8%a7%d8%af%d9%87_%d8%b3%d8%a7%d8%b2%db%8c\" >\u0639\u062f\u0645 \u062a\u0639\u0645\u06cc\u0645 \u0628\u0639\u062f \u0627\u0632 \u0633\u0627\u062f\u0647 \u0633\u0627\u0632\u06cc<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-11\" href=\"https:\/\/rasanegaar.com\/blog\/%db%8c%d8%a7%d8%af%da%af%db%8c%d8%b1%db%8c-%d9%85%d8%a7%d8%b4%db%8c%d9%86%db%8c-%d8%aa%d8%b7%d8%a8%db%8c%d9%82-%d8%a8%db%8c%d8%b4-%d8%a7%d8%b2-%d8%ad%d8%af-%d8%af%d9%88%d8%b3%d8%aa-%d8%b4%d9%85%d8%a7\/#%d8%a7%d9%81%d8%b2%d8%a7%db%8c%d8%b4_%d8%af%d8%a7%d8%af%d9%87_%d9%87%d8%a7_%d8%a8%d8%a7_%da%a9%d9%84%d8%a7%d8%b3_imagedatagenerator_keras\" >\u0627\u0641\u0632\u0627\u06cc\u0634 \u062f\u0627\u062f\u0647 \u0647\u0627 \u0628\u0627 \u06a9\u0644\u0627\u0633 ImageDataGenerator Keras<\/a><\/li><\/ul><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-12\" href=\"https:\/\/rasanegaar.com\/blog\/%db%8c%d8%a7%d8%af%da%af%db%8c%d8%b1%db%8c-%d9%85%d8%a7%d8%b4%db%8c%d9%86%db%8c-%d8%aa%d8%b7%d8%a8%db%8c%d9%82-%d8%a8%db%8c%d8%b4-%d8%a7%d8%b2-%d8%ad%d8%af-%d8%af%d9%88%d8%b3%d8%aa-%d8%b4%d9%85%d8%a7\/#%d9%86%d8%aa%db%8c%d8%ac%d9%87%d8%9f\" >\u0646\u062a\u06cc\u062c\u0647\u061f<\/a><\/li><\/ul><\/nav><\/div>\n<span class=\"span-reading-time rt-reading-time\" style=\"display: block;\"><span class=\"rt-label rt-prefix\">\u0632\u0645\u0627\u0646 \u0644\u0627\u0632\u0645 \u0628\u0631\u0627\u06cc \u0645\u0637\u0627\u0644\u0639\u0647: <\/span> <span class=\"rt-time\"> 18<\/span> <span class=\"rt-label rt-postfix\">\u062f\u0642\u06cc\u0642\u0647<\/span><\/span><p> <br \/>\n<\/p>\n<div><noscript><\/noscript><\/p>\n<p>\u0627\u062c\u0627\u0632\u0647 \u062f\u0647\u06cc\u062f \u0639\u0646\u0648\u0627\u0646 \u0628\u0627\u0644\u0642\u0648\u0647 \u062a\u062d\u0631\u06cc\u06a9 \u0622\u0645\u06cc\u0632 \u0631\u0627 \u0628\u0627 \u0627\u06cc\u0646 \u0645\u0642\u062f\u0645\u0647 \u0628\u06cc\u0627\u0646 \u06a9\u0646\u0645:<\/p>\n<blockquote>\n<p>\u062f\u0631\u0633\u062a \u0627\u0633\u062a\u060c \u0647\u06cc\u0686 \u06a9\u0633 \u0646\u0645\u06cc \u062e\u0648\u0627\u0647\u062f <strong>\u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f<\/strong> \u0645\u062f\u0644\u200c\u0647\u0627\u06cc \u0646\u0647\u0627\u06cc\u06cc\u060c \u062f\u0631\u0633\u062a \u0645\u062b\u0644 \u0686\u06cc\u0632\u06cc \u06a9\u0647 \u0647\u06cc\u0686\u200c\u06a9\u0633 \u0646\u0645\u06cc\u200c\u062e\u0648\u0627\u0647\u062f <strong>\u06a9\u0645 \u062a\u0646\u0627\u0633\u0628<\/strong> \u0645\u062f\u0644 \u0647\u0627\u06cc \u067e\u0627\u06cc\u0627\u0646\u06cc<\/p>\n<\/blockquote>\n<p><strong>\u0645\u062f\u0644 \u0647\u0627\u06cc Overfit<\/strong> \u0639\u0645\u0644\u06a9\u0631\u062f \u0639\u0627\u0644\u06cc \u0631\u0648\u06cc \u062f\u0627\u062f\u0647 \u0647\u0627\u06cc \u0622\u0645\u0648\u0632\u0634\u06cc\u060c \u0627\u0645\u0627 \u0646\u0645\u06cc \u062a\u0648\u0627\u0646\u0646\u062f \u0628\u0647 \u062e\u0648\u0628\u06cc \u0628\u0647 \u0646\u0645\u0648\u0646\u0647 \u0647\u0627\u06cc \u062c\u062f\u06cc\u062f \u062a\u0639\u0645\u06cc\u0645 \u062f\u0647\u0646\u062f.  \u0686\u06cc\u0632\u06cc \u06a9\u0647 \u062f\u0631 \u0646\u0647\u0627\u06cc\u062a \u0628\u0647 \u0622\u0646 \u062f\u0633\u062a \u0645\u06cc\u200c\u06cc\u0627\u0628\u06cc\u062f\u060c \u0645\u062f\u0644\u06cc \u0627\u0633\u062a \u06a9\u0647 \u0628\u0647 \u06cc\u06a9 \u0645\u062f\u0644 \u06a9\u0627\u0645\u0644\u0627\u064b \u06a9\u062f\u06af\u0630\u0627\u0631\u06cc \u0634\u062f\u0647 \u0648 \u0645\u062a\u0646\u0627\u0633\u0628 \u0628\u0627 \u06cc\u06a9 \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u062e\u0627\u0635 \u0646\u0632\u062f\u06cc\u06a9 \u0645\u06cc\u200c\u0634\u0648\u062f.<\/p>\n<p><strong>\u0645\u062f\u0644 \u0647\u0627\u06cc Underfit<\/strong> \u0646\u0645\u06cc \u062a\u0648\u0627\u0646\u0646\u062f \u0628\u0647 \u062f\u0627\u062f\u0647 \u0647\u0627\u06cc \u062c\u062f\u06cc\u062f \u062a\u0639\u0645\u06cc\u0645 \u062f\u0647\u0646\u062f\u060c \u0627\u0645\u0627 \u0646\u0645\u06cc \u062a\u0648\u0627\u0646\u0646\u062f \u0645\u062c\u0645\u0648\u0639\u0647 \u0622\u0645\u0648\u0632\u0634\u06cc \u0627\u0635\u0644\u06cc \u0631\u0627 \u0646\u06cc\u0632 \u0645\u062f\u0644 \u06a9\u0646\u0646\u062f.<\/p>\n<p>\u0627\u06cc\u0646 <strong>\u0645\u062f\u0644 \u062f\u0631\u0633\u062a<\/strong> \u062f\u0627\u062f\u0647 \u0627\u06cc \u0627\u0633\u062a \u06a9\u0647 \u0628\u0647 \u06af\u0648\u0646\u0647 \u0627\u06cc \u0628\u0627 \u062f\u0627\u062f\u0647 \u0647\u0627 \u0645\u0637\u0627\u0628\u0642\u062a \u062f\u0627\u0631\u062f \u06a9\u0647 \u0645\u0642\u0627\u062f\u06cc\u0631 \u067e\u06cc\u0634 \u0628\u06cc\u0646\u06cc \u06a9\u0646\u0646\u062f\u0647 \u062e\u0648\u0628\u06cc \u0631\u0627 \u062f\u0631 \u0645\u062c\u0645\u0648\u0639\u0647 \u0622\u0645\u0648\u0632\u0634\u060c \u0627\u0639\u062a\u0628\u0627\u0631 \u0633\u0646\u062c\u06cc \u0648 \u062a\u0633\u062a \u0648 \u0647\u0645\u0686\u0646\u06cc\u0646 \u0646\u0645\u0648\u0646\u0647 \u0647\u0627\u06cc \u062c\u062f\u06cc\u062f \u0627\u0646\u062c\u0627\u0645 \u062f\u0647\u062f.<\/p>\n<h2 id=\"overfittingvsdatascientists\"><span class=\"ez-toc-section\" id=\"%d8%aa%d8%b7%d8%a8%db%8c%d9%82_%d8%a8%db%8c%d8%b4_%d8%a7%d8%b2_%d8%ad%d8%af_%d8%af%d8%b1_%d9%85%d9%82%d8%a7%d8%a8%d9%84_%d8%af%d8%a7%d9%86%d8%b4%d9%85%d9%86%d8%af%d8%a7%d9%86_%d8%af%d8%a7%d8%af%d9%87\"><\/span>\u062a\u0637\u0628\u06cc\u0642 \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u062f\u0631 \u0645\u0642\u0627\u0628\u0644 \u062f\u0627\u0646\u0634\u0645\u0646\u062f\u0627\u0646 \u062f\u0627\u062f\u0647<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>\u0645\u0628\u0627\u0631\u0632\u0647 \u0628\u0627 \u0628\u06cc\u0634\u200c\u0628\u0631\u0627\u0632\u0634\u200c\u0647\u0627 \u0645\u0648\u0631\u062f \u062a\u0648\u062c\u0647 \u0642\u0631\u0627\u0631 \u06af\u0631\u0641\u062a\u0647 \u0627\u0633\u062a\u060c \u0632\u06cc\u0631\u0627 \u0628\u0631\u0627\u06cc \u06cc\u06a9 \u062a\u0627\u0632\u0647\u200c\u06a9\u0627\u0631 \u0648\u0633\u0648\u0633\u0647\u200c\u0627\u0646\u06af\u06cc\u0632\u062a\u0631 \u0648 \u0648\u0633\u0648\u0633\u0647\u200c\u0627\u0646\u06af\u06cc\u0632\u062a\u0631 \u0627\u0633\u062a \u06a9\u0647 \u0648\u0642\u062a\u06cc \u0633\u0641\u0631 \u06cc\u0627\u062f\u06af\u06cc\u0631\u06cc \u0645\u0627\u0634\u06cc\u0646\u06cc \u062e\u0648\u062f \u0631\u0627 \u0634\u0631\u0648\u0639 \u0645\u06cc\u200c\u06a9\u0646\u062f\u060c \u0645\u062f\u0644\u200c\u0647\u0627\u06cc \u0627\u0636\u0627\u0641\u0647\u200c\u0641\u06cc\u062a \u0627\u06cc\u062c\u0627\u062f \u06a9\u0646\u062f.  \u062f\u0631 \u0633\u0631\u0627\u0633\u0631 \u06a9\u062a\u0627\u0628\u200c\u0647\u0627\u060c \u067e\u0633\u062a\u200c\u0647\u0627\u06cc \u0648\u0628\u0644\u0627\u06af \u0648 \u062f\u0648\u0631\u0647\u200c\u0647\u0627\u060c \u06cc\u06a9 \u0633\u0646\u0627\u0631\u06cc\u0648\u06cc \u0645\u0634\u062a\u0631\u06a9 \u0627\u0631\u0627\u0626\u0647 \u0645\u06cc\u200c\u0634\u0648\u062f:<\/p>\n<blockquote>\n<p>&#8220;\u0627\u06cc\u0646 \u0645\u062f\u0644 \u062f\u0627\u0631\u0627\u06cc \u06cc\u06a9 <strong>\u0645\u06cc\u0632\u0627\u0646 \u062f\u0642\u062a 100%!<\/strong> \u0639\u0627\u0644\u06cc\u0647!  \u06cc\u0627 \u0646\u0647.  \u062f\u0631 \u0648\u0627\u0642\u0639\u060c \u0628\u0647 \u0634\u062f\u062a \u0628\u0647 \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647\u200c\u0647\u0627 \u0648 \u0647\u0646\u06af\u0627\u0645 \u0622\u0632\u0645\u0627\u06cc\u0634 \u0622\u0646 \u062a\u0646\u0627\u0633\u0628 \u062f\u0627\u0631\u062f \u0631\u0648\u06cc \u0646\u0645\u0648\u0646\u0647 \u0647\u0627\u06cc \u062c\u062f\u06cc\u062f\u060c \u0622\u0646 \u0631\u0627 \u0627\u0646\u062c\u0627\u0645 \u0645\u06cc \u062f\u0647\u062f \u0628\u0627 <strong>\u0641\u0642\u0637 X%<\/strong>\u060c \u06a9\u0647 \u0628\u0631\u0627\u0628\u0631 \u0628\u0627 \u062d\u062f\u0633 \u0632\u062f\u0646 \u062a\u0635\u0627\u062f\u0641\u06cc \u0627\u0633\u062a.&#8221;<\/p>\n<\/blockquote>\n<p>\u067e\u0633 \u0627\u0632 \u0627\u06cc\u0646 \u0628\u062e\u0634 \u0647\u0627\u060c \u06a9\u0644 \u0641\u0635\u0644 \u0647\u0627\u06cc \u06a9\u062a\u0627\u0628 \u0648 \u062f\u0648\u0631\u0647 \u0628\u0647 \u0622\u0646 \u0627\u062e\u062a\u0635\u0627\u0635 \u062f\u0627\u062f\u0647 \u0634\u062f\u0647 \u0627\u0633\u062a <strong>\u0645\u0628\u0627\u0631\u0632\u0647 \u0628\u0627 \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0645\u0646\u0627\u0633\u0628<\/strong> \u0648 \u0631\u0648\u0634 \u0627\u062c\u062a\u0646\u0627\u0628 \u0627\u0632 \u0622\u0646  \u0627\u06cc\u0646 \u06a9\u0644\u0645\u0647 \u0628\u0647 \u0639\u0646\u0648\u0627\u0646 \u06cc\u06a9 \u0646\u0646\u06af \u0634\u062f <em>\u0628\u0647 \u0637\u0648\u0631 \u06a9\u0644\u06cc \u0686\u06cc\u0632 \u0628\u062f<\/em>.  \u0648 \u0627\u06cc\u0646\u062c\u0627\u0633\u062a \u06a9\u0647 \u0645\u0641\u0647\u0648\u0645 \u06a9\u0644\u06cc \u0645\u0637\u0631\u062d \u0645\u06cc \u0634\u0648\u062f:<\/p>\n<blockquote>\n<p>&#8220;\u0645\u0646 \u0628\u0627\u06cc\u062f \u0628\u0647 \u0647\u0631 \u0642\u06cc\u0645\u062a\u06cc \u0627\u0632 \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u062c\u0644\u0648\u06af\u06cc\u0631\u06cc \u06a9\u0646\u0645.&#8221;<\/p>\n<\/blockquote>\n<p>\u0628\u0647 \u0622\u0646 \u062a\u0648\u062c\u0647 \u0628\u06cc\u0634\u062a\u0631\u06cc \u0646\u0633\u0628\u062a \u0628\u0647 \u0639\u062f\u0645 \u062a\u0646\u0627\u0633\u0628 \u062f\u0627\u062f\u0647 \u0634\u062f\u0647 \u0627\u0633\u062a\u060c \u06a9\u0647 \u0628\u0647 \u0647\u0645\u0627\u0646 \u0627\u0646\u062f\u0627\u0632\u0647 &#8220;\u0628\u062f&#8221; \u0627\u0633\u062a.  \u0634\u0627\u06cc\u0627\u0646 \u0630\u06a9\u0631 \u0627\u0633\u062a \u06a9\u0647 \u00ab\u0628\u062f\u00bb \u06cc\u06a9 \u0627\u0635\u0637\u0644\u0627\u062d \u062f\u0644\u062e\u0648\u0627\u0647 \u0627\u0633\u062a \u0648 \u0647\u06cc\u0686 \u06cc\u06a9 \u0627\u0632 \u0627\u06cc\u0646 \u0634\u0631\u0627\u06cc\u0637 \u0630\u0627\u062a\u0627\u064b \u00ab\u062e\u0648\u0628\u00bb \u06cc\u0627 \u00ab\u0628\u062f\u00bb \u0646\u06cc\u0633\u062a\u0646\u062f.  \u0628\u0631\u062e\u06cc \u0645\u0645\u06a9\u0646 \u0627\u0633\u062a \u0627\u062f\u0639\u0627 \u06a9\u0646\u0646\u062f \u06a9\u0647 \u0645\u062f\u0644 \u0647\u0627\u06cc \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0627\u0632 \u0646\u0638\u0631 \u0641\u0646\u06cc \u0628\u06cc\u0634\u062a\u0631 \u0647\u0633\u062a\u0646\u062f <em>\u0645\u0641\u06cc\u062f<\/em>\u060c \u0632\u06cc\u0631\u0627 \u062d\u062f\u0627\u0642\u0644 \u0639\u0645\u0644\u06a9\u0631\u062f \u062e\u0648\u0628\u06cc \u062f\u0627\u0631\u0646\u062f \u0631\u0648\u06cc <em>\u0628\u0631\u062e\u06cc \u0627\u0632 \u062f\u0627\u062f\u0647 \u0647\u0627<\/em> \u062f\u0631 \u062d\u0627\u0644\u06cc \u06a9\u0647 \u0645\u062f\u0644 \u0647\u0627\u06cc \u0632\u06cc\u0631\u0646\u0648\u06cc\u0633 \u0639\u0645\u0644\u06a9\u0631\u062f \u062e\u0648\u0628\u06cc \u062f\u0627\u0631\u0646\u062f \u0631\u0648\u06cc <em>\u0627\u0637\u0644\u0627\u0639\u0627\u062a\u06cc \u0648\u062c\u0648\u062f \u0646\u062f\u0627\u0631\u062f<\/em>\u060c \u0627\u0645\u0627 \u062a\u0648\u0647\u0645 \u0645\u0648\u0641\u0642\u06cc\u062a \u06a9\u0627\u0646\u062f\u06cc\u062f\u0627\u06cc \u062e\u0648\u0628\u06cc \u0628\u0631\u0627\u06cc \u063a\u0644\u0628\u0647 \u0628\u0631 \u0627\u06cc\u0646 \u0645\u0632\u06cc\u062a \u0627\u0633\u062a.<\/p>\n<p>\u0628\u0631\u0627\u06cc \u0645\u0631\u062c\u0639 \u060c \u0628\u06cc\u0627\u06cc\u06cc\u062f \u0645\u0634\u0648\u0631\u062a \u06a9\u0646\u06cc\u0645 <em>Google Trends<\/em> \u0648 <em>Google Ngram Viewer<\/em>.  Google Trends \u0631\u0648\u0646\u062f\u0647\u0627\u06cc \u062f\u0627\u062f\u0647\u200c\u0647\u0627\u06cc \u062c\u0633\u062a\u062c\u0648 \u0631\u0627 \u0646\u0645\u0627\u06cc\u0634 \u0645\u06cc\u200c\u062f\u0647\u062f\u060c \u062f\u0631 \u062d\u0627\u0644\u06cc \u06a9\u0647 Google Ngram Viewer \u062a\u0639\u062f\u0627\u062f \u0648\u0642\u0648\u0639 \u0622\u0646 \u0631\u0627 \u0645\u06cc\u200c\u0634\u0645\u0627\u0631\u062f <em>n-\u06af\u0631\u0645<\/em> (\u062a\u0648\u0627\u0644\u06cc \u0627\u0632 <em>n<\/em> \u0645\u0648\u0627\u0631\u062f\u06cc \u0645\u0627\u0646\u0646\u062f \u06a9\u0644\u0645\u0627\u062a) \u062f\u0631 \u0627\u062f\u0628\u06cc\u0627\u062a\u060c \u062a\u062c\u0632\u06cc\u0647 \u0648 \u062a\u062d\u0644\u06cc\u0644 \u062a\u0639\u062f\u0627\u062f \u0632\u06cc\u0627\u062f\u06cc \u06a9\u062a\u0627\u0628 \u062f\u0631 \u0637\u0648\u0644 \u0627\u0639\u0635\u0627\u0631:<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/rasanegar.com\/blog\/wp-content\/uploads\/2024\/01\/overfitting-is-your-friend-not-your-foe-1.jpg\" alt=\"\u0631\u0648\u0646\u062f\u0647\u0627\u06cc \u062c\u0633\u062a\u062c\u0648\u06cc \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0645\u0646\u0627\u0633\u0628 \u062f\u0631 \u0645\u0642\u0627\u0628\u0644 \u0639\u062f\u0645 \u062a\u0646\u0627\u0633\u0628 \u0648 \u0646\u0645\u0627\u06cc\u0634\u06af\u0631 n-gram\" title=\"\"><\/p>\n<p>\u0647\u0645\u0647 \u062f\u0631 \u0645\u0648\u0631\u062f \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0648 \u0628\u06cc\u0634\u062a\u0631 \u062f\u0631 \u0632\u0645\u06cc\u0646\u0647 \u0627\u062c\u062a\u0646\u0627\u0628 \u0627\u0632 \u0622\u0646 \u0635\u062d\u0628\u062a \u0645\u06cc \u06a9\u0646\u0646\u062f &#8211; \u06a9\u0647 \u0627\u063a\u0644\u0628 \u0627\u0648\u0642\u0627\u062a \u0645\u0631\u062f\u0645 \u0631\u0627 \u0628\u0647 \u06cc\u06a9 \u062a\u0635\u0648\u0631 \u06a9\u0644\u06cc \u0633\u0648\u0642 \u0645\u06cc \u062f\u0647\u062f <strong>\u0630\u0627\u062a\u0627 \u06cc\u06a9 \u0686\u06cc\u0632 \u0628\u062f<\/strong>.<\/p>\n<p>\u0627\u06cc\u0646 \u0647\u0633\u062a <em>\u062f\u0631\u0633\u062a \u0627\u0633\u062a\u060c \u0648\u0627\u0642\u0639\u06cc<\/em>\u060c \u0628\u0647 \u0627\u0644\u0641 <em>\u062f\u0631\u062c\u0647<\/em>.  \u0628\u0644\u0647 &#8211; \u0634\u0645\u0627 \u0646\u0645\u06cc \u062e\u0648\u0627\u0647\u06cc\u062f \u06a9\u0647 \u0645\u062f\u0644 \u067e\u0627\u06cc\u0627\u0646 \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0627\u0641\u0632\u0627\u06cc\u0634 \u06cc\u0627\u0628\u062f \u060c \u062f\u0631 \u063a\u06cc\u0631 \u0627\u06cc\u0646 \u0635\u0648\u0631\u062a \u060c \u0639\u0645\u0644\u0627\u064b \u0628\u06cc \u0641\u0627\u06cc\u062f\u0647 \u0627\u0633\u062a.  \u0627\u0645\u0627 \u0634\u0645\u0627 \u0628\u0644\u0627\u0641\u0627\u0635\u0644\u0647 \u0628\u0647 \u0645\u062f\u0644 \u067e\u0627\u06cc\u0627\u0646 \u0646\u0645\u06cc \u0631\u0633\u06cc\u062f &#8211; \u0628\u0627\u0631\u0647\u0627 \u0648 \u0628\u0627\u0631\u0647\u0627 \u0622\u0646 \u0631\u0627 \u0628\u0627 \u0647\u0627\u06cc\u067e\u0631\u067e\u0627\u0631\u0627\u0645\u062a\u0631\u0647\u0627\u06cc \u0645\u062e\u062a\u0644\u0641 \u062a\u063a\u06cc\u06cc\u0631 \u0645\u06cc \u062f\u0647\u06cc\u062f.  \u062f\u0631 \u0637\u06cc \u0627\u06cc\u0646 process \u062c\u0627\u06cc\u06cc \u0627\u0633\u062a \u06a9\u0647 \u0634\u0645\u0627 <strong>\u0646\u0628\u0627\u06cc\u062f \u0628\u0647 \u062f\u06cc\u062f\u0646 \u0627\u062a\u0641\u0627\u0642\u0627\u062a \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u062a\u0648\u062c\u0647 \u06a9\u0631\u062f<\/strong> &#8211; \u0627\u06cc\u0646 \u06cc\u06a9 <strong><em>\u0646\u0634\u0627\u0646\u0647 \u062e\u0648\u0628<\/em><\/strong>\u060c \u0627\u06af\u0631 \u0686\u0647\u060c <strong>\u0646\u062a\u06cc\u062c\u0647 \u062e\u0648\u0628\u06cc \u0646\u06cc\u0633\u062a<\/strong>.<\/p>\n<h2 id=\"howoverfittingisntasbadasitsmadeouttobe\"><span class=\"ez-toc-section\" id=\"%da%86%d9%82%d8%af%d8%b1_%d8%a8%db%8c%d8%b4_%d8%a7%d8%b2_%d8%ad%d8%af_%d8%a8%d8%af_%d9%86%db%8c%d8%b3%d8%aa_%d8%a8%d9%87_%d9%87%d9%85%d8%a7%d9%86_%d8%a7%d9%86%d8%af%d8%a7%d8%b2%d9%87_%da%a9%d9%87_%d8%b3%d8%a7%d8%ae%d8%aa%d9%87_%d8%b4%d8%af%d9%87_%d8%a7%d8%b3%d8%aa\"><\/span>\u0686\u0642\u062f\u0631 \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0628\u062f \u0646\u06cc\u0633\u062a \u0628\u0647 \u0647\u0645\u0627\u0646 \u0627\u0646\u062f\u0627\u0632\u0647 \u06a9\u0647 \u0633\u0627\u062e\u062a\u0647 \u0634\u062f\u0647 \u0627\u0633\u062a<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<blockquote>\n<p><strong>\u06cc\u06a9 \u0645\u062f\u0644 \u0648 \u0645\u0639\u0645\u0627\u0631\u06cc \u06a9\u0647 \u062a\u0648\u0627\u0646\u0627\u06cc\u06cc \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0628\u0631\u0627\u0632\u0634 \u0631\u0627 \u062f\u0627\u0631\u062f\u060c \u0627\u06af\u0631 \u0622\u0646 \u0631\u0627 \u0633\u0627\u062f\u0647 \u06a9\u0646\u06cc\u062f (\u0648\/\u06cc\u0627 \u062f\u0627\u062f\u0647 \u0647\u0627 \u0631\u0627 \u062a\u063a\u06cc\u06cc\u0631 \u062f\u0647\u06cc\u062f) \u0628\u0647 \u0627\u062d\u062a\u0645\u0627\u0644 \u0632\u06cc\u0627\u062f \u062a\u0648\u0627\u0646\u0627\u06cc\u06cc \u062a\u0639\u0645\u06cc\u0645 \u062e\u0648\u0628\u06cc \u0628\u0647 \u0646\u0645\u0648\u0646\u0647 \u0647\u0627\u06cc \u062c\u062f\u06cc\u062f \u0631\u0627 \u062f\u0627\u0631\u062f.<\/strong><\/p>\n<\/blockquote>\n<ul>\n<li><small>\u0628\u0639\u0636\u06cc \u0627\u0648\u0642\u0627\u062a \u060c \u0641\u0642\u0637 \u0645\u0631\u0628\u0648\u0637 \u0628\u0647 \u0645\u062f\u0644 \u0646\u06cc\u0633\u062a \u060c \u0647\u0645\u0627\u0646\u0637\u0648\u0631 \u06a9\u0647 \u0628\u0639\u062f\u0627\u064b \u0645\u06cc \u0628\u06cc\u0646\u06cc\u0645. <\/small><\/li>\n<\/ul>\n<p>\u0627\u06af\u0631 \u06cc\u06a9 \u0645\u062f\u0644 <em>\u0645\u06cc \u062a\u0648\u0627\u0646<\/em> \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f\u060c \u0622\u0646 \u0631\u0627 \u0628\u0647 \u0627\u0646\u062f\u0627\u0632\u0647 \u06a9\u0627\u0641\u06cc \u0627\u0633\u062a <em>\u0638\u0631\u0641\u06cc\u062a \u0622\u0646\u062a\u0631\u0648\u067e\u06cc\u06a9<\/em> \u0628\u0631\u0627\u06cc \u0627\u0633\u062a\u062e\u0631\u0627\u062c \u0648\u06cc\u0698\u06af\u06cc \u0647\u0627 (\u0628\u0647 \u0631\u0648\u0634\u06cc \u0645\u0639\u0646\u06cc \u062f\u0627\u0631 \u0648 \u063a\u06cc\u0631 \u0645\u0639\u0646\u0627\u062f\u0627\u0631) \u0627\u0632 \u062f\u0627\u062f\u0647 \u0647\u0627.  \u0627\u0632 \u0622\u0646\u062c\u0627\u060c \u06cc\u0627 \u0627\u06cc\u0646 \u0627\u0633\u062a \u06a9\u0647 \u0645\u062f\u0644 \u0628\u06cc\u0634 \u0627\u0632 \u0638\u0631\u0641\u06cc\u062a \u0622\u0646\u062a\u0631\u0648\u067e\u06cc\u06a9 \u0645\u0648\u0631\u062f \u0646\u06cc\u0627\u0632 (\u067e\u06cc\u0686\u06cc\u062f\u06af\u06cc\/\u0642\u062f\u0631\u062a) \u062f\u0627\u0631\u062f \u06cc\u0627 \u0627\u06cc\u0646\u06a9\u0647 \u062e\u0648\u062f \u062f\u0627\u062f\u0647 \u06a9\u0627\u0641\u06cc \u0646\u06cc\u0633\u062a (\u0645\u0648\u0631\u062f \u0628\u0633\u06cc\u0627\u0631 \u0631\u0627\u06cc\u062c).<\/p>\n<p>\u0628\u06cc\u0627\u0646\u06cc\u0647 \u0645\u0639\u06a9\u0648\u0633 \u0646\u06cc\u0632 \u0645\u06cc \u062a\u0648\u0627\u0646\u062f \u0635\u062d\u06cc\u062d \u0628\u0627\u0634\u062f \u060c \u0627\u0645\u0627 \u0628\u0647 \u0646\u062f\u0631\u062a.  \u0627\u06af\u0631 \u0645\u062f\u0644 \u06cc\u0627 \u0645\u0639\u0645\u0627\u0631\u06cc \u0645\u0639\u06cc\u0646\u06cc \u0645\u0646\u0627\u0633\u0628 \u0646\u06cc\u0633\u062a\u060c \u0645\u06cc\u200c\u062a\u0648\u0627\u0646\u06cc\u062f \u0645\u062f\u0644 \u0631\u0627 \u062a\u063a\u06cc\u06cc\u0631 \u062f\u0647\u06cc\u062f \u062a\u0627 \u0628\u0628\u06cc\u0646\u06cc\u062f \u0622\u06cc\u0627 \u0648\u06cc\u0698\u06af\u06cc\u200c\u0647\u0627\u06cc \u062e\u0627\u0635\u06cc \u0631\u0627 \u0627\u0646\u062a\u062e\u0627\u0628 \u0645\u06cc\u200c\u06a9\u0646\u062f \u06cc\u0627 \u062e\u06cc\u0631\u060c \u0627\u0645\u0627 \u0646\u0648\u0639 \u0645\u062f\u0644 \u0645\u0645\u06a9\u0646 \u0627\u0633\u062a \u0628\u0631\u0627\u06cc \u0627\u06cc\u0646 \u06a9\u0627\u0631 \u0627\u0634\u062a\u0628\u0627\u0647 \u0628\u0627\u0634\u062f \u0648 \u0634\u0645\u0627 \u0646\u062a\u0648\u0627\u0646\u06cc\u062f \u062f\u0627\u062f\u0647\u200c\u0647\u0627 \u0631\u0627 \u0628\u0627 \u0622\u0646 \u062a\u0637\u0628\u06cc\u0642 \u062f\u0647\u06cc\u062f. \u0645\u0647\u0645 \u0627\u0633\u062a \u06a9\u0647 \u0686\u0647 \u06a9\u0627\u0631\u06cc \u0627\u0646\u062c\u0627\u0645 \u0645\u06cc \u062f\u0647\u06cc\u062f  \u0628\u0631\u062e\u06cc \u0627\u0632 \u0645\u062f\u0644 \u0647\u0627 \u0641\u0642\u0637 \u062f\u0631 \u0633\u0637\u062d\u06cc \u0627\u0632 \u062f\u0642\u062a \u06af\u06cc\u0631 \u0645\u06cc \u06a9\u0646\u0646\u062f\u060c \u0632\u06cc\u0631\u0627 \u0646\u0645\u06cc \u062a\u0648\u0627\u0646\u0646\u062f \u0648\u06cc\u0698\u06af\u06cc \u0647\u0627\u06cc \u06a9\u0627\u0641\u06cc \u0628\u0631\u0627\u06cc \u062a\u0645\u0627\u06cc\u0632 \u0628\u06cc\u0646 \u06a9\u0644\u0627\u0633 \u0647\u0627\u06cc \u062e\u0627\u0635 \u06cc\u0627 \u067e\u06cc\u0634 \u0628\u06cc\u0646\u06cc \u0645\u0642\u0627\u062f\u06cc\u0631 \u0631\u0627 \u0627\u0633\u062a\u062e\u0631\u0627\u062c \u06a9\u0646\u0646\u062f.<\/p>\n<p>\u06a9\u0647 \u062f\u0631 <strong>\u067e\u062e\u062a \u0648 \u067e\u0632<\/strong>\u060c \u06cc\u06a9 \u0642\u06cc\u0627\u0633 \u0645\u0639\u06a9\u0648\u0633 \u0627\u06cc\u062c\u0627\u062f \u0645\u06cc \u0634\u0648\u062f.  \u0628\u0647\u062a\u0631 \u0627\u0633\u062a \u0632\u0648\u062f\u062a\u0631 \u0627\u0632 \u0646\u0645\u06a9 \u06a9\u0645\u062a\u0631\u06cc \u062f\u0631 \u062e\u0648\u0631\u0634 \u0627\u0633\u062a\u0641\u0627\u062f\u0647 \u06a9\u0646\u06cc\u062f \u0631\u0648\u06cc\u060c \u0647\u0645\u0627\u0646\u0637\u0648\u0631 \u06a9\u0647 \u0647\u0645\u06cc\u0634\u0647 \u0645\u06cc \u062a\u0648\u0627\u0646\u06cc\u062f \u0628\u0639\u062f\u0627\u064b \u0646\u0645\u06a9 \u0627\u0636\u0627\u0641\u0647 \u06a9\u0646\u06cc\u062f \u062a\u0627 \u0637\u0639\u0645 \u0622\u0646 \u0631\u0627 \u0628\u062e\u0648\u0631\u06cc\u062f \u060c \u0627\u0645\u0627 \u0633\u062e\u062a \u0627\u0633\u062a \u06a9\u0647 \u06cc\u06a9 \u0628\u0627\u0631 \u0627\u0632 \u0642\u0628\u0644 \u0622\u0646 \u0631\u0627 \u0628\u0631\u062f\u0627\u0631\u06cc\u062f.<\/p>\n<p>\u06a9\u0647 \u062f\u0631 <strong>\u0641\u0631\u0627\u06af\u06cc\u0631\u06cc \u0645\u0627\u0634\u06cc\u0646<\/strong> &#8211; \u0628\u0631\u0639\u06a9\u0633 \u0627\u0633\u062a.  \u0628\u0647\u062a\u0631 \u0627\u0633\u062a \u06cc\u06a9 \u0645\u062f\u0644 overfit \u062f\u0627\u0634\u062a\u0647 \u0628\u0627\u0634\u06cc\u062f\u060c \u0633\u067e\u0633 \u0622\u0646 \u0631\u0627 \u0633\u0627\u062f\u0647 \u06a9\u0646\u06cc\u062f\u060c \u0647\u0627\u06cc\u067e\u0631\u067e\u0627\u0631\u0627\u0645\u062a\u0631\u0647\u0627 \u0631\u0627 \u062a\u063a\u06cc\u06cc\u0631 \u062f\u0647\u06cc\u062f\u060c \u062f\u0627\u062f\u0647 \u0647\u0627 \u0631\u0627 \u062a\u0642\u0648\u06cc\u062a \u06a9\u0646\u06cc\u062f \u0648 \u063a\u06cc\u0631\u0647 \u0631\u0627 \u0628\u0647 \u062e\u0648\u0628\u06cc \u062a\u0639\u0645\u06cc\u0645 \u062f\u0647\u06cc\u062f\u060c \u0627\u0645\u0627 \u0627\u0646\u062c\u0627\u0645 \u0628\u0631\u0639\u06a9\u0633 \u0622\u0646 (\u062f\u0631 \u062a\u0646\u0638\u06cc\u0645\u0627\u062a \u0639\u0645\u0644\u06cc) \u062f\u0634\u0648\u0627\u0631\u062a\u0631 \u0627\u0633\u062a.  \u0627\u062c\u062a\u0646\u0627\u0628 \u0627\u0632 \u0646\u0635\u0628 \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f <em>\u0642\u0628\u0644 \u0627\u0632<\/em> \u0627\u06cc\u0646 \u0627\u062a\u0641\u0627\u0642 \u0645\u0645\u06a9\u0646 \u0627\u0633\u062a \u062e\u06cc\u0644\u06cc \u062e\u0648\u0628 \u0634\u0645\u0627 \u0631\u0627 \u0627\u0632 \u06cc\u0627\u0641\u062a\u0646 \u0645\u062f\u0644 \u0645\u0646\u0627\u0633\u0628 \u0648\/\u06cc\u0627 \u0645\u0639\u0645\u0627\u0631\u06cc \u0628\u0631\u0627\u06cc \u0645\u062f\u062a \u0632\u0645\u0627\u0646 \u0637\u0648\u0644\u0627\u0646\u06cc \u062a\u0631 \u062f\u0648\u0631 \u06a9\u0646\u062f.<\/p>\n<p>\u062f\u0631 \u0639\u0645\u0644 \u060c \u0648 \u062f\u0631 \u0628\u0631\u062e\u06cc \u0627\u0632 \u062c\u0630\u0627\u0628 \u062a\u0631\u06cc\u0646 \u0645\u0648\u0627\u0631\u062f \u0627\u0633\u062a\u0641\u0627\u062f\u0647 \u0627\u0632 \u06cc\u0627\u062f\u06af\u06cc\u0631\u06cc \u0645\u0627\u0634\u06cc\u0646 \u0648 \u06cc\u0627\u062f\u06af\u06cc\u0631\u06cc \u0639\u0645\u06cc\u0642 \u060c \u06a9\u0627\u0631 \u062e\u0648\u0627\u0647\u06cc\u062f \u06a9\u0631\u062f \u0631\u0648\u06cc \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u0647\u0627\u06cc\u06cc \u06a9\u0647 \u062f\u0631 \u0622\u0646 \u0645\u0634\u06a9\u0644 \u062f\u0627\u0631\u06cc\u062f.  \u0627\u06cc\u0646\u0647\u0627 \u0645\u062c\u0645\u0648\u0639\u0647\u200c\u0647\u0627\u06cc \u062f\u0627\u062f\u0647\u200c\u0627\u06cc \u0647\u0633\u062a\u0646\u062f \u06a9\u0647 \u0645\u0639\u0645\u0648\u0644\u0627\u064b \u0627\u0632 \u0622\u0646\u200c\u0647\u0627 \u0627\u0633\u062a\u0641\u0627\u062f\u0647 \u0646\u0645\u06cc\u200c\u06a9\u0646\u06cc\u062f\u060c \u0628\u062f\u0648\u0646 \u0627\u06cc\u0646\u06a9\u0647 \u062a\u0648\u0627\u0646\u0627\u06cc\u06cc \u067e\u06cc\u062f\u0627 \u06a9\u0631\u062f\u0646 \u0645\u062f\u0644\u200c\u0647\u0627 \u0648 \u0645\u0639\u0645\u0627\u0631\u06cc\u200c\u0647\u0627\u06cc\u06cc \u0631\u0627 \u062f\u0627\u0634\u062a\u0647 \u0628\u0627\u0634\u06cc\u062f \u06a9\u0647 \u0628\u062a\u0648\u0627\u0646\u0646\u062f \u0628\u0647 \u062e\u0648\u0628\u06cc \u062a\u0639\u0645\u06cc\u0645 \u062f\u0627\u062f\u0647 \u0648 \u0648\u06cc\u0698\u06af\u06cc\u200c\u0647\u0627 \u0631\u0627 \u0627\u0633\u062a\u062e\u0631\u0627\u062c \u06a9\u0646\u0646\u062f.<\/p>\n<p>\u0647\u0645\u0686\u0646\u06cc\u0646 \u0634\u0627\u06cc\u0627\u0646 \u0630\u06a9\u0631 \u0627\u0633\u062a \u06a9\u0647 \u062a\u0641\u0627\u0648\u062a \u0628\u06cc\u0646 \u0622\u0646\u0686\u0647 \u0645\u0646 \u0645\u06cc \u06af\u0648\u06cc\u0645 <strong>\u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0648\u0627\u0642\u0639\u06cc<\/strong> \u0648 <strong>\u0627\u0636\u0627\u0641\u0647 \u062a\u0646\u0627\u0633\u0628 \u062c\u0632\u0626\u06cc<\/strong>.  \u0645\u062f\u0644\u06cc \u06a9\u0647 \u0627\u0632 \u06cc\u06a9 \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u0627\u0633\u062a\u0641\u0627\u062f\u0647 \u0645\u06cc \u06a9\u0646\u062f \u0648 \u0628\u0647 \u062f\u0642\u062a 60 \u066a \u062f\u0633\u062a \u0645\u06cc \u06cc\u0627\u0628\u062f \u0631\u0648\u06cc \u0645\u062c\u0645\u0648\u0639\u0647 \u0622\u0645\u0648\u0632\u0634\u06cc\u060c \u062a\u0646\u0647\u0627 \u0628\u0627 40% \u0631\u0648\u06cc \u0645\u062c\u0645\u0648\u0639\u0647\u200c\u0647\u0627\u06cc \u0627\u0639\u062a\u0628\u0627\u0631\u0633\u0646\u062c\u06cc \u0648 \u0622\u0632\u0645\u0627\u06cc\u0634 \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0628\u062e\u0634\u06cc \u0627\u0632 \u062f\u0627\u062f\u0647\u200c\u0647\u0627 \u0631\u0627 \u0628\u0631\u0627\u0632\u0634 \u0645\u06cc\u200c\u06a9\u0646\u0646\u062f.  \u0628\u0627 \u0627\u06cc\u0646 \u062d\u0627\u0644\u060c \u0627\u06cc\u0646\u0637\u0648\u0631 \u0646\u06cc\u0633\u062a <strong>\u0648\u0627\u0642\u0639\u0627 \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0645\u0646\u0627\u0633\u0628<\/strong> \u0628\u0647 \u0645\u0639\u0646\u0627\u06cc \u062a\u062d\u062a \u0627\u0644\u0634\u0639\u0627\u0639 \u0642\u0631\u0627\u0631 \u062f\u0627\u062f\u0646 \u06a9\u0644 \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647\u060c \u0648 \u062f\u0633\u062a\u06cc\u0627\u0628\u06cc \u0628\u0647 \u0646\u0631\u062e \u062f\u0642\u062a \u0646\u0632\u062f\u06cc\u06a9 \u0628\u0647 100\u066a (\u06a9\u0627\u0630\u0628)\u060c \u062f\u0631 \u062d\u0627\u0644\u06cc \u06a9\u0647 \u0627\u0639\u062a\u0628\u0627\u0631\u0633\u0646\u062c\u06cc \u0648 \u0645\u062c\u0645\u0648\u0639\u0647 \u0647\u0627\u06cc \u0622\u0632\u0645\u0627\u06cc\u0634\u06cc \u0622\u0646\u060c \u0645\u062b\u0644\u0627\u064b 40\u066a \u067e\u0627\u06cc\u06cc\u0646 \u0627\u0633\u062a.<\/p>\n<p>\u0645\u062f\u0644\u06cc \u06a9\u0647 \u062a\u0627 \u062d\u062f\u06cc \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0627\u0632 \u0622\u0646 \u0627\u0633\u062a\u0641\u0627\u062f\u0647 \u0645\u06cc \u06a9\u0646\u062f \u060c \u0622\u0646 \u0686\u06cc\u0632\u06cc \u0646\u06cc\u0633\u062a \u06a9\u0647 \u0628\u062a\u0648\u0627\u0646\u062f \u0628\u0627 \u0633\u0627\u062f\u0647 \u0633\u0627\u0632\u06cc \u0628\u0647 \u062e\u0648\u0628\u06cc \u062a\u0639\u0645\u06cc\u0645 \u062f\u0647\u062f \u060c \u0647\u0645\u0627\u0646\u0637\u0648\u0631 \u06a9\u0647 \u0646\u062f\u0627\u0631\u062f <em>\u06a9\u0627\u0641\u06cc<\/em> \u0638\u0631\u0641\u06cc\u062a \u0622\u0646\u062a\u0631\u0648\u067e\u06cc\u06a9 \u0628\u0631\u0627\u06cc \u062a\u0646\u0627\u0633\u0628 \u0648\u0627\u0642\u0639\u06cc (\u0628\u06cc\u0634 \u0627\u0632).  \u0628\u0647 \u0645\u062d\u0636 \u0627\u0646\u062c\u0627\u0645\u060c \u0627\u0633\u062a\u062f\u0644\u0627\u0644 \u0645\u0646 \u0627\u0639\u0645\u0627\u0644 \u0645\u06cc \u0634\u0648\u062f\u060c \u0627\u06af\u0631\u0686\u0647 \u0645\u0648\u0641\u0642\u06cc\u062a \u0631\u0627 \u062a\u0636\u0645\u06cc\u0646 \u0646\u0645\u06cc \u06a9\u0646\u062f\u060c \u0647\u0645\u0627\u0646\u0637\u0648\u0631 \u06a9\u0647 \u062f\u0631 \u0628\u062e\u0634 \u0647\u0627\u06cc \u0628\u0639\u062f\u06cc \u062a\u0648\u0636\u06cc\u062d \u062f\u0627\u062f\u0647 \u0634\u062f.<\/p>\n<h2 id=\"casestudyfriendlyoverfittingargument\"><span class=\"ez-toc-section\" id=\"%d9%85%d8%b7%d8%a7%d9%84%d8%b9%d9%87_%d9%85%d9%88%d8%b1%d8%af%db%8c_%e2%80%93_%d8%a7%d8%b3%d8%aa%d8%af%d9%84%d8%a7%d9%84_%d8%a8%db%8c%d8%b4_%d8%a7%d8%b2_%d8%ad%d8%af_%d8%af%d9%88%d8%b3%d8%aa%d8%a7%d9%86%d9%87\"><\/span>\u0645\u0637\u0627\u0644\u0639\u0647 \u0645\u0648\u0631\u062f\u06cc &#8211; \u0627\u0633\u062a\u062f\u0644\u0627\u0644 \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u062f\u0648\u0633\u062a\u0627\u0646\u0647<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>\u0627\u06cc\u0646 <a target=\"_blank\" rel=\"nofollow noopener noreferrer\" href=\"http:\/\/yann.lecun.com\/exdb\/mnist\/\">\u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u0627\u0631\u0642\u0627\u0645 \u062f\u0633\u062a \u0646\u0648\u06cc\u0633 MNIST<\/a>\u060c \u06af\u0631\u062f\u0622\u0648\u0631\u06cc \u0634\u062f\u0647 \u062a\u0648\u0633\u0637 Yann LeCun \u06cc\u06a9\u06cc \u0627\u0632 \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u0647\u0627\u06cc \u0645\u0639\u06cc\u0627\u0631 \u06a9\u0644\u0627\u0633\u06cc\u06a9 \u0627\u0633\u062a \u06a9\u0647 \u0628\u0631\u0627\u06cc \u0645\u062f\u0644 \u0647\u0627\u06cc \u0637\u0628\u0642\u0647 \u0628\u0646\u062f\u06cc \u0622\u0645\u0648\u0632\u0634\u06cc \u0627\u0633\u062a\u0641\u0627\u062f\u0647 \u0645\u06cc \u0634\u0648\u062f.  LeCun \u0628\u0647 \u0637\u0648\u0631 \u06af\u0633\u062a\u0631\u062f\u0647 \u0627\u06cc \u0628\u0647 \u0639\u0646\u0648\u0627\u0646 \u06cc\u06a9\u06cc \u0627\u0632 \u0628\u0646\u06cc\u0627\u0646\u06af\u0630\u0627\u0631\u0627\u0646 \u06cc\u0627\u062f\u06af\u06cc\u0631\u06cc \u0639\u0645\u06cc\u0642 \u062f\u0631 \u0646\u0638\u0631 \u06af\u0631\u0641\u062a\u0647 \u0645\u06cc \u0634\u0648\u062f &#8211; \u0628\u0627 \u0645\u0634\u0627\u0631\u06a9\u062a \u062f\u0631 \u0627\u06cc\u0646 \u0632\u0645\u06cc\u0646\u0647 \u06a9\u0647 \u0627\u06a9\u062b\u0631 \u0622\u0646\u0647\u0627 \u0646\u0645\u06cc \u062a\u0648\u0627\u0646\u0646\u062f \u062f\u0631 \u06a9\u0645\u0631\u0628\u0646\u062f \u062e\u0648\u062f \u0642\u0631\u0627\u0631 \u062f\u0647\u0646\u062f\u060c \u0648 \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u0627\u0631\u0642\u0627\u0645 \u062f\u0633\u062a \u0646\u0648\u06cc\u0633 MNIST \u06cc\u06a9\u06cc \u0627\u0632 \u0627\u0648\u0644\u06cc\u0646 \u0645\u0639\u06cc\u0627\u0631\u0647\u0627\u06cc \u0627\u0635\u0644\u06cc \u0645\u0648\u0631\u062f \u0627\u0633\u062a\u0641\u0627\u062f\u0647 \u0628\u0631\u0627\u06cc \u0645\u0631\u0627\u062d\u0644 \u0627\u0648\u0644\u06cc\u0647 \u0634\u0628\u06a9\u0647 \u0647\u0627\u06cc \u0639\u0635\u0628\u06cc \u06a9\u0627\u0646\u0648\u0644\u0648\u0634\u0646 \u0628\u0648\u062f. .<\/p>\n<blockquote>\n<p>\u0647\u0645\u0686\u0646\u06cc\u0646 \u0627\u06cc\u0646 \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0645\u0648\u0631\u062f \u0627\u0633\u062a\u0641\u0627\u062f\u0647 \u0642\u0631\u0627\u0631 \u06af\u0631\u0641\u062a\u0647 \u0627\u0633\u062a\u060c \u0628\u0647 \u0637\u0648\u0631 \u0628\u0627\u0644\u0642\u0648\u0647.<\/p>\n<\/blockquote>\n<p>\u0647\u06cc\u0686 \u0645\u0634\u06a9\u0644\u06cc \u062f\u0631 \u062e\u0648\u062f \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u0648\u062c\u0648\u062f \u0646\u062f\u0627\u0631\u062f\u060c \u0648 \u0646\u0647 \u0628\u0627 LeCun \u06a9\u0647 \u0622\u0646 \u0631\u0627 \u0627\u06cc\u062c\u0627\u062f \u06a9\u0631\u062f\u0647 \u0627\u0633\u062a &#8211; \u062f\u0631 \u0648\u0627\u0642\u0639 \u0628\u0633\u06cc\u0627\u0631 \u062e\u0648\u0628 \u0627\u0633\u062a\u060c \u0627\u0645\u0627 \u06cc\u0627\u0641\u062a\u0646 \u0646\u0645\u0648\u0646\u0647 \u0628\u0647 \u0645\u062b\u0627\u0644 \u0631\u0648\u06cc \u0647\u0645\u0627\u0646 \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u0622\u0646\u0644\u0627\u06cc\u0646 \u06a9\u0633\u0644 \u06a9\u0646\u0646\u062f\u0647 \u0627\u0633\u062a.  \u062f\u0631 \u06cc\u06a9 \u0646\u0642\u0637\u0647 &#8211; <strong>\u0645\u0627 \u062e\u0648\u062f\u0645\u0627\u0646 \u0631\u0627 \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0622\u0645\u0627\u062f\u0647 \u0645\u06cc \u06a9\u0646\u06cc\u0645<\/strong> \u0646\u06af\u0627\u0647 \u06a9\u0631\u062f\u0646 \u0628\u0647 \u0622\u0646  \u0686\u0642\u062f\u0631\u061f  \u062f\u0631 \u0627\u06cc\u0646\u062c\u0627 \u062a\u0644\u0627\u0634 \u0645\u0646 \u0628\u0631\u0627\u06cc \u0641\u0647\u0631\u0633\u062a \u06a9\u0631\u062f\u0646 \u062f\u0647 \u0631\u0642\u0645 \u0627\u0648\u0644 MNIST \u0627\u0632 \u0628\u0627\u0644\u0627\u06cc \u0633\u0631 \u0645\u0646 \u0627\u0633\u062a:<\/p>\n<pre><code class=\"hljs\">5, 0, 4, 1, 9, 2, 2, 4, 3\n<\/code><\/pre>\n<p>\u0686\u0647 \u06a9\u0627\u0631 \u06a9\u0631\u062f\u0645\u061f<\/p>\n<pre><code class=\"hljs\"><span class=\"hljs-keyword\">from<\/span> tensorflow <span class=\"hljs-keyword\">import<\/span> keras\n<span class=\"hljs-keyword\">import<\/span> numpy <span class=\"hljs-keyword\">as<\/span> np\n<span class=\"hljs-keyword\">import<\/span> matplotlib.pyplot <span class=\"hljs-keyword\">as<\/span> plt\n\n\n(X_train_full, Y_train_full), (X_test, Y_test) = keras.datasets.mnist.load_data()\n\nX_valid, X_train = X_train_full(:<span class=\"hljs-number\">5000<\/span>)\/<span class=\"hljs-number\">255.0<\/span>, X_train_full(<span class=\"hljs-number\">5000<\/span>:)\/<span class=\"hljs-number\">255.0<\/span>\nY_valid, Y_train = Y_train_full(:<span class=\"hljs-number\">5000<\/span>), Y_train_full(<span class=\"hljs-number\">5000<\/span>:)\n\nX_test = X_test\/<span class=\"hljs-number\">255.0<\/span>\n\n\nfig, ax = plt.subplots(<span class=\"hljs-number\">1<\/span>, <span class=\"hljs-number\">10<\/span>, figsize=(<span class=\"hljs-number\">10<\/span>,<span class=\"hljs-number\">2<\/span>))\n<span class=\"hljs-keyword\">for<\/span> i <span class=\"hljs-keyword\">in<\/span> <span class=\"hljs-built_in\">range<\/span>(<span class=\"hljs-number\">10<\/span>):\n    ax(i).imshow(X_train_full(i))\n    ax(i).axis(<span class=\"hljs-string\">'off'<\/span>)\n    plt.subplots_adjust(wspace=<span class=\"hljs-number\">1<\/span>) \n\nplt.show()\n<\/code><\/pre>\n<p><img decoding=\"async\" src=\"https:\/\/rasanegar.com\/blog\/wp-content\/uploads\/2024\/01\/overfitting-is-your-friend-not-your-foe-2.png\" alt=\"\" title=\"\"><\/p>\n<p>\u062a\u0642\u0631\u06cc\u0628\u0627\u064b \u0648\u062c\u0648\u062f \u062f\u0627\u0631\u062f.<\/p>\n<blockquote>\n<p>\u0645\u0646 \u0627\u0632 \u0627\u06cc\u0646 \u0641\u0631\u0635\u062a \u0628\u0631\u0627\u06cc \u062f\u0631\u062e\u0648\u0627\u0633\u062a \u0639\u0645\u0648\u0645\u06cc \u0627\u0632 \u0647\u0645\u0647 \u0633\u0627\u0632\u0646\u062f\u06af\u0627\u0646 \u0645\u062d\u062a\u0648\u0627 \u0627\u0633\u062a\u0641\u0627\u062f\u0647 \u062e\u0648\u0627\u0647\u0645 \u06a9\u0631\u062f \u06a9\u0647 \u0627\u0632 \u0627\u06cc\u0646 \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u0641\u0631\u0627\u062a\u0631 \u0627\u0632 \u0642\u0633\u0645\u062a \u0647\u0627\u06cc \u0645\u0642\u062f\u0645\u0627\u062a\u06cc \u0627\u0633\u062a\u0641\u0627\u062f\u0647 \u0646\u06a9\u0646\u0646\u062f\u060c \u062c\u0627\u06cc\u06cc \u06a9\u0647 \u0645\u06cc \u062a\u0648\u0627\u0646 \u0627\u0632 \u0633\u0627\u062f\u06af\u06cc \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u0628\u0631\u0627\u06cc \u06a9\u0627\u0647\u0634 \u0645\u0627\u0646\u0639 \u0648\u0631\u0648\u062f \u0627\u0633\u062a\u0641\u0627\u062f\u0647 \u06a9\u0631\u062f. <em>\u0644\u0637\u0641\u0627<\/em>.<\/p>\n<\/blockquote>\n<p>\u0639\u0644\u0627\u0648\u0647 \u0628\u0631 \u0627\u06cc\u0646\u060c \u0627\u06cc\u0646 \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u0633\u0627\u062e\u062a\u0646 \u0645\u062f\u0644\u06cc \u0631\u0627 \u06a9\u0647 \u0645\u0646\u0627\u0633\u0628 \u0646\u06cc\u0633\u062a \u062f\u0634\u0648\u0627\u0631 \u0645\u06cc \u06a9\u0646\u062f.  \u0627\u06cc\u0646 \u062e\u06cc\u0644\u06cc \u0633\u0627\u062f\u0647 \u0627\u0633\u062a &#8211; \u0648 \u062d\u062a\u06cc \u06cc\u06a9 \u0646\u0633\u0628\u062a\u0627\u064b \u06a9\u0648\u0686\u06a9 <strong>Perceptron \u0686\u0646\u062f \u0644\u0627\u06cc\u0647 (MLP)<\/strong> \u0637\u0628\u0642\u0647 \u0628\u0646\u062f\u06cc \u06a9\u0646\u0646\u062f\u0647 \u0633\u0627\u062e\u062a\u0647 \u0634\u062f\u0647 \u0628\u0627 \u062a\u0639\u062f\u0627\u062f \u0628\u0635\u0631\u06cc \u0644\u0627\u06cc\u0647 \u0647\u0627 \u0648 \u0646\u0648\u0631\u0648\u0646 \u0647\u0627 \u062f\u0631 \u0647\u0631 \u0644\u0627\u06cc\u0647 \u0645\u06cc \u062a\u0648\u0627\u0646\u062f \u0628\u0647 \u0631\u0627\u062d\u062a\u06cc \u0628\u0647 \u062f\u0642\u062a \u0628\u0627\u0644\u0627\u06cc 98 \u062f\u0631\u0635\u062f \u0628\u0631\u0633\u062f. \u0631\u0648\u06cc \u0645\u062c\u0645\u0648\u0639\u0647 \u0622\u0645\u0648\u0632\u0634 \u060c \u0622\u0632\u0645\u0627\u06cc\u0634 \u0648 \u0627\u0639\u062a\u0628\u0627\u0631 \u0633\u0646\u062c\u06cc. <a target=\"_blank\" href=\"https:\/\/github.com\/StackAbuse\/friendly-overfitting-argument\/blob\/main\/Friendly-Ovetfitting-Argument.ipynb\" rel=\"noopener\">\u0627\u06cc\u0646\u062c\u0627 \u06cc\u06a9 Jupyter \u0646\u0648\u062a \u0628\u0648\u06a9<\/a> \u0627\u0632 \u06cc\u06a9 MLP \u0633\u0627\u062f\u0647 \u062f\u0633\u062a\u06cc\u0627\u0628\u06cc \u0628\u0647 \u062f\u0642\u062a 98 ~ \u0631\u0648\u06cc \u0647\u0645 \u0645\u062c\u0645\u0648\u0639\u0647\u200c\u0647\u0627\u06cc \u0622\u0645\u0648\u0632\u0634\u060c \u0627\u0639\u062a\u0628\u0627\u0631\u0633\u0646\u062c\u06cc \u0648 \u062a\u0633\u062a\u060c \u06a9\u0647 \u0645\u0646 \u0628\u0627 \u067e\u06cc\u0634\u200c\u0641\u0631\u0636\u200c\u0647\u0627\u06cc \u0645\u0639\u0642\u0648\u0644 \u0627\u06cc\u062c\u0627\u062f \u06a9\u0631\u062f\u0645.<\/p>\n<blockquote>\n<p>\u0645\u0646 \u062d\u062a\u06cc \u0628\u0647 \u062e\u0648\u062f \u0632\u062d\u0645\u062a \u0646\u062f\u0627\u062f\u0647 \u0627\u0645 \u06a9\u0647 \u0622\u0646 \u0631\u0627 \u062a\u0646\u0638\u06cc\u0645 \u06a9\u0646\u0645 \u062a\u0627 \u0639\u0645\u0644\u06a9\u0631\u062f \u0628\u0647\u062a\u0631\u06cc \u0646\u0633\u0628\u062a \u0628\u0647 \u062a\u0646\u0638\u06cc\u0645\u0627\u062a \u0627\u0648\u0644\u06cc\u0647 \u062f\u0627\u0634\u062a\u0647 \u0628\u0627\u0634\u062f.<\/p>\n<\/blockquote>\n<h3 id=\"thecifar10andcifar100datasets\"><span class=\"ez-toc-section\" id=\"%d9%85%d8%ac%d9%85%d9%88%d8%b9%d9%87_%d8%af%d8%a7%d8%af%d9%87_%d9%87%d8%a7%db%8c_cifar10_%d9%88_cifar100\"><\/span>\u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u0647\u0627\u06cc CIFAR10 \u0648 CIFAR100<span class=\"ez-toc-section-end\"><\/span><\/h3>\n<p>\u0628\u06cc\u0627\u06cc\u06cc\u062f \u0627\u0632 \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647\u200c\u0627\u06cc \u0627\u0633\u062a\u0641\u0627\u062f\u0647 \u06a9\u0646\u06cc\u0645 \u06a9\u0647 \u067e\u06cc\u0686\u06cc\u062f\u0647\u200c\u062a\u0631 \u0627\u0632 \u0627\u0631\u0642\u0627\u0645 \u062f\u0633\u062a\u200c\u0646\u0648\u06cc\u0633 MNIST \u0627\u0633\u062a\u060c \u0648 \u0628\u0627\u0639\u062b \u0645\u06cc\u200c\u0634\u0648\u062f \u06cc\u06a9 MLP \u0633\u0627\u062f\u0647 \u06a9\u0645\u062a\u0631 \u0634\u0648\u062f\u060c \u0627\u0645\u0627 \u0628\u0647 \u0627\u0646\u062f\u0627\u0632\u0647\u200c\u0627\u06cc \u0633\u0627\u062f\u0647 \u0627\u0633\u062a \u06a9\u0647 \u0628\u0647 \u06cc\u06a9 CNN \u0628\u0627 \u0627\u0646\u062f\u0627\u0632\u0647 \u0645\u0646\u0627\u0633\u0628 \u0627\u062c\u0627\u0632\u0647 \u0645\u06cc\u200c\u062f\u0647\u062f \u0648\u0627\u0642\u0639\u0627\u064b \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0645\u0646\u0627\u0633\u0628 \u0628\u0627\u0634\u062f. \u0631\u0648\u06cc \u0622\u06cc \u062a\u06cc.  \u06cc\u06a9 \u06a9\u0627\u0646\u062f\u06cc\u062f\u0627\u06cc \u062e\u0648\u0628 \u0627\u0633\u062a <a target=\"_blank\" rel=\"nofollow noopener noreferrer\" href=\"https:\/\/www.cs.toronto.edu\/~kriz\/cifar.html\"><em>\u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 CIFAR<\/em><\/a>.<\/p>\n<blockquote>\n<p>10 \u06a9\u0644\u0627\u0633 \u062a\u0635\u0648\u06cc\u0631 \u062f\u0631 CIFAR10 \u0648 100 \u062f\u0631 CIFAR100 \u0648\u062c\u0648\u062f \u062f\u0627\u0631\u062f.  \u0639\u0644\u0627\u0648\u0647 \u0628\u0631 \u0627\u06cc\u0646\u060c \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 CIFAR100 \u062f\u0627\u0631\u0627\u06cc 20 \u062e\u0627\u0646\u0648\u0627\u062f\u0647 \u0627\u0633\u062a <em>\u0645\u0634\u0627\u0628\u0647<\/em> \u06a9\u0644\u0627\u0633\u200c\u0647\u0627\u060c \u0628\u0647 \u0627\u06cc\u0646 \u0645\u0639\u0646\u06cc \u06a9\u0647 \u0634\u0628\u06a9\u0647 \u0639\u0644\u0627\u0648\u0647 \u0628\u0631 \u0627\u06cc\u0646 \u0628\u0627\u06cc\u062f \u062a\u0641\u0627\u0648\u062a\u200c\u0647\u0627\u06cc \u062c\u0632\u0626\u06cc \u0628\u06cc\u0646 \u06a9\u0644\u0627\u0633\u200c\u0647\u0627\u06cc \u0645\u0634\u0627\u0628\u0647\u060c \u0627\u0645\u0627 \u0645\u062a\u0641\u0627\u0648\u062a \u0631\u0627 \u0628\u06cc\u0627\u0645\u0648\u0632\u062f.  \u0627\u06cc\u0646\u0647\u0627 \u0628\u0647 \u0639\u0646\u0648\u0627\u0646 \u0634\u0646\u0627\u062e\u062a\u0647 \u0645\u06cc \u0634\u0648\u0646\u062f <strong>&#8220;\u0628\u0631\u0686\u0633\u0628 \u0647\u0627\u06cc \u062e\u0648\u0628&#8221; (100)<\/strong> \u0648 <strong>&#8220;\u0628\u0631\u0686\u0633\u0628 \u0647\u0627\u06cc \u062f\u0631\u0634\u062a&#8221; (20)<\/strong> \u0648 \u067e\u06cc\u0634\u200c\u0628\u06cc\u0646\u06cc \u0627\u06cc\u0646\u0647\u0627 \u0628\u0631\u0627\u0628\u0631 \u0627\u0633\u062a \u0628\u0627 \u067e\u06cc\u0634\u200c\u0628\u06cc\u0646\u06cc \u0637\u0628\u0642\u0647 \u062e\u0627\u0635 \u06cc\u0627 \u0641\u0642\u0637 \u062e\u0627\u0646\u0648\u0627\u062f\u0647\u200c\u0627\u06cc \u06a9\u0647 \u0628\u0647 \u0622\u0646 \u062a\u0639\u0644\u0642 \u062f\u0627\u0631\u062f.<\/p>\n<\/blockquote>\n<p>\u0628\u0631\u0627\u06cc \u0645\u062b\u0627\u0644\u060c \u062f\u0631 \u0627\u06cc\u0646\u062c\u0627 \u06cc\u06a9 \u0633\u0648\u067e\u0631\u06a9\u0644\u0627\u0633 (\u0628\u0631\u0686\u0633\u0628 \u062f\u0631\u0634\u062a) \u0648 \u0632\u06cc\u0631 \u06a9\u0644\u0627\u0633\u200c\u0647\u0627 (\u0628\u0631\u0686\u0633\u0628\u200c\u0647\u0627\u06cc \u0638\u0631\u06cc\u0641) \u0648\u062c\u0648\u062f \u062f\u0627\u0631\u062f:<\/p>\n<table class=\"table table-striped\">\n<tbody>\n<tr>\n<td>\u0633\u0648\u067e\u0631\u06a9\u0644\u0627\u0633<\/td>\n<td>\u0632\u06cc\u0631 \u06a9\u0644\u0627\u0633 \u0647\u0627<\/td>\n<\/tr>\n<tr>\n<td>\u0638\u0631\u0648\u0641 \u063a\u0630\u0627<\/td>\n<td>\u0628\u0637\u0631\u06cc\u060c \u06a9\u0627\u0633\u0647\u060c \u0642\u0648\u0637\u06cc\u060c \u0641\u0646\u062c\u0627\u0646\u060c \u0628\u0634\u0642\u0627\u0628<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u06cc\u06a9 \u0641\u0646\u062c\u0627\u0646 \u06cc\u06a9 \u0627\u0633\u062a\u0648\u0627\u0646\u0647 \u0627\u0633\u062a\u060c \u0634\u0628\u06cc\u0647 \u0628\u0647 \u0642\u0648\u0637\u06cc \u0646\u0648\u0634\u0627\u0628\u0647\u060c \u0648 \u0645\u0645\u06a9\u0646 \u0627\u0633\u062a \u0628\u0631\u062e\u06cc \u0628\u0637\u0631\u06cc \u0647\u0627 \u0646\u06cc\u0632 \u0686\u0646\u06cc\u0646 \u0628\u0627\u0634\u0646\u062f.  \u0627\u0632 \u0622\u0646\u062c\u0627\u06cc\u06cc \u06a9\u0647 \u0627\u06cc\u0646 \u0648\u06cc\u0698\u06af\u06cc\u200c\u0647\u0627\u06cc \u0633\u0637\u062d \u067e\u0627\u06cc\u06cc\u0646 \u0646\u0633\u0628\u062a\u0627\u064b \u0645\u0634\u0627\u0628\u0647 \u0647\u0633\u062a\u0646\u062f\u060c \u0628\u0647 \u0631\u0627\u062d\u062a\u06cc \u0645\u06cc\u200c\u062a\u0648\u0627\u0646 \u0647\u0645\u0647 \u0622\u0646\u0647\u0627 \u0631\u0627 \u062f\u0631 \u0646\u0638\u0631 \u06af\u0631\u0641\u062a <em>&#8220;\u063a\u0630\u0627 container&#8221;<\/em> \u062f\u0633\u062a\u0647\u060c \u0627\u0645\u0627 \u0627\u0646\u062a\u0632\u0627\u0639 \u0633\u0637\u062d \u0628\u0627\u0644\u0627\u062a\u0631 \u0645\u0648\u0631\u062f \u0646\u06cc\u0627\u0632 \u0627\u0633\u062a \u062a\u0627 \u0628\u0647 \u062f\u0631\u0633\u062a\u06cc \u062d\u062f\u0633 \u0628\u0632\u0646\u06cc\u0645 \u06a9\u0647 \u0622\u06cc\u0627 \u0686\u06cc\u0632\u06cc a \u0627\u0633\u062a \u06cc\u0627 \u062e\u06cc\u0631 <em>&#8220;\u0641\u0646\u062c\u0627\u0646&#8221;<\/em> \u06cc\u0627 \u0627\u0644\u0641 <em>&#8220;\u0645\u06cc \u062a\u0648\u0627\u0646&#8221;<\/em>.<\/p>\n<p>\u0686\u06cc\u0632\u06cc \u06a9\u0647 \u0627\u06cc\u0646 \u06a9\u0627\u0631 \u0631\u0627 \u0633\u062e\u062a\u200c\u062a\u0631 \u0645\u06cc\u200c\u06a9\u0646\u062f \u0627\u06cc\u0646 \u0627\u0633\u062a \u06a9\u0647 CIFAR10 \u062f\u0627\u0631\u0627\u06cc 6000 \u062a\u0635\u0648\u06cc\u0631 \u062f\u0631 \u0647\u0631 \u06a9\u0644\u0627\u0633 \u0627\u0633\u062a\u060c \u062f\u0631 \u062d\u0627\u0644\u06cc \u06a9\u0647 CIFAR100 \u062f\u0627\u0631\u0627\u06cc 600 \u062a\u0635\u0648\u06cc\u0631 \u062f\u0631 \u0647\u0631 \u06a9\u0644\u0627\u0633 \u0627\u0633\u062a \u06a9\u0647 \u0628\u0647 \u0634\u0628\u06a9\u0647 \u062a\u0635\u0627\u0648\u06cc\u0631 \u06a9\u0645\u062a\u0631\u06cc \u0628\u0631\u0627\u06cc \u06cc\u0627\u062f\u06af\u06cc\u0631\u06cc \u062a\u0641\u0627\u0648\u062a\u200c\u0647\u0627\u06cc \u0628\u0633\u06cc\u0627\u0631 \u0638\u0631\u06cc\u0641 \u0645\u06cc\u200c\u062f\u0647\u062f.  \u0641\u0646\u062c\u0627\u0646 \u0647\u0627\u06cc \u0628\u062f\u0648\u0646 \u062f\u0633\u062a\u0647 \u0648\u062c\u0648\u062f \u062f\u0627\u0631\u0646\u062f \u0648 \u0642\u0648\u0637\u06cc \u0647\u0627\u06cc \u0628\u062f\u0648\u0646 \u0628\u0631\u062c\u0633\u062a\u06af\u06cc \u0646\u06cc\u0632 \u0648\u062c\u0648\u062f \u062f\u0627\u0631\u0646\u062f.  \u0627\u0632 \u06cc\u06a9 \u0646\u0645\u0627\u06cc\u0647 &#8211; \u0645\u0645\u06a9\u0646 \u0627\u0633\u062a \u062a\u0634\u062e\u06cc\u0635 \u0622\u0646\u0647\u0627 \u062e\u06cc\u0644\u06cc \u0622\u0633\u0627\u0646 \u0646\u0628\u0627\u0634\u062f.<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/rasanegar.com\/blog\/wp-content\/uploads\/2024\/01\/overfitting-is-your-friend-not-your-foe-3.png\" alt=\"\" title=\"\"><\/p>\n<p>\u0627\u06cc\u0646\u062c\u0627\u0633\u062a \u06a9\u0647 \u0645\u062b\u0644\u0627\u064b \u0627\u0644\u0641 <em>\u067e\u0631\u0633\u067e\u062a\u0631\u0648\u0646 \u0686\u0646\u062f \u0644\u0627\u06cc\u0647<\/em> \u0628\u0647 \u0633\u0627\u062f\u06af\u06cc \u0642\u062f\u0631\u062a \u0627\u0646\u062a\u0632\u0627\u0639\u06cc \u0628\u0631\u0627\u06cc \u06cc\u0627\u062f\u06af\u06cc\u0631\u06cc \u0631\u0627 \u0646\u062f\u0627\u0631\u062f \u0648 \u0645\u062d\u06a9\u0648\u0645 \u0628\u0647 \u0634\u06a9\u0633\u062a \u0627\u0633\u062a \u06a9\u0647 \u0628\u0647 \u0637\u0631\u0632 \u0648\u062d\u0634\u062a\u0646\u0627\u06a9\u06cc \u0646\u0627\u062f\u0631\u0633\u062a \u0627\u0633\u062a. <em>\u0634\u0628\u06a9\u0647 \u0647\u0627\u06cc \u0639\u0635\u0628\u06cc \u06a9\u0627\u0646\u0648\u0644\u0648\u0634\u0646\u0627\u0644<\/em> \u0628\u0631 \u0627\u0633\u0627\u0633 \u0633\u0627\u062e\u062a\u0647 \u0634\u062f\u0647 \u0627\u0646\u062f \u0631\u0648\u06cc \u0631\u0627 <a target=\"_blank\" rel=\"nofollow noopener noreferrer\" href=\"https:\/\/link.springer.com\/article\/10.1007\/BF00344251\">\u0646\u0626\u0648\u06af\u0646\u06cc\u062a\u0631\u0648\u0646<\/a>\u060c \u06a9\u0647 \u0646\u06a9\u0627\u062a\u06cc \u0631\u0627 \u0627\u0632 \u0639\u0644\u0648\u0645 \u0627\u0639\u0635\u0627\u0628 \u0648 \u062a\u0634\u062e\u06cc\u0635 \u0627\u0644\u06af\u0648\u06cc \u0633\u0644\u0633\u0644\u0647 \u0645\u0631\u0627\u062a\u0628\u06cc \u06a9\u0647 \u0645\u063a\u0632 \u0627\u0646\u062c\u0627\u0645 \u0645\u06cc \u062f\u0647\u062f \u06af\u0631\u0641\u062a\u0647 \u0634\u062f\u0647 \u0627\u0633\u062a.  \u0627\u06cc\u0646 \u0634\u0628\u06a9\u0647 \u0647\u0627 \u0645\u06cc \u062a\u0648\u0627\u0646\u0646\u062f \u0648\u06cc\u0698\u06af\u06cc \u0647\u0627\u06cc\u06cc \u0645\u0627\u0646\u0646\u062f \u0627\u06cc\u0646 \u0631\u0627 \u0627\u0633\u062a\u062e\u0631\u0627\u062c \u06a9\u0646\u0646\u062f \u0648 \u062f\u0631 \u0627\u06cc\u0646 \u06a9\u0627\u0631 \u0628\u0631\u062a\u0631\u06cc \u067e\u06cc\u062f\u0627 \u06a9\u0646\u0646\u062f.  \u0628\u0647 \u062d\u062f\u06cc \u06a9\u0647 \u063a\u0627\u0644\u0628\u0627\u064b \u0628\u0647\u200c\u062e\u0648\u0628\u06cc \u0628\u0631 \u0631\u0648\u06cc \u0647\u0645 \u0642\u0631\u0627\u0631 \u0645\u06cc\u200c\u06af\u06cc\u0631\u0646\u062f \u0648 \u062f\u0631 \u0646\u0647\u0627\u06cc\u062a \u0646\u0645\u06cc\u200c\u062a\u0648\u0627\u0646 \u0622\u0646\u200c\u0647\u0627 \u0631\u0627 \u0647\u0645\u0627\u0646\u0637\u0648\u0631 \u06a9\u0647 \u0647\u0633\u062a \u0627\u0633\u062a\u0641\u0627\u062f\u0647 \u06a9\u0631\u062f &#8211; \u062c\u0627\u06cc\u06cc \u06a9\u0647 \u0645\u0627 \u0645\u0639\u0645\u0648\u0644\u0627\u064b \u0628\u0647 \u062e\u0627\u0637\u0631 \u062a\u0648\u0627\u0646\u0627\u06cc\u06cc \u062a\u0639\u0645\u06cc\u0645 \u0628\u062e\u0634\u06cc \u0627\u0632 \u062f\u0642\u062a \u0631\u0627 \u0642\u0631\u0628\u0627\u0646\u06cc \u0645\u06cc\u200c\u06a9\u0646\u06cc\u0645.<\/p>\n<p>\u0628\u06cc\u0627\u06cc\u06cc\u062f \u062f\u0648 \u0645\u0639\u0645\u0627\u0631\u06cc \u0634\u0628\u06a9\u0647 \u0645\u062a\u0641\u0627\u0648\u062a \u0631\u0627 \u0622\u0645\u0648\u0632\u0634 \u062f\u0647\u06cc\u0645 \u0631\u0648\u06cc \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u0647\u0627\u06cc CIFAR10 \u0648 CIFAR100 \u0628\u0647 \u0639\u0646\u0648\u0627\u0646 \u062a\u0635\u0648\u06cc\u0631\u06cc \u0627\u0632 \u0646\u0638\u0631 \u0645\u0646.<\/p>\n<blockquote>\n<p>\u0647\u0645\u0686\u0646\u06cc\u0646 \u062f\u0631 \u0627\u06cc\u0646\u062c\u0627\u0633\u062a \u06a9\u0647 \u0645\u06cc\u200c\u062a\u0648\u0627\u0646\u06cc\u0645 \u0628\u0628\u06cc\u0646\u06cc\u0645 \u06a9\u0647 \u0686\u06af\u0648\u0646\u0647 \u062d\u062a\u06cc \u0632\u0645\u0627\u0646\u06cc \u06a9\u0647 \u06cc\u06a9 \u0634\u0628\u06a9\u0647 \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0628\u0631\u0627\u0632\u0634 \u0645\u06cc\u200c\u06a9\u0646\u062f\u060c \u062a\u0636\u0645\u06cc\u0646\u06cc \u0646\u06cc\u0633\u062a \u06a9\u0647 \u062e\u0648\u062f \u0634\u0628\u06a9\u0647 \u062f\u0631 \u0635\u0648\u0631\u062a \u0633\u0627\u062f\u0647\u200c\u0633\u0627\u0632\u06cc\u060c \u0642\u0637\u0639\u0627\u064b \u0628\u0647 \u062e\u0648\u0628\u06cc \u062a\u0639\u0645\u06cc\u0645 \u0645\u06cc\u200c\u06cc\u0627\u0628\u062f &#8211; \u0627\u06af\u0631 \u0633\u0627\u062f\u0647\u200c\u0633\u0627\u0632\u06cc \u0634\u0648\u062f\u060c \u0645\u0645\u06a9\u0646 \u0627\u0633\u062a \u0642\u0627\u062f\u0631 \u0628\u0647 \u062a\u0639\u0645\u06cc\u0645 \u0646\u0628\u0627\u0634\u062f\u060c \u0627\u06af\u0631\u0686\u0647 \u06af\u0631\u0627\u06cc\u0634\u06cc \u0648\u062c\u0648\u062f \u062f\u0627\u0631\u062f.  \u0634\u0628\u06a9\u0647 \u0645\u0645\u06a9\u0646 \u0627\u0633\u062a \u0628\u0627\u0634\u062f <strong>\u062f\u0631\u0633\u062a<\/strong>\u060c \u0627\u0645\u0627 <strong>\u062f\u0627\u062f\u0647 \u0647\u0627<\/strong> \u0645\u0645\u06a9\u0646 \u0627\u0633\u062a \u06a9\u0627\u0641\u06cc \u0646\u0628\u0627\u0634\u062f<\/p>\n<\/blockquote>\n<p>\u062f\u0631 \u0645\u0648\u0631\u062f CIFAR100 &#8211; \u0641\u0642\u0637 500 \u062a\u0635\u0648\u06cc\u0631 \u0628\u0631\u0627\u06cc \u0622\u0645\u0648\u0632\u0634 (\u0648 100 \u062a\u0635\u0648\u06cc\u0631 \u0628\u0631\u0627\u06cc \u0622\u0632\u0645\u0627\u06cc\u0634) \u062f\u0631 \u0647\u0631 \u06a9\u0644\u0627\u0633 \u0628\u0631\u0627\u06cc \u06cc\u06a9 CNN \u0633\u0627\u062f\u0647 \u06a9\u0627\u0641\u06cc \u0646\u06cc\u0633\u062a. <em>\u0648\u0627\u0642\u0639\u0627<\/em> \u062e\u0648\u0628 \u062a\u0639\u0645\u06cc\u0645 \u062f\u0647\u06cc\u062f \u0631\u0648\u06cc \u06a9\u0644 100 \u06a9\u0644\u0627\u0633\u060c \u0648 \u0645\u0627 \u0628\u0627\u06cc\u062f \u0628\u0631\u0627\u06cc \u06a9\u0645\u06a9 \u0628\u0647 \u0622\u0646\u060c \u0627\u0641\u0632\u0627\u06cc\u0634 \u062f\u0627\u062f\u0647 \u0631\u0627 \u0627\u0646\u062c\u0627\u0645 \u062f\u0647\u06cc\u0645.  \u062d\u062a\u06cc \u0628\u0627 \u0627\u0641\u0632\u0627\u06cc\u0634 \u062f\u0627\u062f\u0647\u200c\u0647\u0627\u060c \u0645\u0645\u06a9\u0646 \u0627\u0633\u062a \u0634\u0628\u06a9\u0647 \u0628\u0633\u06cc\u0627\u0631 \u062f\u0642\u06cc\u0642\u06cc \u0628\u0647 \u062f\u0633\u062a \u0646\u06cc\u0627\u0648\u0631\u06cc\u0645 \u0632\u06cc\u0631\u0627 \u06a9\u0627\u0631\u0647\u0627\u06cc \u0632\u06cc\u0627\u062f\u06cc \u0645\u06cc\u200c\u062a\u0648\u0627\u0646\u06cc\u062f \u0628\u0631\u0627\u06cc \u062f\u0627\u062f\u0647\u200c\u0647\u0627 \u0627\u0646\u062c\u0627\u0645 \u062f\u0647\u06cc\u062f.  \u0627\u06af\u0631 \u0647\u0645\u06cc\u0646 \u0645\u0639\u0645\u0627\u0631\u06cc \u0628\u0647 \u062e\u0648\u0628\u06cc \u0639\u0645\u0644 \u06a9\u0646\u062f \u0631\u0648\u06cc CIFAR10\u060c \u0627\u0645\u0627 \u0646\u0647 CIFAR100 &#8211; \u0628\u0647 \u0627\u06cc\u0646 \u0645\u0639\u0646\u06cc \u0627\u0633\u062a \u06a9\u0647 \u0628\u0647 \u0633\u0627\u062f\u06af\u06cc \u0646\u0645\u06cc \u062a\u0648\u0627\u0646\u062f \u0627\u0632 \u0628\u0631\u062e\u06cc \u0627\u0632 \u062c\u0632\u0626\u06cc\u0627\u062a \u0631\u06cc\u0632\u062f\u0627\u0646\u0647 \u062a\u0631 \u06a9\u0647 \u062a\u0641\u0627\u0648\u062a \u0631\u0627 \u0628\u06cc\u0646 \u0627\u062c\u0633\u0627\u0645 \u0627\u0633\u062a\u0648\u0627\u0646\u0647 \u0627\u06cc \u06a9\u0647 \u0628\u0647 \u0639\u0646\u0648\u0627\u0646 \u0645\u062b\u0627\u0644 &#8220;\u0641\u0646\u062c\u0627\u0646&#8221;\u060c &#8220;\u0642\u0648\u0637\u06cc&#8221; \u0648 &#8220;\u0628\u0637\u0631\u06cc&#8221; \u0645\u06cc \u0646\u0627\u0645\u06cc\u0645\u060c \u0645\u062a\u0645\u0627\u06cc\u0632 \u06a9\u0646\u062f.<\/p>\n<blockquote>\n<p>\u0627\u06cc\u0646 <a rel=\"nofollow noopener noreferrer\" target=\"_blank\" href=\"https:\/\/paperswithcode.com\/sota\/image-classification-on-cifar-100\">\u0627\u06a9\u062b\u0631\u06cc\u062a \u0642\u0631\u06cc\u0628 \u0628\u0647 \u0627\u062a\u0641\u0627\u0642<\/a> \u0627\u0632 \u0645\u0639\u0645\u0627\u0631\u06cc \u0634\u0628\u06a9\u0647 \u0647\u0627\u06cc \u067e\u06cc\u0634\u0631\u0641\u062a\u0647 \u06a9\u0647 \u0628\u0647 \u062f\u0642\u062a \u0628\u0627\u0644\u0627\u06cc\u06cc \u062f\u0633\u062a \u0645\u06cc \u06cc\u0627\u0628\u0646\u062f \u0631\u0648\u06cc \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 CIFAR100 \u0627\u0641\u0632\u0627\u06cc\u0634 \u062f\u0627\u062f\u0647 \u0631\u0627 \u0627\u0646\u062c\u0627\u0645 \u0645\u06cc \u062f\u0647\u062f \u06cc\u0627 \u0645\u062c\u0645\u0648\u0639\u0647 \u0622\u0645\u0648\u0632\u0634\u06cc \u0631\u0627 \u06af\u0633\u062a\u0631\u0634 \u0645\u06cc \u062f\u0647\u062f.<\/p>\n<\/blockquote>\n<p>\u0628\u06cc\u0634\u062a\u0631 \u0622\u0646\u0647\u0627 <em>\u0628\u0627\u06cc\u062f<\/em>\u0648 \u0627\u06cc\u0646 \u0646\u0634\u0627\u0646\u0647 \u0645\u0647\u0646\u062f\u0633\u06cc \u0628\u062f \u0646\u06cc\u0633\u062a.  \u062f\u0631 \u0648\u0627\u0642\u0639 &#8211; \u0627\u06cc\u0646 \u0648\u0627\u0642\u0639\u06cc\u062a \u06a9\u0647 \u0645\u0627 \u0645\u06cc\u200c\u062a\u0648\u0627\u0646\u06cc\u0645 \u0627\u06cc\u0646 \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647\u200c\u0647\u0627 \u0631\u0627 \u06af\u0633\u062a\u0631\u0634 \u062f\u0647\u06cc\u0645 \u0648 \u0628\u0647 \u0634\u0628\u06a9\u0647\u200c\u0647\u0627 \u06a9\u0645\u06a9 \u06a9\u0646\u06cc\u0645 \u062a\u0627 \u0628\u0647\u062a\u0631 \u062a\u0639\u0645\u06cc\u0645 \u067e\u06cc\u062f\u0627 \u06a9\u0646\u0646\u062f\u060c \u0646\u0634\u0627\u0646\u0647\u200c\u0627\u06cc \u0627\u0632 \u0646\u0628\u0648\u063a \u0645\u0647\u0646\u062f\u0633\u06cc \u0627\u0633\u062a.<\/p>\n<p>\u0628\u0647\u200c\u0639\u0644\u0627\u0648\u0647\u060c \u0645\u0646 \u0627\u0632 \u0647\u0631 \u0627\u0646\u0633\u0627\u0646\u06cc \u062f\u0639\u0648\u062a \u0645\u06cc\u200c\u06a9\u0646\u0645 \u0627\u06af\u0631 \u0645\u062a\u0642\u0627\u0639\u062f \u0634\u062f\u0647 \u0627\u0633\u062a \u06a9\u0647 \u0637\u0628\u0642\u0647\u200c\u0628\u0646\u062f\u06cc \u062a\u0635\u0627\u0648\u06cc\u0631 \u0628\u0627 \u062a\u0635\u0627\u0648\u06cc\u0631\u06cc \u0628\u0647 \u06a9\u0648\u0686\u06a9\u06cc 32\u00d732 \u06a9\u0627\u0631 \u0633\u062e\u062a\u06cc \u0646\u06cc\u0633\u062a\u060c \u0627\u0645\u062a\u062d\u0627\u0646 \u06a9\u0646\u062f \u0648 \u062d\u062f\u0633 \u0628\u0632\u0646\u062f \u06a9\u0647 \u0627\u06cc\u0646\u0647\u0627 \u0686\u06cc\u0633\u062a\u0646\u062f:<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/rasanegar.com\/blog\/wp-content\/uploads\/2024\/01\/overfitting-is-your-friend-not-your-foe-4.png\" alt=\"\" title=\"\"><\/p>\n<p>\u0627\u0633\u062a <em>\u062a\u0635\u0648\u06cc\u0631 4<\/em> \u0686\u0646\u062f \u067e\u0631\u062a\u0642\u0627\u0644\u061f  \u062a\u0648\u067e \u0647\u0627\u06cc \u067e\u06cc\u0646\u06af \u067e\u0646\u06af\u061f  \u0632\u0631\u062f\u0647 \u062a\u062e\u0645 \u0645\u0631\u063a\u061f  \u062e\u0648\u0628\u060c \u0627\u062d\u062a\u0645\u0627\u0644\u0627\u064b \u0632\u0631\u062f\u0647 \u062a\u062e\u0645 \u0645\u0631\u063a \u0646\u06cc\u0633\u062a\u060c \u0627\u0645\u0627 \u0627\u06cc\u0646 \u0646\u06cc\u0627\u0632 \u0628\u0647 \u062f\u0627\u0646\u0634 \u0642\u0628\u0644\u06cc \u062f\u0627\u0631\u062f \u0631\u0648\u06cc &#8220;\u062a\u062e\u0645 \u0645\u0631\u063a&#8221; \u0686\u06cc\u0633\u062a \u0648 \u0622\u06cc\u0627 \u0627\u062d\u062a\u0645\u0627\u0644 \u062f\u0627\u0631\u062f \u0632\u0631\u062f\u0647 \u0647\u0627 \u0631\u0627 \u062f\u0631 \u062d\u0627\u0644\u062a \u0646\u0634\u0633\u062a\u0647 \u067e\u06cc\u062f\u0627 \u06a9\u0646\u06cc\u062f \u0631\u0648\u06cc \u062c\u062f\u0648\u0644\u06cc \u06a9\u0647 \u06cc\u06a9 \u0634\u0628\u06a9\u0647 \u0646\u062e\u0648\u0627\u0647\u062f \u062f\u0627\u0634\u062a.  \u0645\u06cc\u0632\u0627\u0646 \u062f\u0627\u0646\u0634 \u0642\u0628\u0644\u06cc \u06a9\u0647 \u0645\u0645\u06a9\u0646 \u0627\u0633\u062a \u062f\u0631 \u0645\u0648\u0631\u062f \u062c\u0647\u0627\u0646 \u062f\u0627\u0634\u062a\u0647 \u0628\u0627\u0634\u06cc\u062f \u0648 \u0645\u06cc\u0632\u0627\u0646 \u062a\u0623\u062b\u06cc\u0631 \u0622\u0646 \u0628\u0631 \u0622\u0646\u0686\u0647 \u06a9\u0647 \u0645\u06cc \u0628\u06cc\u0646\u06cc\u062f \u0631\u0627 \u062f\u0631 \u0646\u0638\u0631 \u0628\u06af\u06cc\u0631\u06cc\u062f.<\/p>\n<h3 id=\"importingthedata\"><span class=\"ez-toc-section\" id=\"%d9%88%d8%a7%d8%b1%d8%af_%da%a9%d8%b1%d8%af%d9%86_%d8%af%d8%a7%d8%af%d9%87_%d9%87%d8%a7\"><\/span>\u0648\u0627\u0631\u062f \u06a9\u0631\u062f\u0646 \u062f\u0627\u062f\u0647 \u0647\u0627<span class=\"ez-toc-section-end\"><\/span><\/h3>\n<p>\u0645\u0627 \u0627\u0632 Keras \u0628\u0647\u200c\u0639\u0646\u0648\u0627\u0646 \u06a9\u062a\u0627\u0628\u062e\u0627\u0646\u0647 \u06cc\u0627\u062f\u06af\u06cc\u0631\u06cc \u0639\u0645\u06cc\u0642 \u0627\u0646\u062a\u062e\u0627\u0628\u06cc \u0627\u0633\u062a\u0641\u0627\u062f\u0647 \u062e\u0648\u0627\u0647\u06cc\u0645 \u06a9\u0631\u062f\u060c \u0627\u0645\u0627 \u062f\u0631 \u0635\u0648\u0631\u062a \u062a\u0645\u0627\u06cc\u0644 \u0645\u06cc\u200c\u062a\u0648\u0627\u0646\u06cc\u062f \u06a9\u062a\u0627\u0628\u062e\u0627\u0646\u0647\u200c\u0647\u0627\u06cc \u062f\u06cc\u06af\u0631 \u06cc\u0627 \u062d\u062a\u06cc \u0645\u062f\u0644\u200c\u0647\u0627\u06cc \u0633\u0641\u0627\u0631\u0634\u06cc \u062e\u0648\u062f \u0631\u0627 \u062f\u0646\u0628\u0627\u0644 \u06a9\u0646\u06cc\u062f.<\/p>\n<p>\u0627\u0645\u0627 \u0627\u0628\u062a\u062f\u0627\u060c \u0627\u062c\u0627\u0632\u0647 \u062f\u0647\u06cc\u062f \u0622\u0646 \u0631\u0627 \u0628\u0627\u0631\u06af\u0630\u0627\u0631\u06cc \u06a9\u0646\u06cc\u0645\u060c \u062f\u0627\u062f\u0647 \u0647\u0627 \u0631\u0627 \u062f\u0631 \u06cc\u06a9 \u0645\u062c\u0645\u0648\u0639\u0647 \u0622\u0645\u0648\u0632\u0634\u06cc\u060c \u0622\u0632\u0645\u0627\u06cc\u0634\u06cc \u0648 \u0627\u0639\u062a\u0628\u0627\u0631 \u0633\u0646\u062c\u06cc \u062c\u062f\u0627 \u06a9\u0646\u06cc\u0645\u060c \u0648 \u0645\u0642\u0627\u062f\u06cc\u0631 \u062a\u0635\u0648\u06cc\u0631 \u0631\u0627 \u0639\u0627\u062f\u06cc \u06a9\u0646\u06cc\u0645. <code>0..1<\/code>:<\/p>\n<pre><code class=\"hljs\"><span class=\"hljs-keyword\">from<\/span> tensorflow <span class=\"hljs-keyword\">import<\/span> keras\n<span class=\"hljs-keyword\">import<\/span> numpy <span class=\"hljs-keyword\">as<\/span> np\n<span class=\"hljs-keyword\">import<\/span> matplotlib.pyplot <span class=\"hljs-keyword\">as<\/span> plt\n\n\n(X_train_full, Y_train_full), (X_test, Y_test) = keras.datasets.cifar10.load_data()\n\nX_valid, X_train = X_train_full(:<span class=\"hljs-number\">5000<\/span>)\/<span class=\"hljs-number\">255.0<\/span>, X_train_full(<span class=\"hljs-number\">5000<\/span>:)\/<span class=\"hljs-number\">255.0<\/span>\nY_valid, Y_train = Y_train_full(:<span class=\"hljs-number\">5000<\/span>), Y_train_full(<span class=\"hljs-number\">5000<\/span>:)\n\nX_test = X_test\/<span class=\"hljs-number\">255.0<\/span>\n<\/code><\/pre>\n<p>\u0633\u067e\u0633\u060c \u0627\u062c\u0627\u0632\u0647 \u062f\u0647\u06cc\u062f \u0628\u0631\u062e\u06cc \u0627\u0632 \u062a\u0635\u0627\u0648\u06cc\u0631 \u0645\u0648\u062c\u0648\u062f \u062f\u0631 \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u0631\u0627 \u062a\u062c\u0633\u0645 \u06a9\u0646\u06cc\u0645 \u062a\u0627 \u0627\u06cc\u062f\u0647 \u0627\u06cc \u062f\u0631 \u0645\u0648\u0631\u062f \u0622\u0646\u0686\u0647 \u06a9\u0647 \u062f\u0631 \u0645\u0642\u0627\u0628\u0644 \u0622\u0646 \u0642\u0631\u0627\u0631 \u062f\u0627\u0631\u06cc\u0645 \u0628\u0647 \u062f\u0633\u062a \u0622\u0648\u0631\u06cc\u0645:<\/p>\n<pre><code class=\"hljs\">fig, ax = plt.subplots(<span class=\"hljs-number\">5<\/span>, <span class=\"hljs-number\">5<\/span>, figsize=(<span class=\"hljs-number\">10<\/span>, <span class=\"hljs-number\">10<\/span>))\nax = ax.ravel()\n\n\nclass_names = (<span class=\"hljs-string\">'Airplane'<\/span>, <span class=\"hljs-string\">'Automobile'<\/span>, <span class=\"hljs-string\">'Bird'<\/span>, <span class=\"hljs-string\">'Cat'<\/span>, <span class=\"hljs-string\">'Deer'<\/span>, <span class=\"hljs-string\">'Dog'<\/span>, <span class=\"hljs-string\">'Frog'<\/span>, <span class=\"hljs-string\">'Horse'<\/span>, <span class=\"hljs-string\">'Ship'<\/span>, <span class=\"hljs-string\">'Truck'<\/span>)\n\n<span class=\"hljs-keyword\">for<\/span> i <span class=\"hljs-keyword\">in<\/span> <span class=\"hljs-built_in\">range<\/span>(<span class=\"hljs-number\">25<\/span>):\n    ax(i).imshow(X_train_full(i))\n    ax(i).set_title(class_names(Y_train_full(i)(<span class=\"hljs-number\">0<\/span>)))\n    ax(i).axis(<span class=\"hljs-string\">'off'<\/span>)\n    plt.subplots_adjust(wspace=<span class=\"hljs-number\">1<\/span>) \n\nplt.show()\n<\/code><\/pre>\n<p><img decoding=\"async\" src=\"https:\/\/rasanegar.com\/blog\/wp-content\/uploads\/2024\/01\/overfitting-is-your-friend-not-your-foe-5.png\" alt=\"\" title=\"\"><\/p>\n<h3 id=\"underfittingmultilayerperceptron\"><span class=\"ez-toc-section\" id=\"%d8%b9%d8%af%d9%85_%d8%aa%d9%86%d8%a7%d8%b3%d8%a8_%d9%be%d8%b1%d8%b3%d9%be%d8%aa%d8%b1%d9%88%d9%86_%da%86%d9%86%d8%af%d9%84%d8%a7%db%8c%d9%87\"><\/span>\u0639\u062f\u0645 \u062a\u0646\u0627\u0633\u0628 \u067e\u0631\u0633\u067e\u062a\u0631\u0648\u0646 \u0686\u0646\u062f\u0644\u0627\u06cc\u0647<span class=\"ez-toc-section-end\"><\/span><\/h3>\n<p>\u062a\u0642\u0631\u06cc\u0628\u0627\u064b \u0645\u0647\u0645 \u0646\u06cc\u0633\u062a \u06a9\u0647 \u0645\u0627 \u0686\u0647 \u06a9\u0627\u0631\u06cc \u0627\u0646\u062c\u0627\u0645 \u0645\u06cc \u062f\u0647\u06cc\u0645\u060c MLP \u0622\u0646\u0642\u062f\u0631 \u062e\u0648\u0628 \u0639\u0645\u0644 \u0646\u0645\u06cc \u06a9\u0646\u062f.  \u0642\u0637\u0639\u0627\u064b \u0628\u0647 \u0633\u0637\u062d\u06cc \u0627\u0632 \u062f\u0642\u062a \u0628\u0631 \u0627\u0633\u0627\u0633 \u062e\u0648\u0627\u0647\u062f \u0631\u0633\u06cc\u062f \u0631\u0648\u06cc \u062a\u0648\u0627\u0644\u06cc \u062e\u0627\u0645 \u0627\u0637\u0644\u0627\u0639\u0627\u062a\u06cc \u06a9\u0647 \u0648\u0627\u0631\u062f \u0645\u06cc \u0634\u0648\u062f &#8211; \u0627\u0645\u0627 \u0627\u06cc\u0646 \u062a\u0639\u062f\u0627\u062f \u0645\u062d\u062f\u0648\u062f \u0627\u0633\u062a \u0648 \u0627\u062d\u062a\u0645\u0627\u0644\u0627\u064b \u062e\u06cc\u0644\u06cc \u0632\u06cc\u0627\u062f \u0646\u062e\u0648\u0627\u0647\u062f \u0628\u0648\u062f.<\/p>\n<p>\u0634\u0628\u06a9\u0647 \u062f\u0631 \u06cc\u06a9 \u0646\u0642\u0637\u0647 \u0634\u0631\u0648\u0639 \u0628\u0647 \u0646\u0635\u0628 \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0645\u06cc\u200c\u06a9\u0646\u062f \u0648 \u062a\u0648\u0627\u0644\u06cc\u200c\u0647\u0627\u06cc \u0645\u0634\u062e\u0635 \u062f\u0627\u062f\u0647\u200c\u0647\u0627 \u0631\u0627 \u0646\u0634\u0627\u0646 \u0645\u06cc\u200c\u062f\u0647\u062f\u060c \u0627\u0645\u0627 \u0647\u0645\u0686\u0646\u0627\u0646 \u062f\u0642\u062a \u067e\u0627\u06cc\u06cc\u0646\u06cc \u062e\u0648\u0627\u0647\u062f \u062f\u0627\u0634\u062a. \u0631\u0648\u06cc \u0645\u062c\u0645\u0648\u0639\u0647 \u0622\u0645\u0648\u0632\u0634\u06cc \u062d\u062a\u06cc \u062f\u0631 \u0647\u0646\u06af\u0627\u0645 \u0646\u0635\u0628 \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f\u060c \u06a9\u0647 \u0628\u0647\u062a\u0631\u06cc\u0646 \u0632\u0645\u0627\u0646 \u0628\u0631\u0627\u06cc \u0645\u062a\u0648\u0642\u0641 \u06a9\u0631\u062f\u0646 \u0622\u0645\u0648\u0632\u0634 \u0622\u0646 \u0627\u0633\u062a\u060c \u0632\u06cc\u0631\u0627 \u0628\u0647 \u0633\u0627\u062f\u06af\u06cc \u0646\u0645\u06cc \u062a\u0648\u0627\u0646\u062f \u0628\u0647 \u062e\u0648\u0628\u06cc \u0628\u0627 \u062f\u0627\u062f\u0647 \u0647\u0627 \u0633\u0627\u0632\u06af\u0627\u0631 \u0634\u0648\u062f. <small>\u0645\u06cc \u062f\u0627\u0646\u06cc\u062f \u0634\u0628\u06a9\u0647 \u0647\u0627\u06cc \u0622\u0645\u0648\u0632\u0634\u06cc \u0631\u062f\u067e\u0627\u06cc \u06a9\u0631\u0628\u0646 \u062f\u0627\u0631\u0646\u062f.<\/small><\/p>\n<p>\u0628\u06cc\u0627\u06cc\u06cc\u062f \u06cc\u06a9 \u0631\u0627 \u0627\u0636\u0627\u0641\u0647 \u06a9\u0646\u06cc\u0645 <code>EarlyStopping<\/code> \u0628\u0631\u0627\u06cc \u062c\u0644\u0648\u06af\u06cc\u0631\u06cc \u0627\u0632 \u0627\u062c\u0631\u0627\u06cc \u0634\u0628\u06a9\u0647 \u0641\u0631\u0627\u062a\u0631 \u0627\u0632 \u0646\u0642\u0637\u0647 \u0639\u0642\u0644 \u0633\u0644\u06cc\u0645\u060c \u0648 \u062a\u0646\u0638\u06cc\u0645 \u06a9\u0646\u06cc\u062f <code>epochs<\/code> \u0628\u0647 \u0639\u062f\u062f\u06cc \u0641\u0631\u0627\u062a\u0631 \u0627\u0632 \u0622\u0646\u0686\u0647 \u06a9\u0647 \u0622\u0646 \u0631\u0627 \u0627\u062c\u0631\u0627 \u062e\u0648\u0627\u0647\u06cc\u0645 \u06a9\u0631\u062f (\u0628\u0646\u0627\u0628\u0631\u0627\u06cc\u0646 <code>EarlyStopping<\/code> \u0645\u06cc \u062a\u0648\u0627\u0646\u062f \u0648\u0627\u0631\u062f \u0634\u0648\u062f).<\/p>\n<p>\u0645\u0627 \u0627\u0632 Sequential API \u0628\u0631\u0627\u06cc \u0627\u0636\u0627\u0641\u0647 \u06a9\u0631\u062f\u0646 \u0686\u0646\u062f \u0644\u0627\u06cc\u0647 \u0627\u0633\u062a\u0641\u0627\u062f\u0647 \u062e\u0648\u0627\u0647\u06cc\u0645 \u06a9\u0631\u062f <code>BatchNormalization<\/code> \u0648 \u06a9\u0645\u06cc \u0627\u0632 <code>Dropout<\/code>.  \u0622\u0646\u0647\u0627 \u0628\u0647 \u062a\u0639\u0645\u06cc\u0645 \u06a9\u0645\u06a9 \u0645\u06cc \u06a9\u0646\u0646\u062f \u0648 \u0645\u0627 \u062d\u062f\u0627\u0642\u0644 \u0645\u06cc \u062e\u0648\u0627\u0647\u06cc\u0645 <em>\u062a\u0644\u0627\u0634 \u06a9\u0631\u062f\u0646<\/em> \u0628\u0631\u0627\u06cc \u0627\u06cc\u0646\u06a9\u0647 \u0627\u06cc\u0646 \u0645\u062f\u0644 \u0686\u06cc\u0632\u06cc \u06cc\u0627\u062f \u0628\u06af\u06cc\u0631\u062f.<\/p>\n<p>\u0647\u0627\u06cc\u067e\u0631\u067e\u0627\u0631\u0627\u0645\u062a\u0631\u0647\u0627\u06cc \u0627\u0635\u0644\u06cc \u06a9\u0647 \u0645\u06cc \u062a\u0648\u0627\u0646\u06cc\u0645 \u062f\u0631 \u0627\u06cc\u0646\u062c\u0627 \u062a\u063a\u06cc\u06cc\u0631 \u062f\u0647\u06cc\u0645 \u0639\u0628\u0627\u0631\u062a\u0646\u062f \u0627\u0632: \u062a\u0639\u062f\u0627\u062f \u0644\u0627\u06cc\u0647 \u0647\u0627\u060c \u0627\u0646\u062f\u0627\u0632\u0647 \u0622\u0646\u0647\u0627\u060c \u062a\u0648\u0627\u0628\u0639 \u0641\u0639\u0627\u0644 \u0633\u0627\u0632\u06cc\u060c \u0645\u0642\u062f\u0627\u0631\u062f\u0647\u06cc \u0627\u0648\u0644\u06cc\u0647 \u0647\u0633\u062a\u0647 \u0648 \u0646\u0631\u062e \u0627\u0646\u0635\u0631\u0627\u0641\u060c \u0648 \u062f\u0631 \u0627\u06cc\u0646\u062c\u0627 \u06cc\u06a9 \u062a\u0646\u0638\u06cc\u0645 &#8220;\u0645\u0646\u0627\u0633\u0628&#8221; \u0648\u062c\u0648\u062f \u062f\u0627\u0631\u062f:<\/p>\n<pre><code class=\"hljs\">checkpoint = keras.callbacks.ModelCheckpoint(<span class=\"hljs-string\">\"simple_dense.h5\"<\/span>, save_best_only=<span class=\"hljs-literal\">True<\/span>)\nearly_stopping = keras.callbacks.EarlyStopping(patience=<span class=\"hljs-number\">10<\/span>, restore_best_weights=<span class=\"hljs-literal\">True<\/span>)\n\nmodel = keras.Sequential((\n  keras.layers.Flatten(input_shape=(<span class=\"hljs-number\">32<\/span>, <span class=\"hljs-number\">32<\/span>, <span class=\"hljs-number\">3<\/span>)),\n  keras.layers.BatchNormalization(),\n  keras.layers.Dense(<span class=\"hljs-number\">75<\/span>),\n    \n  keras.layers.Dense((<span class=\"hljs-number\">50<\/span>), activation=<span class=\"hljs-string\">'elu'<\/span>),\n  keras.layers.BatchNormalization(),\n  keras.layers.Dropout(<span class=\"hljs-number\">0.1<\/span>),\n    \n  keras.layers.Dense((<span class=\"hljs-number\">50<\/span>), activation=<span class=\"hljs-string\">'elu'<\/span>),\n  keras.layers.BatchNormalization(),\n  keras.layers.Dropout(<span class=\"hljs-number\">0.1<\/span>),\n    \n  keras.layers.Dense(<span class=\"hljs-number\">10<\/span>, activation=<span class=\"hljs-string\">'softmax'<\/span>)\n))\n\nmodel.<span class=\"hljs-built_in\">compile<\/span>(loss=<span class=\"hljs-string\">\"sparse_categorical_crossentropy\"<\/span>,\n              optimizer=keras.optimizers.Nadam(learning_rate=<span class=\"hljs-number\">1e-4<\/span>),\n              metrics=(<span class=\"hljs-string\">\"accuracy\"<\/span>))\n\nhistory = model.fit(X_train, \n                    Y_train, \n                    epochs=<span class=\"hljs-number\">150<\/span>, \n                    validation_data=(X_valid, Y_valid),\n                    callbacks=(checkpoint, early_stopping))\n<\/code><\/pre>\n<p>\u0628\u06cc\u0627\u06cc\u06cc\u062f \u0628\u0628\u06cc\u0646\u06cc\u0645 \u0622\u06cc\u0627 \u0641\u0631\u0636\u06cc\u0647 \u0622\u063a\u0627\u0632\u06cc\u0646 \u062f\u0631\u0633\u062a \u0627\u0633\u062a \u06cc\u0627 \u062e\u06cc\u0631 &#8211; \u0634\u0631\u0648\u0639 \u0628\u0647 \u06cc\u0627\u062f\u06af\u06cc\u0631\u06cc \u0648 \u062a\u0639\u0645\u06cc\u0645 \u062a\u0627 \u062d\u062f\u06cc \u0645\u06cc \u06a9\u0646\u062f\u060c \u0627\u0645\u0627 \u062f\u0631 \u0646\u0647\u0627\u06cc\u062a \u062f\u0642\u062a \u067e\u0627\u06cc\u06cc\u0646\u06cc \u062e\u0648\u0627\u0647\u062f \u062f\u0627\u0634\u062a. \u0631\u0648\u06cc \u0647\u0645 \u0645\u062c\u0645\u0648\u0639\u0647 \u0622\u0645\u0648\u0632\u0634\u06cc \u0648 \u0647\u0645 \u0645\u062c\u0645\u0648\u0639\u0647 \u062a\u0633\u062a \u0648 \u0627\u0639\u062a\u0628\u0627\u0631 \u0633\u0646\u062c\u06cc\u060c \u06a9\u0647 \u0645\u0646\u062c\u0631 \u0628\u0647 \u062f\u0642\u062a \u06a9\u0644\u06cc \u067e\u0627\u06cc\u06cc\u0646 \u0645\u06cc \u0634\u0648\u062f.<\/p>\n<p>\u0628\u0631\u0627\u06cc CIFAR10\u060c \u0634\u0628\u06a9\u0647 \u0628\u0647 \u062e\u0648\u0628\u06cc \u0639\u0645\u0644 \u0645\u06cc \u06a9\u0646\u062f:<\/p>\n<pre><code class=\"hljs\">Epoch <span class=\"hljs-number\">1<\/span>\/<span class=\"hljs-number\">150<\/span>\n<span class=\"hljs-number\">1407<\/span>\/<span class=\"hljs-number\">1407<\/span> (==============================) - 5s 3ms\/step - loss: <span class=\"hljs-number\">1.9706<\/span> - accuracy: <span class=\"hljs-number\">0.3108<\/span> - val_loss: <span class=\"hljs-number\">1.6841<\/span> - val_accuracy: <span class=\"hljs-number\">0.4100<\/span>\n...\nEpoch <span class=\"hljs-number\">50<\/span>\/<span class=\"hljs-number\">150<\/span>\n<span class=\"hljs-number\">1407<\/span>\/<span class=\"hljs-number\">1407<\/span> (==============================) - 4s 3ms\/step - loss: <span class=\"hljs-number\">1.2927<\/span> - accuracy: <span class=\"hljs-number\">0.5403<\/span> - val_loss: <span class=\"hljs-number\">1.3893<\/span> - val_accuracy: <span class=\"hljs-number\">0.5122<\/span>\n<\/code><\/pre>\n<p>\u0628\u06cc\u0627\u06cc\u06cc\u062f \u0646\u06af\u0627\u0647\u06cc \u0628\u0647 \u062a\u0627\u0631\u06cc\u062e\u0686\u0647 \u06cc\u0627\u062f\u06af\u06cc\u0631\u06cc \u0622\u0646 \u0628\u06cc\u0646\u062f\u0627\u0632\u06cc\u0645:<\/p>\n<pre><code class=\"hljs\">pd.DataFrame(history.history).plot()\nplt.show()\n\nmodel.evaluate(X_test, Y_test)\n<\/code><\/pre>\n<p><img decoding=\"async\" src=\"https:\/\/rasanegar.com\/blog\/wp-content\/uploads\/2024\/01\/overfitting-is-your-friend-not-your-foe-6.png\" alt=\"\" title=\"\"><\/p>\n<pre><code class=\"hljs\">313\/313 (==============================) - 0s 926us\/step - loss: 1.3836 - accuracy: 0.5058\n(1.383605718612671, 0.5058000087738037)\n<\/code><\/pre>\n<p>\u062f\u0642\u062a \u06a9\u0644\u06cc \u062a\u0627 50% \u0645\u06cc \u0631\u0633\u062f \u0648 \u0634\u0628\u06a9\u0647 \u062e\u06cc\u0644\u06cc \u0633\u0631\u06cc\u0639 \u0628\u0647 \u0627\u06cc\u0646\u062c\u0627 \u0645\u06cc \u0631\u0633\u062f \u0648 \u0634\u0631\u0648\u0639 \u0628\u0647 \u0641\u0644\u0627\u062a \u0645\u06cc \u06a9\u0646\u062f.  5\/10 \u062a\u0635\u0627\u0648\u06cc\u0631 \u0628\u0647 \u062f\u0631\u0633\u062a\u06cc \u0637\u0628\u0642\u0647 \u0628\u0646\u062f\u06cc \u0634\u062f\u0647 \u0628\u0647 \u0646\u0638\u0631 \u0645\u06cc \u0631\u0633\u062f \u0645\u0627\u0646\u0646\u062f \u067e\u0631\u062a\u0627\u0628 \u06cc\u06a9 \u0633\u06a9\u0647 \u060c \u0627\u0645\u0627 \u0628\u0647 \u06cc\u0627\u062f \u062f\u0627\u0634\u062a\u0647 \u0628\u0627\u0634\u06cc\u062f \u06a9\u0647 \u062f\u0631 \u0627\u06cc\u0646\u062c\u0627 10 \u06a9\u0644\u0627\u0633 \u0648\u062c\u0648\u062f \u062f\u0627\u0631\u062f \u060c \u0628\u0646\u0627\u0628\u0631\u0627\u06cc\u0646 \u0627\u06af\u0631 \u0628\u0647 \u0637\u0648\u0631 \u062a\u0635\u0627\u062f\u0641\u06cc \u062d\u062f\u0633 \u0645\u06cc \u0632\u062f\u0646\u062f \u060c \u0645\u06cc \u062a\u0648\u0627\u0646\u0633\u062a \u0631\u0648\u06cc \u062d\u062f\u0633 \u0645\u062a\u0648\u0633\u0637 \u200b\u200b\u06cc\u06a9 \u062a\u0635\u0648\u06cc\u0631 \u0627\u0632 \u062f\u0647.  \u0628\u06cc\u0627\u06cc\u06cc\u062f \u0628\u0647 \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 CIFAR100 \u0628\u0631\u0648\u06cc\u0645\u060c \u06a9\u0647 \u0647\u0645\u0686\u0646\u06cc\u0646 \u0628\u0647 \u0634\u0628\u06a9\u0647 \u0627\u06cc \u0628\u0627 \u062d\u062f\u0627\u0642\u0644 \u06a9\u0645\u06cc \u0642\u062f\u0631\u062a \u0628\u06cc\u0634\u062a\u0631 \u0646\u06cc\u0627\u0632 \u062f\u0627\u0631\u062f\u060c \u0632\u06cc\u0631\u0627 \u0646\u0645\u0648\u0646\u0647 \u0647\u0627\u06cc \u0622\u0645\u0648\u0632\u0634\u06cc \u06a9\u0645\u062a\u0631\u06cc \u062f\u0631 \u0647\u0631 \u06a9\u0644\u0627\u0633 \u0648\u062c\u0648\u062f \u062f\u0627\u0631\u062f \u0648 \u0647\u0645\u0686\u0646\u06cc\u0646 \u062a\u0639\u062f\u0627\u062f \u06a9\u0644\u0627\u0633 \u0647\u0627\u06cc \u0628\u0633\u06cc\u0627\u0631 \u0628\u0627\u0644\u0627\u062a\u0631\u06cc \u0648\u062c\u0648\u062f \u062f\u0627\u0631\u062f:<\/p>\n<pre><code class=\"hljs\">checkpoint = keras.callbacks.ModelCheckpoint(<span class=\"hljs-string\">\"bigger_dense.h5\"<\/span>, save_best_only=<span class=\"hljs-literal\">True<\/span>)\nearly_stopping = keras.callbacks.EarlyStopping(patience=<span class=\"hljs-number\">10<\/span>, restore_best_weights=<span class=\"hljs-literal\">True<\/span>)\n\n\n(X_train_full, Y_train_full), (X_test, Y_test) = keras.datasets.cifar100.load_data()\n\n\nmodel1 = keras.Sequential((\n  keras.layers.Flatten(input_shape=(<span class=\"hljs-number\">32<\/span>, <span class=\"hljs-number\">32<\/span>, <span class=\"hljs-number\">3<\/span>)),\n  keras.layers.BatchNormalization(),\n  keras.layers.Dense(<span class=\"hljs-number\">256<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, kernel_initializer=<span class=\"hljs-string\">\"he_normal\"<\/span>),\n    \n  keras.layers.Dense(<span class=\"hljs-number\">128<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>),\n  keras.layers.BatchNormalization(),\n  keras.layers.Dropout(<span class=\"hljs-number\">0.1<\/span>),\n\n  keras.layers.Dense(<span class=\"hljs-number\">100<\/span>, activation=<span class=\"hljs-string\">'softmax'<\/span>)\n))\n\n\nmodel1.<span class=\"hljs-built_in\">compile<\/span>(loss=<span class=\"hljs-string\">\"sparse_categorical_crossentropy\"<\/span>,\n              optimizer=keras.optimizers.Nadam(learning_rate=<span class=\"hljs-number\">1e-4<\/span>),\n              metrics=(<span class=\"hljs-string\">\"accuracy\"<\/span>))\n\nhistory = model1.fit(X_train, \n                    Y_train, \n                    epochs=<span class=\"hljs-number\">150<\/span>, \n                    validation_data=(X_valid, Y_valid),\n                    callbacks=(checkpoint, early_stopping))\n<\/code><\/pre>\n<p>\u0639\u0645\u0644\u06a9\u0631\u062f \u0634\u0628\u06a9\u0647 \u0646\u0633\u0628\u062a\u0627\u064b \u0628\u062f \u0627\u0633\u062a:<\/p>\n<pre><code class=\"hljs\">Epoch <span class=\"hljs-number\">1<\/span>\/<span class=\"hljs-number\">150<\/span>\n<span class=\"hljs-number\">1407<\/span>\/<span class=\"hljs-number\">1407<\/span> (==============================) - 13s 9ms\/step - loss: <span class=\"hljs-number\">4.2260<\/span> - accuracy: <span class=\"hljs-number\">0.0836<\/span> - val_loss: <span class=\"hljs-number\">3.8682<\/span> - val_accuracy: <span class=\"hljs-number\">0.1238<\/span>\n...\nEpoch <span class=\"hljs-number\">24<\/span>\/<span class=\"hljs-number\">150<\/span>\n<span class=\"hljs-number\">1407<\/span>\/<span class=\"hljs-number\">1407<\/span> (==============================) - 12s 8ms\/step - loss: <span class=\"hljs-number\">2.3598<\/span> - accuracy: <span class=\"hljs-number\">0.4006<\/span> - val_loss: <span class=\"hljs-number\">3.3577<\/span> - val_accuracy: <span class=\"hljs-number\">0.2434<\/span>\n<\/code><\/pre>\n<p>\u0648 \u0628\u06cc\u0627\u06cc\u06cc\u062f \u062a\u0627\u0631\u06cc\u062e\u0686\u0647 \u067e\u06cc\u0634\u0631\u0641\u062a \u0622\u0646 \u0631\u0627 \u062a\u0631\u0633\u06cc\u0645 \u06a9\u0646\u06cc\u0645 \u0648 \u0647\u0645\u0686\u0646\u06cc\u0646 \u0622\u0646 \u0631\u0627 \u0627\u0631\u0632\u06cc\u0627\u0628\u06cc \u06a9\u0646\u06cc\u0645 \u0631\u0648\u06cc \u0645\u062c\u0645\u0648\u0639\u0647 \u062a\u0633\u062a (\u06a9\u0647 \u0627\u062d\u062a\u0645\u0627\u0644\u0627\u064b \u0628\u0647 \u062e\u0648\u0628\u06cc \u0645\u062c\u0645\u0648\u0639\u0647 \u0627\u0639\u062a\u0628\u0627\u0631 \u0633\u0646\u062c\u06cc \u0639\u0645\u0644 \u0645\u06cc \u06a9\u0646\u062f):<\/p>\n<pre><code class=\"hljs\">pd.DataFrame(history.history).plot()\nplt.show()\n\nmodel.evaluate(X_test, Y_test)\n<\/code><\/pre>\n<p><img decoding=\"async\" src=\"https:\/\/rasanegar.com\/blog\/wp-content\/uploads\/2024\/01\/overfitting-is-your-friend-not-your-foe-7.png\" alt=\"\" title=\"\"><\/p>\n<pre><code class=\"hljs\">313\/313 (==============================) - 0s 2ms\/step - loss: 3.2681 - accuracy: 0.2408\n(3.2681326866149902, 0.24079999327659607)\n<\/code><\/pre>\n<p>\u0647\u0645\u0627\u0646\u0637\u0648\u0631 \u06a9\u0647 \u0627\u0646\u062a\u0638\u0627\u0631 \u0645\u06cc \u0631\u0641\u062a\u060c \u0634\u0628\u06a9\u0647 \u0646\u062a\u0648\u0627\u0646\u0633\u062a \u062f\u0627\u062f\u0647 \u0647\u0627 \u0631\u0627 \u0628\u0647 \u062e\u0648\u0628\u06cc \u062f\u0631\u06a9 \u06a9\u0646\u062f.  \u062f\u0631 \u0646\u0647\u0627\u06cc\u062a \u062f\u0642\u062a \u0627\u0636\u0627\u0641\u0647 40% \u0648 \u062f\u0642\u062a \u0648\u0627\u0642\u0639\u06cc 24% \u062f\u0627\u0634\u062a.<\/p>\n<p>\u062f\u0642\u062a \u0628\u0647 40% \u0645\u062d\u062f\u0648\u062f \u0634\u062f &#8211; \u0627\u06cc\u0646\u0637\u0648\u0631 \u0646\u0628\u0648\u062f <em>\u0648\u0627\u0642\u0639\u0627<\/em> \u0642\u0627\u062f\u0631 \u0628\u0647 \u062a\u0637\u0628\u06cc\u0642 \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u0647\u0627 \u0627\u0633\u062a\u060c \u062d\u062a\u06cc \u0627\u06af\u0631 \u0628\u0627 \u0628\u0631\u062e\u06cc \u0627\u0632 \u0642\u0633\u0645\u062a \u0647\u0627\u06cc \u0622\u0646 \u06a9\u0647 \u0628\u0627 \u062a\u0648\u062c\u0647 \u0628\u0647 \u0645\u0639\u0645\u0627\u0631\u06cc \u0645\u062d\u062f\u0648\u062f \u0642\u0627\u062f\u0631 \u0628\u0647 \u062a\u0634\u062e\u06cc\u0635 \u0622\u0646 \u0628\u0648\u062f\u060c \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0628\u0631\u0627\u0632\u0634 \u062f\u0627\u0634\u062a\u0647 \u0628\u0627\u0634\u062f.  \u0627\u06cc\u0646 \u0645\u062f\u0644 \u0638\u0631\u0641\u06cc\u062a \u0622\u0646\u062a\u0631\u0648\u067e\u06cc\u06a9 \u0644\u0627\u0632\u0645 \u0631\u0627 \u0646\u062f\u0627\u0631\u062f \u062a\u0627 \u0628\u0647 \u062e\u0627\u0637\u0631 \u0627\u0633\u062a\u062f\u0644\u0627\u0644 \u0645\u0646 \u0648\u0627\u0642\u0639\u0627\u064b \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0645\u0646\u0627\u0633\u0628 \u0628\u0627\u0634\u062f.<\/p>\n<p>\u0627\u06cc\u0646 \u0645\u062f\u0644 \u0648 \u0645\u0639\u0645\u0627\u0631\u06cc \u0622\u0646 \u0628\u0647 \u0633\u0627\u062f\u06af\u06cc \u0628\u0631\u0627\u06cc \u0627\u06cc\u0646 \u06a9\u0627\u0631 \u0645\u0646\u0627\u0633\u0628 \u0646\u06cc\u0633\u062a &#8211; \u0648 \u062f\u0631 \u062d\u0627\u0644\u06cc \u06a9\u0647 \u0645\u0627 \u0645\u06cc \u062a\u0648\u0627\u0646\u06cc\u0645 \u0627\u0632 \u0646\u0638\u0631 \u0641\u0646\u06cc \u0622\u0646 \u0631\u0627 \u0628\u0647 \u062f\u0633\u062a \u0628\u06cc\u0627\u0648\u0631\u06cc\u0645 (\u0628\u06cc\u0634 \u0627\u0632 \u0622\u0646) \u0645\u062a\u0646\u0627\u0633\u0628 \u0628\u0627\u0634\u062f \u060c \u0627\u0645\u0627 \u0647\u0646\u0648\u0632 \u0647\u0645 \u062f\u0631 \u0637\u0648\u0644\u0627\u0646\u06cc \u0645\u062f\u062a \u0645\u0634\u06a9\u0644\u0627\u062a\u06cc \u062e\u0648\u0627\u0647\u062f \u062f\u0627\u0634\u062a.  \u0628\u0647 \u0639\u0646\u0648\u0627\u0646 \u0645\u062b\u0627\u0644\u060c \u0627\u062c\u0627\u0632\u0647 \u062f\u0647\u06cc\u062f \u0622\u0646 \u0631\u0627 \u0628\u0647 \u06cc\u06a9 \u0634\u0628\u06a9\u0647 \u0628\u0632\u0631\u06af\u062a\u0631 \u062a\u0628\u062f\u06cc\u0644 \u06a9\u0646\u06cc\u0645\u060c \u06a9\u0647 \u0627\u0632 \u0646\u0638\u0631 \u062a\u0626\u0648\u0631\u06cc \u0628\u0647 \u0622\u0646 \u0627\u062c\u0627\u0632\u0647 \u0645\u06cc \u062f\u0647\u062f \u0627\u0644\u06af\u0648\u0647\u0627\u06cc \u067e\u06cc\u0686\u06cc\u062f\u0647 \u062a\u0631\u06cc \u0631\u0627 \u062a\u0634\u062e\u06cc\u0635 \u062f\u0647\u062f:<\/p>\n<pre><code class=\"hljs\">model2 = keras.Sequential((\n  keras.layers.Flatten(input_shape=(<span class=\"hljs-number\">32<\/span>, <span class=\"hljs-number\">32<\/span>, <span class=\"hljs-number\">3<\/span>)),\n  keras.layers.BatchNormalization(),\n  keras.layers.Dense(<span class=\"hljs-number\">512<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, kernel_initializer=<span class=\"hljs-string\">\"he_normal\"<\/span>),\n    \n  keras.layers.Dense(<span class=\"hljs-number\">256<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>),\n  keras.layers.BatchNormalization(),\n  keras.layers.Dropout(<span class=\"hljs-number\">0.1<\/span>),\n    \n  keras.layers.Dense(<span class=\"hljs-number\">128<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>),\n  keras.layers.BatchNormalization(),\n  keras.layers.Dropout(<span class=\"hljs-number\">0.1<\/span>),\n\n  keras.layers.Dense(<span class=\"hljs-number\">100<\/span>, activation=<span class=\"hljs-string\">'softmax'<\/span>)\n))\n<\/code><\/pre>\n<p>\u0627\u06af\u0631\u0686\u0647\u060c \u0627\u06cc\u0646 \u0627\u0635\u0644\u0627\u064b \u0628\u0647\u062a\u0631 \u0646\u06cc\u0633\u062a:<\/p>\n<pre><code class=\"hljs\">Epoch <span class=\"hljs-number\">24<\/span>\/<span class=\"hljs-number\">150<\/span>\n<span class=\"hljs-number\">1407<\/span>\/<span class=\"hljs-number\">1407<\/span> (==============================) - 28s 20ms\/step - loss: <span class=\"hljs-number\">2.1202<\/span> - accuracy: <span class=\"hljs-number\">0.4507<\/span> - val_loss: <span class=\"hljs-number\">3.2796<\/span> - val_accuracy: <span class=\"hljs-number\">0.2528<\/span>\n<\/code><\/pre>\n<p>\u0627\u06cc\u0646 \u0628\u0633\u06cc\u0627\u0631 \u067e\u06cc\u0686\u06cc\u062f\u0647\u200c\u062a\u0631 \u0627\u0633\u062a (\u0686\u06af\u0627\u0644\u06cc \u0645\u0646\u0641\u062c\u0631 \u0645\u06cc\u200c\u0634\u0648\u062f)\u060c \u0627\u0645\u0627 \u0628\u0647 \u0633\u0627\u062f\u06af\u06cc \u0646\u0645\u06cc\u200c\u062a\u0648\u0627\u0646\u062f \u062e\u06cc\u0644\u06cc \u0628\u06cc\u0634\u062a\u0631 \u0627\u0633\u062a\u062e\u0631\u0627\u062c \u06a9\u0646\u062f:<\/p>\n<pre><code class=\"hljs\">model1.summary()\nmodel2.summary()\n<\/code><\/pre>\n<pre><code class=\"hljs\">Model: <span class=\"hljs-string\">\"sequential_17\"<\/span>\n...\nTotal params: <span class=\"hljs-number\">845<\/span>,<span class=\"hljs-number\">284<\/span>\nTrainable params: <span class=\"hljs-number\">838<\/span>,<span class=\"hljs-number\">884<\/span>\nNon-trainable params: <span class=\"hljs-number\">6<\/span>,<span class=\"hljs-number\">400<\/span>\n_________________________________________________________________\nModel: <span class=\"hljs-string\">\"sequential_18\"<\/span>\n...\nTotal params: <span class=\"hljs-number\">1<\/span>,<span class=\"hljs-number\">764<\/span>,<span class=\"hljs-number\">324<\/span>\nTrainable params: <span class=\"hljs-number\">1<\/span>,<span class=\"hljs-number\">757<\/span>,<span class=\"hljs-number\">412<\/span>\nNon-trainable params: <span class=\"hljs-number\">6<\/span>,<span class=\"hljs-number\">912<\/span>\n<\/code><\/pre>\n<h3 id=\"overfittingconvolutionalneuralnetworkoncifar10\"><span class=\"ez-toc-section\" id=\"%d8%a8%db%8c%d8%b4_%d8%a7%d8%b2_%d8%ad%d8%af_%d8%a8%d8%b1%d8%a7%d8%b2%d8%b4_%d8%b4%d8%a8%da%a9%d9%87_%d8%b9%d8%b5%d8%a8%db%8c_%da%a9%d8%a7%d9%86%d9%88%d9%84%d9%88%d8%b4%d9%86%d8%a7%d9%84_%d8%b1%d9%88%db%8c_cifar10\"><\/span>\u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0628\u0631\u0627\u0632\u0634 \u0634\u0628\u06a9\u0647 \u0639\u0635\u0628\u06cc \u06a9\u0627\u0646\u0648\u0644\u0648\u0634\u0646\u0627\u0644 \u0631\u0648\u06cc CIFAR10<span class=\"ez-toc-section-end\"><\/span><\/h3>\n<p>\u062d\u0627\u0644\u0627 \u0628\u06cc\u0627\u06cc\u06cc\u062f \u0633\u0639\u06cc \u06a9\u0646\u06cc\u0645 \u06a9\u0627\u0631\u06cc \u0645\u062a\u0641\u0627\u0648\u062a \u0627\u0646\u062c\u0627\u0645 \u062f\u0647\u06cc\u0645.  \u062a\u063a\u06cc\u06cc\u0631 \u0628\u0647 \u06cc\u06a9 CNN \u0628\u0647 \u0637\u0648\u0631 \u0642\u0627\u0628\u0644 \u062a\u0648\u062c\u0647\u06cc \u0628\u0647 \u0627\u0633\u062a\u062e\u0631\u0627\u062c \u0648\u06cc\u0698\u06af\u06cc\u200c\u0647\u0627 \u0627\u0632 \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u06a9\u0645\u06a9 \u0645\u06cc\u200c\u06a9\u0646\u062f \u0648 \u062f\u0631 \u0646\u062a\u06cc\u062c\u0647 \u0628\u0647 \u0645\u062f\u0644 \u0627\u062c\u0627\u0632\u0647 \u0645\u06cc\u200c\u062f\u0647\u062f <em>\u0628\u0631\u0627\u0633\u062a\u06cc<\/em> overfit\u060c \u0628\u0647 \u062f\u0642\u062a \u0628\u0633\u06cc\u0627\u0631 \u0628\u0627\u0644\u0627\u062a\u0631 (\u062a\u0648\u0647\u0645) \u0645\u06cc \u0631\u0633\u062f.<\/p>\n<p>\u0645\u0627 \u0622\u0646 \u0631\u0627 \u0628\u06cc\u0631\u0648\u0646 \u0645\u06cc \u0627\u0646\u062f\u0627\u0632\u06cc\u0645 <code>EarlyStopping<\/code> \u062a\u0645\u0627\u0633 \u0628\u06af\u06cc\u0631\u06cc\u062f \u062a\u0627 \u0627\u062c\u0627\u0632\u0647 \u062f\u0647\u06cc\u062f \u06a9\u0627\u0631 \u062e\u0648\u062f \u0631\u0627 \u0627\u0646\u062c\u0627\u0645 \u062f\u0647\u062f.  \u0639\u0644\u0627\u0648\u0647 \u0628\u0631 \u0627\u06cc\u0646\u060c \u0645\u0627 \u0627\u0633\u062a\u0641\u0627\u062f\u0647 \u0646\u062e\u0648\u0627\u0647\u06cc\u0645 \u06a9\u0631\u062f <code>Dropout<\/code> \u0644\u0627\u06cc\u0647 \u0647\u0627\u060c \u0648 \u062f\u0631 \u0639\u0648\u0636 \u0633\u0639\u06cc \u06a9\u0646\u06cc\u062f \u0634\u0628\u06a9\u0647 \u0631\u0627 \u0645\u062c\u0628\u0648\u0631 \u06a9\u0646\u06cc\u062f \u062a\u0627 \u0648\u06cc\u0698\u06af\u06cc \u0647\u0627 \u0631\u0627 \u0627\u0632 \u0637\u0631\u06cc\u0642 \u0644\u0627\u06cc\u0647 \u0647\u0627\u06cc \u0628\u06cc\u0634\u062a\u0631\u06cc \u06cc\u0627\u062f \u0628\u06af\u06cc\u0631\u062f.<\/p>\n<div class=\"alert alert-note\">\n<div class=\"flex\">\n<div class=\"flex-shrink-0 mr-3\"><\/div>\n<div class=\"w-full\">\n<p><strong>\u062a\u0648\u062c\u0647 \u062f\u0627\u0634\u062a\u0647 \u0628\u0627\u0634\u06cc\u062f:<\/strong> \u062e\u0627\u0631\u062c \u0627\u0632 \u0632\u0645\u06cc\u0646\u0647 \u062a\u0644\u0627\u0634 \u0628\u0631\u0627\u06cc \u0627\u062b\u0628\u0627\u062a \u0627\u0633\u062a\u062f\u0644\u0627\u0644\u060c \u0627\u06cc\u0646 \u062a\u0648\u0635\u06cc\u0647 \u0648\u062d\u0634\u062a\u0646\u0627\u06a9\u06cc \u062e\u0648\u0627\u0647\u062f \u0628\u0648\u062f.  \u0627\u06cc\u0646 \u0628\u0631\u0639\u06a9\u0633 \u06a9\u0627\u0631\u06cc \u0627\u0633\u062a \u06a9\u0647 \u0645\u06cc \u062e\u0648\u0627\u0647\u06cc\u062f \u062f\u0631 \u067e\u0627\u06cc\u0627\u0646 \u0627\u0646\u062c\u0627\u0645 \u062f\u0647\u06cc\u062f.  Dropout \u0628\u0647 \u0634\u0628\u06a9\u0647\u200c\u0647\u0627 \u06a9\u0645\u06a9 \u0645\u06cc\u200c\u06a9\u0646\u062f \u062a\u0627 \u0628\u0627 \u0648\u0627\u062f\u0627\u0631 \u06a9\u0631\u062f\u0646 \u0646\u0648\u0631\u0648\u0646\u200c\u0647\u0627\u06cc\u06cc \u06a9\u0647 \u0631\u0647\u0627 \u0646\u0634\u062f\u0647\u200c\u0627\u0646\u062f\u060c \u0633\u0633\u062a\u06cc \u0631\u0627 \u062f\u0631\u06cc\u0627\u0641\u062a \u06a9\u0646\u0646\u062f\u060c \u0628\u0647\u062a\u0631 \u062a\u0639\u0645\u06cc\u0645 \u067e\u06cc\u062f\u0627 \u06a9\u0646\u0646\u062f.  \u0627\u062c\u0628\u0627\u0631 \u0634\u0628\u06a9\u0647 \u0628\u0647 \u06cc\u0627\u062f\u06af\u06cc\u0631\u06cc \u0627\u0632 \u0637\u0631\u06cc\u0642 \u0644\u0627\u06cc\u0647 \u0647\u0627\u06cc \u0628\u06cc\u0634\u062a\u0631 \u0628\u0647 \u0627\u062d\u062a\u0645\u0627\u0644 \u0632\u06cc\u0627\u062f \u0645\u0646\u062c\u0631 \u0628\u0647 \u06cc\u06a9 \u0645\u062f\u0644 \u0627\u0636\u0627\u0641\u0647 \u0628\u0631\u0627\u0632\u0634 \u0645\u06cc \u0634\u0648\u062f.<\/p>\n<\/p><\/div><\/div><\/div>\n<p>\u062f\u0644\u06cc\u0644 \u0627\u06cc\u0646\u06a9\u0647 \u0645\u0646 \u0628\u0647 \u0637\u0648\u0631 \u0647\u062f\u0641\u0645\u0646\u062f \u0627\u06cc\u0646 \u06a9\u0627\u0631 \u0631\u0627 \u0627\u0646\u062c\u0627\u0645 \u0645\u06cc \u062f\u0647\u0645 \u0627\u06cc\u0646 \u0627\u0633\u062a \u06a9\u0647 \u0628\u0647 \u0634\u0628\u06a9\u0647 \u0627\u062c\u0627\u0632\u0647 \u0645\u06cc \u062f\u0647\u0645 \u0628\u0647 \u0637\u0648\u0631 \u0648\u062d\u0634\u062a\u0646\u0627\u06a9\u06cc \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0645\u0646\u0627\u0633\u0628 \u0634\u0648\u062f <strong>\u0646\u0634\u0627\u0646\u0647 \u0627\u06cc \u0627\u0632 \u062a\u0648\u0627\u0646\u0627\u06cc\u06cc \u0622\u0646 \u062f\u0631 \u062a\u0634\u062e\u06cc\u0635 \u0648\u0627\u0642\u0639\u06cc \u0648\u06cc\u0698\u06af\u06cc \u0647\u0627\u060c \u0642\u0628\u0644 \u0627\u0632 \u0633\u0627\u062f\u0647 \u0633\u0627\u0632\u06cc \u0648 \u0627\u0636\u0627\u0641\u0647 \u06a9\u0631\u062f\u0646 \u0622\u0646 <code>Dropout<\/code> \u062a\u0627 \u0648\u0627\u0642\u0639\u0627\u064b \u0628\u0647 \u0622\u0646 \u0627\u062c\u0627\u0632\u0647 \u062a\u0639\u0645\u06cc\u0645 \u062f\u0627\u062f\u0647 \u0634\u0648\u062f.<\/strong> \u0627\u06af\u0631 \u0628\u0647 \u062f\u0642\u062a (\u062a\u0648\u0647\u0645) \u0628\u0627\u0644\u0627\u06cc\u06cc \u0628\u0631\u0633\u062f\u060c \u0645\u06cc \u062a\u0648\u0627\u0646\u062f \u0628\u0633\u06cc\u0627\u0631 \u0628\u06cc\u0634\u062a\u0631 \u0627\u0632 \u0645\u062f\u0644 MLP \u0627\u0633\u062a\u062e\u0631\u0627\u062c \u06a9\u0646\u062f\u060c \u0628\u0647 \u0627\u06cc\u0646 \u0645\u0639\u0646\u06cc \u06a9\u0647 \u0645\u06cc \u062a\u0648\u0627\u0646\u06cc\u0645 \u0633\u0627\u062f\u0647 \u0633\u0627\u0632\u06cc \u0622\u0646 \u0631\u0627 \u0622\u063a\u0627\u0632 \u06a9\u0646\u06cc\u0645.<\/p>\n<p>\u0627\u062c\u0627\u0632\u0647 \u062f\u0647\u06cc\u062f \u06cc\u06a9 \u0628\u0627\u0631 \u062f\u06cc\u06af\u0631 \u0627\u0632 API \u0645\u062a\u0648\u0627\u0644\u06cc \u0628\u0631\u0627\u06cc \u0633\u0627\u062e\u062a \u06cc\u06a9 CNN \u0627\u0633\u062a\u0641\u0627\u062f\u0647 \u06a9\u0646\u06cc\u0645 \u0631\u0648\u06cc \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 CIFAR10:<\/p>\n<pre><code class=\"hljs\">checkpoint = keras.callbacks.ModelCheckpoint(<span class=\"hljs-string\">\"overcomplicated_cnn_cifar10.h5\"<\/span>, save_best_only=<span class=\"hljs-literal\">True<\/span>)\n\nmodel = keras.models.Sequential((\n    keras.layers.Conv2D(<span class=\"hljs-number\">64<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, \n                        kernel_initializer=<span class=\"hljs-string\">\"he_normal\"<\/span>, \n                        kernel_regularizer=keras.regularizers.l2(l=<span class=\"hljs-number\">0.01<\/span>), \n                        padding=<span class=\"hljs-string\">'same'<\/span>, \n                        input_shape=(<span class=\"hljs-number\">32<\/span>, <span class=\"hljs-number\">32<\/span>, <span class=\"hljs-number\">3<\/span>)),\n    keras.layers.Conv2D(<span class=\"hljs-number\">64<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.MaxPooling2D(<span class=\"hljs-number\">2<\/span>),\n    \n    keras.layers.Conv2D(<span class=\"hljs-number\">128<\/span>, <span class=\"hljs-number\">2<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.Conv2D(<span class=\"hljs-number\">128<\/span>, <span class=\"hljs-number\">2<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.MaxPooling2D(<span class=\"hljs-number\">2<\/span>),\n    \n    keras.layers.Conv2D(<span class=\"hljs-number\">256<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.Conv2D(<span class=\"hljs-number\">256<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.MaxPooling2D(<span class=\"hljs-number\">2<\/span>),\n    \n    keras.layers.Conv2D(<span class=\"hljs-number\">128<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.Conv2D(<span class=\"hljs-number\">128<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.MaxPooling2D(<span class=\"hljs-number\">2<\/span>),\n    \n    keras.layers.Conv2D(<span class=\"hljs-number\">64<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.Conv2D(<span class=\"hljs-number\">64<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.MaxPooling2D(<span class=\"hljs-number\">2<\/span>),\n    \n    keras.layers.Flatten(),    \n    keras.layers.Dense(<span class=\"hljs-number\">32<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>),\n    keras.layers.Dense(<span class=\"hljs-number\">10<\/span>, activation=<span class=\"hljs-string\">'softmax'<\/span>)\n))\n\nmodel.<span class=\"hljs-built_in\">compile<\/span>(loss=<span class=\"hljs-string\">\"sparse_categorical_crossentropy\"<\/span>,\n              optimizer=keras.optimizers.Adam(learning_rate=<span class=\"hljs-number\">1e-3<\/span>),\n              metrics=(<span class=\"hljs-string\">\"accuracy\"<\/span>))\n\nmodel.summary()\n\nhistory = model.fit(X_train, \n                    Y_train, \n                    epochs=<span class=\"hljs-number\">150<\/span>,\n                    batch_size=<span class=\"hljs-number\">64<\/span>,\n                    validation_data=(X_valid, Y_valid),\n                    callbacks=(checkpoint))\n<\/code><\/pre>\n<p>\u0639\u0627\u0644\u06cc\u060c \u062e\u06cc\u0644\u06cc \u0633\u0631\u06cc\u0639 \u0627\u0636\u0627\u0641\u0647 \u0645\u06cc \u0634\u0648\u062f!  \u062a\u0646\u0647\u0627 \u062f\u0631 \u0686\u0646\u062f \u062f\u0648\u0631\u0647\u060c \u0634\u0631\u0648\u0639 \u0628\u0647 \u062a\u0637\u0628\u06cc\u0642 \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u062f\u0627\u062f\u0647 \u0647\u0627 \u06a9\u0631\u062f \u0648 \u062f\u0631 \u062f\u0648\u0631\u0647 31\u060c \u0628\u0627 \u062f\u0642\u062a \u0627\u0639\u062a\u0628\u0627\u0631 \u0633\u0646\u062c\u06cc \u06a9\u0645\u062a\u0631\u060c \u0628\u0647 98 \u062f\u0631\u0635\u062f \u0631\u0633\u06cc\u062f:<\/p>\n<pre><code class=\"hljs\">Epoch <span class=\"hljs-number\">1<\/span>\/<span class=\"hljs-number\">150<\/span>\n<span class=\"hljs-number\">704<\/span>\/<span class=\"hljs-number\">704<\/span> (==============================) - 149s 210ms\/step - loss: <span class=\"hljs-number\">1.9561<\/span> - accuracy: <span class=\"hljs-number\">0.4683<\/span> - val_loss: <span class=\"hljs-number\">2.5060<\/span> - val_accuracy: <span class=\"hljs-number\">0.3760<\/span>\n...\nEpoch <span class=\"hljs-number\">31<\/span>\/<span class=\"hljs-number\">150<\/span>\n<span class=\"hljs-number\">704<\/span>\/<span class=\"hljs-number\">704<\/span> (==============================) - 149s 211ms\/step - loss: <span class=\"hljs-number\">0.0610<\/span> - accuracy: <span class=\"hljs-number\">0.9841<\/span> - val_loss: <span class=\"hljs-number\">1.0433<\/span> - val_accuracy: <span class=\"hljs-number\">0.6958<\/span>\n<\/code><\/pre>\n<p>\u0627\u0632 \u0622\u0646\u062c\u0627\u06cc\u06cc \u06a9\u0647 \u062a\u0646\u0647\u0627 10 \u06a9\u0644\u0627\u0633 \u062e\u0631\u0648\u062c\u06cc \u0648\u062c\u0648\u062f \u062f\u0627\u0631\u062f\u060c \u062d\u062a\u06cc \u0627\u06af\u0631 \u0645\u0627 \u0633\u0639\u06cc \u06a9\u0631\u062f\u06cc\u0645 \u0622\u0646 \u0631\u0627 \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0646\u0635\u0628 \u06a9\u0646\u06cc\u0645 <em>\u0632\u06cc\u0627\u062f<\/em> \u0628\u0627 \u0627\u06cc\u062c\u0627\u062f \u06cc\u06a9 CNN \u0628\u0632\u0631\u06af \u063a\u06cc\u0631 \u0636\u0631\u0648\u0631\u06cc\u060c \u0635\u062d\u062a \u0627\u0639\u062a\u0628\u0627\u0631\u0633\u0646\u062c\u06cc \u0647\u0646\u0648\u0632 \u0646\u0633\u0628\u062a\u0627\u064b \u0628\u0627\u0644\u0627\u0633\u062a.<\/p>\n<h3 id=\"simplifyingtheconvolutionalneuralnetworkoncifar10\"><span class=\"ez-toc-section\" id=\"%d8%b3%d8%a7%d8%af%d9%87_%d8%b3%d8%a7%d8%b2%db%8c_%d8%b4%d8%a8%da%a9%d9%87_%d8%b9%d8%b5%d8%a8%db%8c_%da%a9%d8%a7%d9%86%d9%88%d9%84%d9%88%d8%b4%d9%86%d8%a7%d9%84_%d8%b1%d9%88%db%8c_cifar10\"><\/span>\u0633\u0627\u062f\u0647 \u0633\u0627\u0632\u06cc \u0634\u0628\u06a9\u0647 \u0639\u0635\u0628\u06cc \u06a9\u0627\u0646\u0648\u0644\u0648\u0634\u0646\u0627\u0644 \u0631\u0648\u06cc CIFAR10<span class=\"ez-toc-section-end\"><\/span><\/h3>\n<p>\u062d\u0627\u0644\u0627 \u0628\u06cc\u0627\u06cc\u06cc\u062f \u0622\u0646 \u0631\u0627 \u0633\u0627\u062f\u0647 \u06a9\u0646\u06cc\u0645 \u062a\u0627 \u0628\u0628\u06cc\u0646\u06cc\u0645 \u0628\u0627 \u06cc\u06a9 \u0645\u0639\u0645\u0627\u0631\u06cc \u0645\u0639\u0642\u0648\u0644 \u062a\u0631 \u0686\u06af\u0648\u0646\u0647 \u0639\u0645\u0644 \u0645\u06cc \u06a9\u0646\u062f.  \u0627\u0636\u0627\u0641\u0647 \u0645\u06cc \u06a9\u0646\u06cc\u0645 <code>BatchNormalization<\/code> \u0648 <code>Dropout<\/code> \u0632\u06cc\u0631\u0627 \u0647\u0631 \u062f\u0648 \u0628\u0647 \u062a\u0639\u0645\u06cc\u0645 \u06a9\u0645\u06a9 \u0645\u06cc \u06a9\u0646\u0646\u062f:<\/p>\n<pre><code class=\"hljs\">checkpoint = keras.callbacks.ModelCheckpoint(<span class=\"hljs-string\">\"simplified_cnn_cifar10.h5\"<\/span>, save_best_only=<span class=\"hljs-literal\">True<\/span>)\nearly_stopping = keras.callbacks.EarlyStopping(patience=<span class=\"hljs-number\">10<\/span>, restore_best_weights=<span class=\"hljs-literal\">True<\/span>)\n\nmodel = keras.models.Sequential((\n    keras.layers.Conv2D(<span class=\"hljs-number\">32<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, kernel_initializer=<span class=\"hljs-string\">\"he_normal\"<\/span>, kernel_regularizer=keras.regularizers.l2(l=<span class=\"hljs-number\">0.01<\/span>), padding=<span class=\"hljs-string\">'same'<\/span>, input_shape=(<span class=\"hljs-number\">32<\/span>, <span class=\"hljs-number\">32<\/span>, <span class=\"hljs-number\">3<\/span>)),\n    keras.layers.BatchNormalization(),\n    keras.layers.Conv2D(<span class=\"hljs-number\">32<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.MaxPooling2D(<span class=\"hljs-number\">2<\/span>),\n    keras.layers.Dropout(<span class=\"hljs-number\">0.4<\/span>),\n    \n    keras.layers.Conv2D(<span class=\"hljs-number\">64<\/span>, <span class=\"hljs-number\">2<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.Conv2D(<span class=\"hljs-number\">64<\/span>, <span class=\"hljs-number\">2<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.MaxPooling2D(<span class=\"hljs-number\">2<\/span>),\n    keras.layers.Dropout(<span class=\"hljs-number\">0.4<\/span>),\n    \n    keras.layers.Conv2D(<span class=\"hljs-number\">128<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.Conv2D(<span class=\"hljs-number\">128<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.MaxPooling2D(<span class=\"hljs-number\">2<\/span>),\n    keras.layers.Dropout(<span class=\"hljs-number\">0.5<\/span>),\n    \n    keras.layers.Flatten(),    \n    keras.layers.Dense(<span class=\"hljs-number\">32<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.Dropout(<span class=\"hljs-number\">0.3<\/span>),\n    keras.layers.Dense(<span class=\"hljs-number\">10<\/span>, activation=<span class=\"hljs-string\">'softmax'<\/span>)\n))\n\nmodel.<span class=\"hljs-built_in\">compile<\/span>(loss=<span class=\"hljs-string\">\"sparse_categorical_crossentropy\"<\/span>,\n              optimizer=keras.optimizers.Adam(learning_rate=<span class=\"hljs-number\">1e-3<\/span>),\n              metrics=(<span class=\"hljs-string\">\"accuracy\"<\/span>))\n\nmodel.summary()\n\nhistory = model.fit(X_train, \n                    Y_train, \n                    epochs=<span class=\"hljs-number\">150<\/span>,\n                    batch_size=<span class=\"hljs-number\">64<\/span>,\n                    validation_data=(X_valid, Y_valid),\n                    callbacks=(checkpoint, early_stopping))\n<\/code><\/pre>\n<p>\u0627\u06cc\u0646 \u0645\u062f\u0644 \u062f\u0627\u0631\u0627\u06cc \u062a\u0639\u062f\u0627\u062f (\u0645\u062a\u0648\u0633\u0637) 323146 \u067e\u0627\u0631\u0627\u0645\u062a\u0631 \u0642\u0627\u0628\u0644 \u0622\u0645\u0648\u0632\u0634 \u0627\u0633\u062a\u060c \u062f\u0631 \u0645\u0642\u0627\u06cc\u0633\u0647 \u0628\u0627 1\u060c579\u060c178 \u0627\u0632 CNN \u0642\u0628\u0644\u06cc.  \u0639\u0645\u0644\u06a9\u0631\u062f \u0622\u0646 \u0686\u06af\u0648\u0646\u0647 \u0627\u0633\u062a\u061f<\/p>\n<pre><code class=\"hljs\">Epoch <span class=\"hljs-number\">1<\/span>\/<span class=\"hljs-number\">150<\/span>\n<span class=\"hljs-number\">704<\/span>\/<span class=\"hljs-number\">704<\/span> (==============================) - 91s 127ms\/step - loss: <span class=\"hljs-number\">2.1327<\/span> - accuracy: <span class=\"hljs-number\">0.3910<\/span> - val_loss: <span class=\"hljs-number\">1.5495<\/span> - val_accuracy: <span class=\"hljs-number\">0.5406<\/span>\n...\nEpoch <span class=\"hljs-number\">52<\/span>\/<span class=\"hljs-number\">150<\/span>\n<span class=\"hljs-number\">704<\/span>\/<span class=\"hljs-number\">704<\/span> (==============================) - 89s 127ms\/step - loss: <span class=\"hljs-number\">0.4091<\/span> - accuracy: <span class=\"hljs-number\">0.8648<\/span> - val_loss: <span class=\"hljs-number\">0.4694<\/span> - val_accuracy: <span class=\"hljs-number\">0.8500<\/span>\n<\/code><\/pre>\n<p>\u062f\u0631 \u0648\u0627\u0642\u0639 \u0628\u0647 \u062f\u0642\u062a 85% \u0628\u0633\u06cc\u0627\u0631 \u0645\u0646\u0627\u0633\u0628\u06cc \u0645\u06cc \u0631\u0633\u062f!  \u062a\u06cc\u063a \u0627\u0648\u06a9\u0627\u0645 \u062f\u0648\u0628\u0627\u0631\u0647 \u0645\u06cc \u0632\u0646\u062f.  \u0628\u06cc\u0627\u06cc\u06cc\u062f \u0646\u06af\u0627\u0647\u06cc \u0628\u0647 \u0628\u0631\u062e\u06cc \u0627\u0632 \u0646\u062a\u0627\u06cc\u062c \u0628\u06cc\u0627\u0646\u062f\u0627\u0632\u06cc\u0645:<\/p>\n<pre><code class=\"hljs\">y_preds = model.predict(X_test)\n<span class=\"hljs-built_in\">print<\/span>(y_preds(<span class=\"hljs-number\">1<\/span>))\n<span class=\"hljs-built_in\">print<\/span>(np.argmax(y_preds(<span class=\"hljs-number\">1<\/span>)))\n\nfig, ax = plt.subplots(<span class=\"hljs-number\">6<\/span>, <span class=\"hljs-number\">6<\/span>, figsize=(<span class=\"hljs-number\">10<\/span>, <span class=\"hljs-number\">10<\/span>))\nax = ax.ravel()\n\n<span class=\"hljs-keyword\">for<\/span> i <span class=\"hljs-keyword\">in<\/span> <span class=\"hljs-built_in\">range<\/span>(<span class=\"hljs-number\">0<\/span>, <span class=\"hljs-number\">36<\/span>):\n    ax(i).imshow(X_test(i))\n    ax(i).set_title(<span class=\"hljs-string\">\"Actual: %s\\nPred: %s\"<\/span> % (class_names(Y_test(i)(<span class=\"hljs-number\">0<\/span>)), class_names(np.argmax(y_preds(i)))))\n    ax(i).axis(<span class=\"hljs-string\">'off'<\/span>)\n    plt.subplots_adjust(wspace=<span class=\"hljs-number\">1<\/span>)\n    \nplt.show()\n<\/code><\/pre>\n<p><img decoding=\"async\" src=\"https:\/\/rasanegar.com\/blog\/wp-content\/uploads\/2024\/01\/overfitting-is-your-friend-not-your-foe-8.png\" alt=\"\" title=\"\"><\/p>\n<p>\u0637\u0628\u0642\u0647 \u0628\u0646\u062f\u06cc \u0646\u0627\u062f\u0631\u0633\u062a \u0627\u0635\u0644\u06cc \u062f\u0648 \u062a\u0635\u0648\u06cc\u0631 \u062f\u0631 \u0627\u06cc\u0646 \u0645\u062c\u0645\u0648\u0639\u0647 \u06a9\u0648\u0686\u06a9 \u0627\u0633\u062a &#8211; \u06cc\u06a9 \u0633\u06af \u0628\u0647 \u0639\u0646\u0648\u0627\u0646 \u06af\u0648\u0632\u0646 \u0637\u0628\u0642\u0647 \u0628\u0646\u062f\u06cc \u0634\u062f (\u0628\u0647 \u0627\u0646\u062f\u0627\u0632\u0647 \u06a9\u0627\u0641\u06cc \u0642\u0627\u0628\u0644 \u0627\u062d\u062a\u0631\u0627\u0645) \u060c \u0627\u0645\u0627 \u06cc\u06a9 \u0646\u0632\u062f\u06cc\u06a9\u06cc \u0627\u0632 \u06cc\u06a9 \u067e\u0631\u0646\u062f\u0647 EMU \u0628\u0647 \u0639\u0646\u0648\u0627\u0646 \u06cc\u06a9 \u06af\u0631\u0628\u0647 \u0637\u0628\u0642\u0647 \u0628\u0646\u062f\u06cc \u0634\u062f (\u0628\u0647 \u0627\u0646\u062f\u0627\u0632\u0647 \u06a9\u0627\u0641\u06cc \u062e\u0646\u062f\u0647 \u062f\u0627\u0631 \u060c \u0628\u0646\u0627\u0628\u0631\u0627\u06cc\u0646 \u0645\u0627 \u0627\u062c\u0627\u0632\u0647 \u062e\u0648\u0627\u0647\u06cc\u0645 \u062f\u0627\u062f \u06a9\u0647 \u0622\u0646 \u0631\u0627 \u0628\u06a9\u0634\u06cc\u062f).<\/p>\n<h3 id=\"overfittingconvolutionalneuralnetworkoncifar100\"><span class=\"ez-toc-section\" id=\"%d8%a8%db%8c%d8%b4_%d8%a7%d8%b2_%d8%ad%d8%af_%d8%a8%d8%b1%d8%a7%d8%b2%d8%b4_%d8%b4%d8%a8%da%a9%d9%87_%d8%b9%d8%b5%d8%a8%db%8c_%da%a9%d8%a7%d9%86%d9%88%d9%84%d9%88%d8%b4%d9%86_%d8%b1%d9%88%db%8c_cifar100\"><\/span>\u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0628\u0631\u0627\u0632\u0634 \u0634\u0628\u06a9\u0647 \u0639\u0635\u0628\u06cc \u06a9\u0627\u0646\u0648\u0644\u0648\u0634\u0646 \u0631\u0648\u06cc CIFAR100<span class=\"ez-toc-section-end\"><\/span><\/h3>\n<p>\u0648\u0642\u062a\u06cc \u0628\u0647 \u0633\u0631\u0627\u063a \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 CIFAR100 \u0645\u06cc \u0631\u0648\u06cc\u0645 \u0686\u0647 \u0627\u062a\u0641\u0627\u0642\u06cc \u0645\u06cc \u0627\u0641\u062a\u062f\u061f<\/p>\n<pre><code class=\"hljs\">checkpoint = keras.callbacks.ModelCheckpoint(<span class=\"hljs-string\">\"overcomplicated_cnn_model_cifar100.h5\"<\/span>, save_best_only=<span class=\"hljs-literal\">True<\/span>)\nearly_stopping = keras.callbacks.EarlyStopping(patience=<span class=\"hljs-number\">10<\/span>, restore_best_weights=<span class=\"hljs-literal\">True<\/span>)\n\nmodel = keras.models.Sequential((\n    keras.layers.Conv2D(<span class=\"hljs-number\">32<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, kernel_initializer=<span class=\"hljs-string\">\"he_normal\"<\/span>, kernel_regularizer=keras.regularizers.l2(l=<span class=\"hljs-number\">0.01<\/span>), padding=<span class=\"hljs-string\">'same'<\/span>, input_shape=(<span class=\"hljs-number\">32<\/span>, <span class=\"hljs-number\">32<\/span>, <span class=\"hljs-number\">3<\/span>)),\n    keras.layers.BatchNormalization(),\n    keras.layers.Conv2D(<span class=\"hljs-number\">32<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.MaxPooling2D(<span class=\"hljs-number\">2<\/span>),\n    \n    keras.layers.Conv2D(<span class=\"hljs-number\">64<\/span>, <span class=\"hljs-number\">2<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.Conv2D(<span class=\"hljs-number\">64<\/span>, <span class=\"hljs-number\">2<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.MaxPooling2D(<span class=\"hljs-number\">2<\/span>),\n    \n    keras.layers.Conv2D(<span class=\"hljs-number\">128<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.Conv2D(<span class=\"hljs-number\">128<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.MaxPooling2D(<span class=\"hljs-number\">2<\/span>),\n    \n    keras.layers.Conv2D(<span class=\"hljs-number\">128<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.Conv2D(<span class=\"hljs-number\">128<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.MaxPooling2D(<span class=\"hljs-number\">2<\/span>),\n    \n    keras.layers.Conv2D(<span class=\"hljs-number\">64<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.Conv2D(<span class=\"hljs-number\">64<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.MaxPooling2D(<span class=\"hljs-number\">2<\/span>),\n    \n    keras.layers.Flatten(),    \n    keras.layers.Dense(<span class=\"hljs-number\">256<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.Dense(<span class=\"hljs-number\">128<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>),\n    keras.layers.BatchNormalization(),\n    \n    keras.layers.Dense(<span class=\"hljs-number\">100<\/span>, activation=<span class=\"hljs-string\">'softmax'<\/span>)\n))\n\nmodel.<span class=\"hljs-built_in\">compile<\/span>(loss=<span class=\"hljs-string\">\"sparse_categorical_crossentropy\"<\/span>,\n              optimizer=keras.optimizers.Adam(learning_rate=<span class=\"hljs-number\">1e-3<\/span>),\n              metrics=(<span class=\"hljs-string\">\"accuracy\"<\/span>))\n\nmodel.summary()\n\nhistory = model.fit(X_train, \n                    Y_train, \n                    epochs=<span class=\"hljs-number\">150<\/span>,\n                    batch_size=<span class=\"hljs-number\">64<\/span>,\n                    validation_data=(X_valid, Y_valid),\n                    callbacks=(checkpoint))\n<\/code><\/pre>\n<pre><code class=\"hljs\">Epoch <span class=\"hljs-number\">1<\/span>\/<span class=\"hljs-number\">150<\/span>\n<span class=\"hljs-number\">704<\/span>\/<span class=\"hljs-number\">704<\/span> (==============================) - 97s 137ms\/step - loss: <span class=\"hljs-number\">4.1752<\/span> - accuracy: <span class=\"hljs-number\">0.1336<\/span> - val_loss: <span class=\"hljs-number\">3.9696<\/span> - val_accuracy: <span class=\"hljs-number\">0.1392<\/span>\n...\nEpoch <span class=\"hljs-number\">42<\/span>\/<span class=\"hljs-number\">150<\/span>\n<span class=\"hljs-number\">704<\/span>\/<span class=\"hljs-number\">704<\/span> (==============================) - 95s 135ms\/step - loss: <span class=\"hljs-number\">0.1543<\/span> - accuracy: <span class=\"hljs-number\">0.9572<\/span> - val_loss: <span class=\"hljs-number\">4.1394<\/span> - val_accuracy: <span class=\"hljs-number\">0.4458<\/span>\n<\/code><\/pre>\n<p>\u0641\u0648\u0642 \u0627\u0644\u0639\u0627\u062f\u0647!  \u062f\u0642\u062a 96% \u0631\u0648\u06cc \u0645\u062c\u0645\u0648\u0639\u0647 \u0622\u0645\u0648\u0632\u0634\u06cc!  \u0647\u0646\u0648\u0632 \u062f\u0642\u062a \u0627\u0639\u062a\u0628\u0627\u0631 44% \u0631\u0627 \u0646\u06af\u0631\u0627\u0646 \u0646\u06a9\u0646\u06cc\u062f.  \u0628\u06cc\u0627\u06cc\u06cc\u062f \u0645\u062f\u0644 \u0631\u0627 \u062e\u06cc\u0644\u06cc \u0633\u0631\u06cc\u0639 \u0633\u0627\u062f\u0647 \u06a9\u0646\u06cc\u0645 \u062a\u0627 \u0622\u0646 \u0631\u0627 \u0628\u0647 \u062a\u0639\u0645\u06cc\u0645 \u0628\u0647\u062a\u0631\u06cc \u0628\u0631\u0633\u0627\u0646\u06cc\u0645.<\/p>\n<h3 id=\"failuretogeneralizeaftersimplification\"><span class=\"ez-toc-section\" id=\"%d8%b9%d8%af%d9%85_%d8%aa%d8%b9%d9%85%db%8c%d9%85_%d8%a8%d8%b9%d8%af_%d8%a7%d8%b2_%d8%b3%d8%a7%d8%af%d9%87_%d8%b3%d8%a7%d8%b2%db%8c\"><\/span>\u0639\u062f\u0645 \u062a\u0639\u0645\u06cc\u0645 \u0628\u0639\u062f \u0627\u0632 \u0633\u0627\u062f\u0647 \u0633\u0627\u0632\u06cc<span class=\"ez-toc-section-end\"><\/span><\/h3>\n<p>\u0648 \u0627\u06cc\u0646\u062c\u0627\u0633\u062a \u06a9\u0647 \u0645\u0634\u062e\u0635 \u0645\u06cc \u0634\u0648\u062f \u06a9\u0647 \u062a\u0648\u0627\u0646\u0627\u06cc\u06cc \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0645\u0646\u0627\u0633\u0628 \u0646\u06cc\u0633\u062a <strong>\u0636\u0645\u0627\u0646\u062a<\/strong> \u06a9\u0647 \u0627\u06cc\u0646 \u0645\u062f\u0644 \u062f\u0631 \u0635\u0648\u0631\u062a \u0633\u0627\u062f\u0647 \u0633\u0627\u0632\u06cc \u0628\u0647\u062a\u0631 \u0645\u06cc \u062a\u0648\u0627\u0646\u062f \u062a\u0639\u0645\u06cc\u0645 \u06cc\u0627\u0628\u062f.  \u062f\u0631 \u0645\u0648\u0631\u062f CIFAR100 \u060c \u062f\u0631 \u0647\u0631 \u06a9\u0644\u0627\u0633 \u0646\u0645\u0648\u0646\u0647 \u0647\u0627\u06cc \u0622\u0645\u0648\u0632\u0634\u06cc \u0632\u06cc\u0627\u062f\u06cc \u0648\u062c\u0648\u062f \u0646\u062f\u0627\u0631\u062f \u060c \u0648 \u0627\u06cc\u0646 \u0627\u062d\u062a\u0645\u0627\u0644\u0627\u064b \u0627\u0632 \u06cc\u0627\u062f\u06af\u06cc\u0631\u06cc \u062e\u0648\u0628 \u0645\u062f\u0644 \u0642\u0628\u0644\u06cc \u062c\u0644\u0648\u06af\u06cc\u0631\u06cc \u0645\u06cc \u06a9\u0646\u062f.  \u0628\u06cc\u0627\u06cc\u06cc\u062f \u0622\u0646 \u0631\u0627 \u0627\u0645\u062a\u062d\u0627\u0646 \u06a9\u0646\u06cc\u0645:<\/p>\n<pre><code class=\"hljs\">checkpoint = keras.callbacks.ModelCheckpoint(<span class=\"hljs-string\">\"simplified_cnn_model_cifar100.h5\"<\/span>, save_best_only=<span class=\"hljs-literal\">True<\/span>)\nearly_stopping = keras.callbacks.EarlyStopping(patience=<span class=\"hljs-number\">10<\/span>, restore_best_weights=<span class=\"hljs-literal\">True<\/span>)\n\nmodel = keras.models.Sequential((\n    keras.layers.Conv2D(<span class=\"hljs-number\">32<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, kernel_initializer=<span class=\"hljs-string\">\"he_normal\"<\/span>, kernel_regularizer=keras.regularizers.l2(l=<span class=\"hljs-number\">0.01<\/span>), padding=<span class=\"hljs-string\">'same'<\/span>, input_shape=(<span class=\"hljs-number\">32<\/span>, <span class=\"hljs-number\">32<\/span>, <span class=\"hljs-number\">3<\/span>)),\n    keras.layers.BatchNormalization(),\n    keras.layers.Conv2D(<span class=\"hljs-number\">32<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.MaxPooling2D(<span class=\"hljs-number\">2<\/span>),\n    keras.layers.Dropout(<span class=\"hljs-number\">0.4<\/span>),\n    \n    keras.layers.Conv2D(<span class=\"hljs-number\">64<\/span>, <span class=\"hljs-number\">2<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.Conv2D(<span class=\"hljs-number\">64<\/span>, <span class=\"hljs-number\">2<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.MaxPooling2D(<span class=\"hljs-number\">2<\/span>),\n    keras.layers.Dropout(<span class=\"hljs-number\">0.4<\/span>),\n    \n    keras.layers.Conv2D(<span class=\"hljs-number\">128<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.Conv2D(<span class=\"hljs-number\">128<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.MaxPooling2D(<span class=\"hljs-number\">2<\/span>),\n    keras.layers.Dropout(<span class=\"hljs-number\">0.5<\/span>),\n    \n    keras.layers.Flatten(),    \n    keras.layers.Dense(<span class=\"hljs-number\">256<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.Dropout(<span class=\"hljs-number\">0.3<\/span>),\n    keras.layers.Dense(<span class=\"hljs-number\">100<\/span>, activation=<span class=\"hljs-string\">'softmax'<\/span>)\n))\n\nmodel.<span class=\"hljs-built_in\">compile<\/span>(loss=<span class=\"hljs-string\">\"sparse_categorical_crossentropy\"<\/span>,\n              optimizer=keras.optimizers.Adam(learning_rate=<span class=\"hljs-number\">1e-3<\/span>),\n              metrics=(<span class=\"hljs-string\">\"accuracy\"<\/span>))\n\nhistory = model.fit(X_train, \n                    Y_train, \n                    epochs=<span class=\"hljs-number\">150<\/span>,\n                    batch_size=<span class=\"hljs-number\">64<\/span>,\n                    validation_data=(X_valid, Y_valid),\n                    callbacks=(checkpoint, early_stopping))\n<\/code><\/pre>\n<pre><code class=\"hljs\">Epoch <span class=\"hljs-number\">1<\/span>\/<span class=\"hljs-number\">150<\/span>\n<span class=\"hljs-number\">704<\/span>\/<span class=\"hljs-number\">704<\/span> (==============================) - 96s 135ms\/step - loss: <span class=\"hljs-number\">4.4432<\/span> - accuracy: <span class=\"hljs-number\">0.1112<\/span> - val_loss: <span class=\"hljs-number\">3.7893<\/span> - val_accuracy: <span class=\"hljs-number\">0.1702<\/span>\n...\nEpoch <span class=\"hljs-number\">48<\/span>\/<span class=\"hljs-number\">150<\/span>\n<span class=\"hljs-number\">704<\/span>\/<span class=\"hljs-number\">704<\/span> (==============================) - 92s 131ms\/step - loss: <span class=\"hljs-number\">1.2550<\/span> - accuracy: <span class=\"hljs-number\">0.6370<\/span> - val_loss: <span class=\"hljs-number\">1.7147<\/span> - val_accuracy: <span class=\"hljs-number\">0.5466<\/span>\n<\/code><\/pre>\n<p>\u0627\u06cc\u0646 \u0641\u0644\u0627\u062a \u0627\u0633\u062a \u0648 \u0648\u0627\u0642\u0639\u0627 \u0646\u0645\u06cc \u062a\u0648\u0627\u0646\u062f \u062f\u0627\u062f\u0647 \u0647\u0627 \u0631\u0627 \u062a\u0639\u0645\u06cc\u0645 \u062f\u0647\u062f.  \u062f\u0631 \u0627\u06cc\u0646 \u0645\u0648\u0631\u062f\u060c \u0645\u0645\u06a9\u0646 \u0627\u0633\u062a \u062a\u0642\u0635\u06cc\u0631 \u0645\u062f\u0644 \u0646\u0628\u0627\u0634\u062f &#8211; \u0634\u0627\u06cc\u062f \u0628\u0631\u0627\u06cc \u06a9\u0627\u0631 \u0645\u0646\u0627\u0633\u0628 \u0628\u0627\u0634\u062f\u060c \u0628\u0647 \u062e\u0635\u0648\u0635 \u0628\u0627 \u062a\u0648\u062c\u0647 \u0628\u0647 \u062f\u0642\u062a \u0628\u0627\u0644\u0627 \u0631\u0648\u06cc \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 CIFAR10\u060c \u06a9\u0647 \u0634\u06a9\u0644 \u0648\u0631\u0648\u062f\u06cc \u06cc\u06a9\u0633\u0627\u0646 \u0648 \u062a\u0635\u0627\u0648\u06cc\u0631 \u0645\u0634\u0627\u0628\u0647 \u062f\u0631 \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u062f\u0627\u0631\u062f.  \u0628\u0647 \u0646\u0638\u0631 \u0645\u06cc\u200c\u0631\u0633\u062f \u06a9\u0647 \u0627\u06cc\u0646 \u0645\u062f\u0644 \u0645\u06cc\u200c\u062a\u0648\u0627\u0646\u062f \u0628\u0627 \u0627\u0634\u06a9\u0627\u0644 \u06a9\u0644\u06cc \u062f\u0642\u06cc\u0642 \u0628\u0627\u0634\u062f\u060c \u0627\u0645\u0627 \u0646\u0647 \u062a\u0645\u0627\u06cc\u0632 \u0628\u06cc\u0646 \u0627\u0634\u06a9\u0627\u0644 \u0638\u0631\u06cc\u0641.<\/p>\n<p>\u0645\u062f\u0644 \u0633\u0627\u062f\u0647 \u062a\u0631 \u062f\u0631 \u0648\u0627\u0642\u0639 \u0627\u0632 \u0646\u0638\u0631 \u062f\u0642\u062a \u0627\u0639\u062a\u0628\u0627\u0631 \u0633\u0646\u062c\u06cc \u0628\u0647\u062a\u0631 \u0627\u0632 \u067e\u06cc\u0686\u06cc\u062f\u0647 \u062a\u0631 \u0639\u0645\u0644 \u0645\u06cc \u06a9\u0646\u062f &#8211; \u0628\u0646\u0627\u0628\u0631\u0627\u06cc\u0646 CNN \u067e\u06cc\u0686\u06cc\u062f\u0647 \u062a\u0631 \u0627\u06cc\u0646 \u062c\u0632\u0626\u06cc\u0627\u062a \u062e\u0648\u0628 \u0631\u0627 \u0628\u0647 \u0647\u06cc\u0686 \u0648\u062c\u0647 \u0628\u0647\u062a\u0631 \u0646\u0645\u06cc \u06a9\u0646\u062f.  \u062f\u0631 \u0627\u06cc\u0646\u062c\u0627\u060c \u0645\u0634\u06a9\u0644 \u0628\u0647 \u0627\u062d\u062a\u0645\u0627\u0644 \u0632\u06cc\u0627\u062f \u062f\u0631 \u0627\u06cc\u0646 \u0648\u0627\u0642\u0639\u06cc\u062a \u0646\u0647\u0641\u062a\u0647 \u0627\u0633\u062a \u06a9\u0647 \u062a\u0646\u0647\u0627 500 \u062a\u0635\u0648\u06cc\u0631 \u0622\u0645\u0648\u0632\u0634\u06cc \u062f\u0631 \u0647\u0631 \u06a9\u0644\u0627\u0633 \u0648\u062c\u0648\u062f \u062f\u0627\u0631\u062f \u06a9\u0647 \u0648\u0627\u0642\u0639\u0627\u064b \u06a9\u0627\u0641\u06cc \u0646\u06cc\u0633\u062a.  \u062f\u0631 \u0634\u0628\u06a9\u0647 \u067e\u06cc\u0686\u06cc\u062f\u0647 \u062a\u0631 \u060c \u0627\u06cc\u0646 \u0645\u0646\u062c\u0631 \u0628\u0647 \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0645\u06cc \u0634\u0648\u062f \u060c \u0632\u06cc\u0631\u0627 \u062a\u0646\u0648\u0639 \u06a9\u0627\u0641\u06cc \u0648\u062c\u0648\u062f \u0646\u062f\u0627\u0631\u062f &#8211; \u062f\u0631 \u0635\u0648\u0631\u062a \u0633\u0627\u062f\u0647 \u06a9\u0631\u062f\u0646 \u0628\u0631\u0627\u06cc \u062c\u0644\u0648\u06af\u06cc\u0631\u06cc \u0627\u0632 \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u060c \u0627\u06cc\u0646 \u0627\u0645\u0631 \u0628\u0627\u0639\u062b \u0645\u06cc \u0634\u0648\u062f \u06a9\u0647 \u0645\u062c\u062f\u062f\u0627\u064b \u06a9\u0645\u0631\u0646\u06af \u0634\u0648\u062f \u060c \u0647\u06cc\u0686 \u062a\u0646\u0648\u0639 \u0648\u062c\u0648\u062f \u0646\u062f\u0627\u0631\u062f.<\/p>\n<blockquote>\n<p>\u0628\u0647 \u0647\u0645\u06cc\u0646 \u062f\u0644\u06cc\u0644 \u0627\u0633\u062a \u06a9\u0647 \u0627\u06a9\u062b\u0631\u06cc\u062a \u0642\u0631\u06cc\u0628 \u0628\u0647 \u0627\u062a\u0641\u0627\u0642 \u0645\u0642\u0627\u0644\u0627\u062a\u06cc \u06a9\u0647 \u0642\u0628\u0644\u0627\u064b \u067e\u06cc\u0648\u0646\u062f \u062f\u0627\u062f\u0647 \u0634\u062f\u0647 \u0627\u0646\u062f \u0648 \u0627\u06a9\u062b\u0631\u06cc\u062a \u0642\u0631\u06cc\u0628 \u0628\u0647 \u0627\u062a\u0641\u0627\u0642 \u0634\u0628\u06a9\u0647 \u0647\u0627 \u062f\u0627\u062f\u0647 \u0647\u0627\u06cc \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 CIFAR100 \u0631\u0627 \u0627\u0641\u0632\u0627\u06cc\u0634 \u0645\u06cc \u062f\u0647\u0646\u062f.<\/p>\n<\/blockquote>\n<p>\u0627\u06cc\u0646 \u0648\u0627\u0642\u0639\u0627\u064b \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u0627\u06cc \u0646\u06cc\u0633\u062a \u06a9\u0647 \u0628\u0647 \u0631\u0627\u062d\u062a\u06cc \u0628\u062a\u0648\u0627\u0646 \u062f\u0642\u062a \u0628\u0627\u0644\u0627\u06cc\u06cc \u0628\u0631\u0627\u06cc \u0622\u0646 \u0628\u0647 \u062f\u0633\u062a \u0622\u0648\u0631\u062f on\u0628\u0631\u062e\u0644\u0627\u0641 \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u0627\u0631\u0642\u0627\u0645 \u062f\u0633\u062a\u200c\u0646\u0648\u06cc\u0633 MNIST\u060c \u0648 \u06cc\u06a9 \u0633\u06cc\u200c\u0627\u0646\u200c\u0627\u0646 \u0633\u0627\u062f\u0647 \u0645\u0627\u0646\u0646\u062f \u0622\u0646\u0686\u0647 \u0645\u0627 \u0645\u06cc\u200c\u0633\u0627\u0632\u06cc\u0645\u060c \u0627\u062d\u062a\u0645\u0627\u0644\u0627\u064b \u0622\u0646 \u0631\u0627 \u0628\u0631\u0627\u06cc \u062f\u0642\u062a \u0628\u0627\u0644\u0627 \u06a9\u0627\u0647\u0634 \u0646\u0645\u06cc\u200c\u062f\u0647\u062f.  \u0641\u0642\u0637 \u062a\u0639\u062f\u0627\u062f \u06a9\u0644\u0627\u0633 \u0647\u0627\u06cc \u06a9\u0627\u0645\u0644\u0627\u064b \u062e\u0627\u0635 \u0631\u0627 \u0628\u0647 \u062e\u0627\u0637\u0631 \u0628\u0633\u067e\u0627\u0631\u06cc\u062f\u060c \u0628\u0631\u062e\u06cc \u0627\u0632 \u062a\u0635\u0627\u0648\u06cc\u0631 \u0686\u0642\u062f\u0631 \u0628\u06cc \u0627\u0637\u0644\u0627\u0639 \u0647\u0633\u062a\u0646\u062f \u0648 <em>\u0686\u0642\u062f\u0631 \u062f\u0627\u0646\u0634 \u0642\u0628\u0644\u06cc \u0627\u0646\u0633\u0627\u0646\u0647\u0627 \u0628\u0627\u06cc\u062f \u0628\u06cc\u0646 \u0627\u06cc\u0646\u0647\u0627 \u062a\u0634\u062e\u06cc\u0635 \u062f\u0647\u0646\u062f<\/em>.<\/p>\n<p>\u0628\u06cc\u0627\u06cc\u06cc\u062f \u0628\u0627 \u0627\u0641\u0632\u0648\u062f\u0646 \u0686\u0646\u062f \u062a\u0635\u0648\u06cc\u0631 \u0648 \u06af\u0633\u062a\u0631\u0634 \u0645\u0635\u0646\u0648\u0639\u06cc \u062f\u0627\u062f\u0647 \u0647\u0627\u06cc \u0622\u0645\u0648\u0632\u0634\u06cc \u062a\u0645\u0627\u0645 \u062a\u0644\u0627\u0634 \u062e\u0648\u062f \u0631\u0627 \u0628\u06a9\u0646\u06cc\u0645 \u062a\u0627 \u062d\u062f\u0627\u0642\u0644 \u0633\u0639\u06cc \u06a9\u0646\u06cc\u0645 \u062f\u0642\u062a \u0628\u0627\u0644\u0627\u062a\u0631\u06cc \u062f\u0627\u0634\u062a\u0647 \u0628\u0627\u0634\u06cc\u0645.  \u0628\u0647 \u062e\u0627\u0637\u0631 \u062f\u0627\u0634\u062a\u0647 \u0628\u0627\u0634\u06cc\u062f \u06a9\u0647 CIFAR100 \u06cc\u06a9 \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u0648\u0627\u0642\u0639\u0627\u064b \u062f\u0634\u0648\u0627\u0631 \u0628\u0631\u0627\u06cc \u062f\u0633\u062a\u06cc\u0627\u0628\u06cc \u0628\u0647 \u062f\u0642\u062a \u0628\u0627\u0644\u0627 \u0627\u0633\u062a. \u0631\u0648\u06cc \u0628\u0627 \u0645\u062f\u0644 \u0647\u0627\u06cc \u0633\u0627\u062f\u0647  \u0645\u062f\u0644 \u0647\u0627\u06cc \u067e\u06cc\u0634\u0631\u0641\u062a\u0647 \u0627\u0632 \u062a\u06a9\u0646\u06cc\u06a9 \u0647\u0627\u06cc \u0645\u062a\u0641\u0627\u0648\u062a \u0648 \u062c\u062f\u06cc\u062f \u0628\u0631\u0627\u06cc \u0627\u0632 \u0628\u06cc\u0646 \u0628\u0631\u062f\u0646 \u062e\u0637\u0627\u0647\u0627 \u0627\u0633\u062a\u0641\u0627\u062f\u0647 \u0645\u06cc \u06a9\u0646\u0646\u062f \u0648 \u0628\u0633\u06cc\u0627\u0631\u06cc \u0627\u0632 \u0627\u06cc\u0646 \u0645\u062f\u0644 \u0647\u0627 \u062d\u062a\u06cc <em>CNN \u0647\u0627<\/em> &#8211; \u0622\u0646\u0647\u0627 \u0647\u0633\u062a\u0646\u062f <em>\u0645\u0628\u062f\u0644 \u0647\u0627<\/em>.<\/p>\n<blockquote>\n<p>\u0627\u06af\u0631 \u062f\u0648\u0633\u062a \u062f\u0627\u0631\u06cc\u062f \u0646\u06af\u0627\u0647\u06cc \u0628\u0647 \u0686\u0634\u0645 \u0627\u0646\u062f\u0627\u0632 \u0627\u06cc\u0646 \u0645\u062f\u0644 \u0647\u0627 \u0628\u06cc\u0646\u062f\u0627\u0632\u06cc\u062f\u060c <a rel=\"nofollow noopener noreferrer\" target=\"_blank\" href=\"https:\/\/paperswithcode.com\/sota\/image-classification-on-cifar-100\">PapersWithCode<\/a> \u0645\u062c\u0645\u0648\u0639\u0647 \u0627\u06cc \u0632\u06cc\u0628\u0627 \u0627\u0632 \u0645\u0642\u0627\u0644\u0627\u062a\u060c \u06a9\u062f \u0645\u0646\u0628\u0639 \u0648 \u0646\u062a\u0627\u06cc\u062c \u0631\u0627 \u0627\u0646\u062c\u0627\u0645 \u062f\u0627\u062f\u0647 \u0627\u0633\u062a.<\/p>\n<\/blockquote>\n<h3 id=\"dataaugmentationwithkerasimagedatageneratorclass\"><span class=\"ez-toc-section\" id=\"%d8%a7%d9%81%d8%b2%d8%a7%db%8c%d8%b4_%d8%af%d8%a7%d8%af%d9%87_%d9%87%d8%a7_%d8%a8%d8%a7_%da%a9%d9%84%d8%a7%d8%b3_imagedatagenerator_keras\"><\/span>\u0627\u0641\u0632\u0627\u06cc\u0634 \u062f\u0627\u062f\u0647 \u0647\u0627 \u0628\u0627 \u06a9\u0644\u0627\u0633 ImageDataGenerator Keras<span class=\"ez-toc-section-end\"><\/span><\/h3>\n<p>\u0622\u06cc\u0627 \u0627\u0641\u0632\u0627\u06cc\u0634 \u062f\u0627\u062f\u0647 \u06a9\u0645\u06a9\u06cc \u062e\u0648\u0627\u0647\u062f \u06a9\u0631\u062f\u061f  \u0645\u0639\u0645\u0648\u0644\u0627\u064b \u0627\u06cc\u0646 \u06a9\u0627\u0631 \u0631\u0627 \u0645\u06cc \u06a9\u0646\u062f\u060c \u0627\u0645\u0627 \u0628\u0627 \u06cc\u06a9 <em>\u062c\u062f\u06cc<\/em> \u0639\u062f\u0645 \u0648\u062c\u0648\u062f \u062f\u0627\u062f\u0647 \u0647\u0627\u06cc \u0622\u0645\u0648\u0632\u0634 \u0645\u0627\u0646\u0646\u062f \u0645\u0627 \u0628\u0627 \u0622\u0646 \u0631\u0648\u0628\u0631\u0648 \u0647\u0633\u062a\u06cc\u0645 \u060c \u0641\u0642\u0637 \u0645\u06cc \u062a\u0648\u0627\u0646\u06cc\u062f \u0628\u0627 \u0686\u0631\u062e\u0634 \u0647\u0627\u06cc \u062a\u0635\u0627\u062f\u0641\u06cc \u060c \u0686\u0631\u062e\u0634 \u060c \u06a9\u0634\u062a \u0648 \u063a\u06cc\u0631\u0647 \u0627\u0646\u062c\u0627\u0645 \u062f\u0647\u06cc\u062f. \u0627\u06af\u0631 \u06cc\u06a9 \u0645\u0639\u0645\u0627\u0631\u06cc \u0646\u062a\u0648\u0627\u0646\u062f \u0628\u0647 \u062e\u0648\u0628\u06cc \u062a\u0639\u0645\u06cc\u0645 \u062f\u0647\u062f \u0631\u0648\u06cc \u06cc\u06a9 \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647\u060c \u0627\u062d\u062a\u0645\u0627\u0644\u0627\u064b \u0622\u0646 \u0631\u0627 \u0627\u0632 \u0637\u0631\u06cc\u0642 \u0627\u0641\u0632\u0627\u06cc\u0634 \u062f\u0627\u062f\u0647 \u0647\u0627 \u062a\u0642\u0648\u06cc\u062a \u062e\u0648\u0627\u0647\u06cc\u062f \u06a9\u0631\u062f\u060c \u0627\u0645\u0627 \u0627\u062d\u062a\u0645\u0627\u0644\u0627\u064b \u0632\u06cc\u0627\u062f \u0646\u062e\u0648\u0627\u0647\u062f \u0628\u0648\u062f.<\/p>\n<p>\u0647\u0645\u0627\u0646\u0637\u0648\u0631 \u06a9\u0647 \u06af\u0641\u062a\u0647 \u0634\u062f\u060c \u0628\u06cc\u0627\u06cc\u06cc\u062f \u0627\u0632 Keras \u0627\u0633\u062a\u0641\u0627\u062f\u0647 \u06a9\u0646\u06cc\u0645 <code>ImageDataGenerator<\/code> \u06a9\u0644\u0627\u0633 \u0628\u0631\u0627\u06cc \u062a\u0648\u0644\u06cc\u062f \u0628\u0631\u062e\u06cc \u062f\u0627\u062f\u0647 \u0647\u0627\u06cc \u0622\u0645\u0648\u0632\u0634\u06cc \u062c\u062f\u06cc\u062f \u0628\u0627 \u062a\u063a\u06cc\u06cc\u0631\u0627\u062a \u062a\u0635\u0627\u062f\u0641\u06cc\u060c \u0628\u0647 \u0627\u0645\u06cc\u062f \u0628\u0647\u0628\u0648\u062f \u062f\u0642\u062a \u0645\u062f\u0644.  \u0627\u06af\u0631 \u0628\u0647\u0628\u0648\u062f \u06cc\u0627\u0628\u062f \u060c \u0646\u0628\u0627\u06cc\u062f \u0628\u0627 \u0645\u0642\u062f\u0627\u0631 \u0632\u06cc\u0627\u062f\u06cc \u0628\u0627\u0634\u062f \u060c \u0648 \u0628\u0647 \u0627\u062d\u062a\u0645\u0627\u0644 \u0632\u06cc\u0627\u062f \u062f\u0648\u0628\u0627\u0631\u0647 \u0645\u06cc \u062a\u0648\u0627\u0646\u062f \u0628\u0647 \u0635\u0648\u0631\u062a \u062c\u0632\u0626\u06cc \u0627\u0632 \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u0647\u0627 \u0628\u062f\u0648\u0646 \u062a\u0648\u0627\u0646\u0627\u06cc\u06cc \u062a\u0639\u0645\u06cc\u0645 \u062e\u0648\u0628 \u06cc\u0627 \u06a9\u0627\u0645\u0644\u0627\u064b \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u062f\u0627\u062f\u0647 \u0647\u0627 \u0628\u0627\u0632\u06af\u0631\u062f\u062f.<\/p>\n<p>\u0628\u0627 \u062a\u0648\u062c\u0647 \u0628\u0647 \u062a\u063a\u06cc\u06cc\u0631\u0627\u062a \u062a\u0635\u0627\u062f\u0641\u06cc \u062b\u0627\u0628\u062a \u062f\u0631 \u062f\u0627\u062f\u0647\u200c\u0647\u0627\u060c \u0645\u062f\u0644 \u06a9\u0645\u062a\u0631 \u0628\u0631\u0627\u0632\u0634 \u0645\u06cc\u200c\u06a9\u0646\u062f \u0631\u0648\u06cc \u0628\u0647 \u0647\u0645\u0627\u0646 \u062a\u0639\u062f\u0627\u062f \u062f\u0648\u0631\u0647\u200c\u0647\u0627\u060c \u0632\u06cc\u0631\u0627 \u062a\u063a\u06cc\u06cc\u0631\u0627\u062a \u0628\u0627\u0639\u062b \u0645\u06cc\u200c\u0634\u0648\u062f \u0628\u0647 \u062f\u0627\u062f\u0647\u200c\u0647\u0627\u06cc \u00ab\u062c\u062f\u06cc\u062f\u00bb \u062a\u0637\u0628\u06cc\u0642 \u06cc\u0627\u0628\u062f.  \u0628\u06cc\u0627\u06cc\u06cc\u062f \u0622\u0646 \u0631\u0627 \u0628\u0631\u0627\u06cc \u0645\u062b\u0644\u0627\u064b 300 \u062f\u0648\u0631\u0647 \u0627\u062c\u0631\u0627 \u06a9\u0646\u06cc\u0645\u060c \u06a9\u0647 \u0628\u0647 \u0637\u0648\u0631 \u0642\u0627\u0628\u0644 \u062a\u0648\u062c\u0647\u06cc \u0628\u06cc\u0634\u062a\u0631 \u0627\u0632 \u0628\u0642\u06cc\u0647 \u0634\u0628\u06a9\u0647 \u0647\u0627\u06cc\u06cc \u0627\u0633\u062a \u06a9\u0647 \u0645\u0627 \u0622\u0645\u0648\u0632\u0634 \u062f\u0627\u062f\u0647 \u0627\u06cc\u0645.  \u0628\u062f\u0648\u0646 \u0627\u06cc\u0646 \u0627\u0645\u06a9\u0627\u0646 \u067e\u0630\u06cc\u0631 \u0627\u0633\u062a <em>\u0639\u0645\u062f\u0647<\/em> \u0645\u062c\u062f\u062f\u0627\u064b \u0628\u0647 \u062f\u0644\u06cc\u0644 \u062a\u063a\u06cc\u06cc\u0631\u0627\u062a \u062a\u0635\u0627\u062f\u0641\u06cc \u0627\u06cc\u062c\u0627\u062f \u0634\u062f\u0647 \u062f\u0631 \u062a\u0635\u0627\u0648\u06cc\u0631 \u062f\u0631 \u062d\u0627\u0644\u06cc \u06a9\u0647 \u0622\u0646\u0647\u0627 \u062f\u0631 \u062d\u0627\u0644 \u062c\u0631\u06cc\u0627\u0646 \u0647\u0633\u062a\u0646\u062f\u060c \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0645\u0646\u0627\u0633\u0628 \u0627\u0633\u062a:<\/p>\n<pre><code class=\"hljs\">checkpoint = keras.callbacks.ModelCheckpoint(<span class=\"hljs-string\">\"augmented_cnn.h5\"<\/span>, save_best_only=<span class=\"hljs-literal\">True<\/span>)\n\nmodel = keras.models.Sequential((\n    keras.layers.Conv2D(<span class=\"hljs-number\">64<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, kernel_initializer=<span class=\"hljs-string\">\"he_normal\"<\/span>, kernel_regularizer=keras.regularizers.l2(l=<span class=\"hljs-number\">0.01<\/span>), padding=<span class=\"hljs-string\">'same'<\/span>, input_shape=(<span class=\"hljs-number\">32<\/span>, <span class=\"hljs-number\">32<\/span>, <span class=\"hljs-number\">3<\/span>)),\n    keras.layers.Conv2D(<span class=\"hljs-number\">64<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.MaxPooling2D(<span class=\"hljs-number\">2<\/span>),\n    keras.layers.Dropout(<span class=\"hljs-number\">0.4<\/span>),\n    \n    keras.layers.Conv2D(<span class=\"hljs-number\">128<\/span>, <span class=\"hljs-number\">2<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.Conv2D(<span class=\"hljs-number\">128<\/span>, <span class=\"hljs-number\">2<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.Conv2D(<span class=\"hljs-number\">128<\/span>, <span class=\"hljs-number\">2<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.MaxPooling2D(<span class=\"hljs-number\">2<\/span>),\n    keras.layers.Dropout(<span class=\"hljs-number\">0.4<\/span>),\n    \n    keras.layers.Conv2D(<span class=\"hljs-number\">256<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.Conv2D(<span class=\"hljs-number\">256<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.Conv2D(<span class=\"hljs-number\">256<\/span>, <span class=\"hljs-number\">3<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>, padding=<span class=\"hljs-string\">'same'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.MaxPooling2D(<span class=\"hljs-number\">2<\/span>),\n    keras.layers.Dropout(<span class=\"hljs-number\">0.4<\/span>),\n    \n    keras.layers.Flatten(),    \n    keras.layers.Dense(<span class=\"hljs-number\">512<\/span>, activation=<span class=\"hljs-string\">'relu'<\/span>),\n    keras.layers.BatchNormalization(),\n    keras.layers.Dropout(<span class=\"hljs-number\">0.3<\/span>),\n    keras.layers.Dense(<span class=\"hljs-number\">100<\/span>, activation=<span class=\"hljs-string\">'softmax'<\/span>)\n))\n\n    \ntrain_datagen = ImageDataGenerator(rotation_range=<span class=\"hljs-number\">30<\/span>,\n        height_shift_range=<span class=\"hljs-number\">0.2<\/span>,\n        width_shift_range=<span class=\"hljs-number\">0.2<\/span>,\n        shear_range=<span class=\"hljs-number\">0.2<\/span>,\n        zoom_range=<span class=\"hljs-number\">0.2<\/span>,\n        horizontal_flip=<span class=\"hljs-literal\">True<\/span>,\n        vertical_flip=<span class=\"hljs-literal\">True<\/span>,\n        fill_mode=<span class=\"hljs-string\">'nearest'<\/span>)\n\nvalid_datagen = ImageDataGenerator()\n\ntrain_datagen.fit(X_train)\nvalid_datagen.fit(X_valid)\n\ntrain_generator = train_datagen.flow(X_train, Y_train, batch_size=<span class=\"hljs-number\">128<\/span>)\nvalid_generator = valid_datagen.flow(X_valid, Y_valid, batch_size=<span class=\"hljs-number\">128<\/span>)\n\nmodel.<span class=\"hljs-built_in\">compile<\/span>(loss=<span class=\"hljs-string\">\"sparse_categorical_crossentropy\"<\/span>,\n              optimizer=keras.optimizers.Adam(learning_rate=<span class=\"hljs-number\">1e-3<\/span>, decay=<span class=\"hljs-number\">1e-6<\/span>),\n              metrics=(<span class=\"hljs-string\">\"accuracy\"<\/span>))\n\nhistory = model.fit(train_generator, \n                    epochs=<span class=\"hljs-number\">300<\/span>,\n                    batch_size=<span class=\"hljs-number\">128<\/span>,\n                    steps_per_epoch=<span class=\"hljs-built_in\">len<\/span>(X_train)\/\/<span class=\"hljs-number\">128<\/span>,\n                    validation_data=valid_generator,\n                    callbacks=(checkpoint))\n<\/code><\/pre>\n<pre><code class=\"hljs\">Epoch 1\/300\n351\/351 (==============================) - 16s 44ms\/step - loss: 5.3788 - accuracy: 0.0487 - val_loss: 5.3474 - val_accuracy: 0.0440\n...\nEpoch 300\/300\n351\/351 (==============================) - 15s 43ms\/step - loss: 1.0571 - accuracy: 0.6895 - val_loss: 2.0005 - val_accuracy: 0.5532\n<\/code><\/pre>\n<p><img decoding=\"async\" src=\"https:\/\/rasanegar.com\/blog\/wp-content\/uploads\/2024\/01\/overfitting-is-your-friend-not-your-foe-9.png\" alt=\"\" title=\"\"><\/p>\n<p>\u0639\u0645\u0644\u06a9\u0631\u062f \u0645\u062f\u0644 \u0628\u0627 ~55\u066a \u0631\u0648\u06cc \u0645\u062c\u0645\u0648\u0639\u0647 \u0627\u0639\u062a\u0628\u0627\u0631 \u0633\u0646\u062c\u06cc\u060c \u0648 \u0647\u0646\u0648\u0632 \u0647\u0645 \u062a\u0627 \u062d\u062f\u06cc \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0628\u0631\u0627\u0632\u0634 \u062f\u0627\u062f\u0647 \u0627\u0633\u062a.  \u0627\u06cc\u0646 <code>val_loss<\/code> \u0627\u0632 \u067e\u0627\u06cc\u06cc\u0646 \u0631\u0641\u062a\u0646 \u0645\u062a\u0648\u0642\u0641 \u0634\u062f\u0647 \u0627\u0633\u062a \u0648 \u062d\u062a\u06cc \u0628\u0627 \u0627\u0631\u062a\u0641\u0627\u0639\u06cc \u0628\u0627\u0644\u0627\u062a\u0631 \u06a9\u0627\u0645\u0644\u0627\u064b \u0635\u062e\u0631\u0647 \u0627\u06cc \u0627\u0633\u062a <code>batch_size<\/code>.<\/p>\n<p>\u0627\u06cc\u0646 \u0634\u0628\u06a9\u0647 \u0628\u0647 \u0633\u0627\u062f\u06af\u06cc \u0646\u0645\u06cc \u062a\u0648\u0627\u0646\u062f \u062f\u0627\u062f\u0647 \u0647\u0627 \u0631\u0627 \u0628\u0627 \u062f\u0642\u062a \u0628\u0627\u0644\u0627 \u0628\u06cc\u0627\u0645\u0648\u0632\u062f \u0648 \u0645\u062a\u0646\u0627\u0633\u0628 \u06a9\u0646\u062f \u060c \u062d\u062a\u06cc \u0627\u06af\u0631 \u062a\u063a\u06cc\u06cc\u0631\u0627\u062a \u0622\u0646 \u062f\u0627\u0631\u0627\u06cc \u0638\u0631\u0641\u06cc\u062a \u0622\u0646\u062a\u0631\u0648\u067e\u06cc \u0628\u0631\u0627\u06cc \u0627\u0641\u0632\u0627\u06cc\u0634 \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u062f\u0627\u062f\u0647 \u0647\u0627 \u0628\u0627\u0634\u062f.<\/p>\n<h2 id=\"conclusion\"><span class=\"ez-toc-section\" id=\"%d9%86%d8%aa%db%8c%d8%ac%d9%87%d8%9f\"><\/span>\u0646\u062a\u06cc\u062c\u0647\u061f<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>\u062a\u0646\u0627\u0633\u0628 \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0630\u0627\u062a\u0627\u064b \u0686\u06cc\u0632 \u0628\u062f\u06cc \u0646\u06cc\u0633\u062a &#8211; \u0641\u0642\u0637 \u0647\u0645\u06cc\u0646 \u0627\u0633\u062a <em>\u06cc\u06a9 \u0686\u06cc\u0632<\/em>.  \u0646\u0647 \u060c \u0634\u0645\u0627 \u0646\u0645\u06cc \u062e\u0648\u0627\u0647\u06cc\u062f \u0645\u062f\u0644 \u0647\u0627\u06cc \u0627\u0646\u062a\u0647\u0627\u06cc\u06cc \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u060c \u0627\u0645\u0627 \u0646\u0628\u0627\u06cc\u062f \u0628\u0647 \u0639\u0646\u0648\u0627\u0646 \u0637\u0627\u0639\u0648\u0646 \u0631\u0641\u062a\u0627\u0631 \u06a9\u0631\u062f \u0648 \u062d\u062a\u06cc \u0645\u06cc \u062a\u0648\u0627\u0646\u062f \u0646\u0634\u0627\u0646\u0647 \u062e\u0648\u0628\u06cc \u0628\u0627\u0634\u062f \u06a9\u0647 \u06cc\u06a9 \u0645\u062f\u0644 \u0628\u0627 \u062a\u0648\u062c\u0647 \u0628\u0647 \u062f\u0627\u062f\u0647 \u0647\u0627\u06cc \u0628\u06cc\u0634\u062a\u0631 \u0648 \u06cc\u06a9 \u0645\u0631\u062d\u0644\u0647 \u0633\u0627\u062f\u0647 \u0633\u0627\u0632\u06cc \u0645\u06cc \u062a\u0648\u0627\u0646\u062f \u0628\u0647\u062a\u0631 \u0639\u0645\u0644 \u06a9\u0646\u062f.  \u0627\u06cc\u0646 \u0628\u0647 \u0647\u06cc\u0686 \u0648\u062c\u0647 \u062a\u0636\u0645\u06cc\u0646 \u0646\u0634\u062f\u0647 \u0627\u0633\u062a \u0648 \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 CIFAR100 \u0628\u0647 \u0639\u0646\u0648\u0627\u0646 \u0646\u0645\u0648\u0646\u0647 \u0627\u06cc \u0627\u0632 \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u0627\u06cc \u0627\u0633\u062a\u0641\u0627\u062f\u0647 \u0634\u062f\u0647 \u0627\u0633\u062a \u06a9\u0647 \u062a\u0639\u0645\u06cc\u0645 \u0628\u0647 \u062e\u0648\u0628\u06cc \u0628\u0647 \u0622\u0646 \u0622\u0633\u0627\u0646 \u0646\u06cc\u0633\u062a.<\/p>\n<p>\u0646\u06a9\u062a\u0647 \u0627\u06cc\u0646 \u0633\u0631 \u0648 \u0635\u062f\u0627 \u06a9\u0631\u062f\u0646\u060c \u0628\u0627\u0632 \u0647\u0645 \u0645\u062e\u0627\u0644\u0641 \u0646\u0628\u0648\u062f\u0646 &#8211; \u0628\u0644\u06a9\u0647 \u062a\u062d\u0631\u06cc\u06a9 \u0628\u062d\u062b \u0627\u0633\u062a \u0631\u0648\u06cc \u0645\u0648\u0636\u0648\u0639\u06cc \u06a9\u0647 \u0628\u0647 \u0646\u0638\u0631 \u0646\u0645\u06cc \u0631\u0633\u062f \u062c\u0627\u06cc\u06af\u0627\u0647 \u0632\u06cc\u0627\u062f\u06cc \u062f\u0627\u0634\u062a\u0647 \u0628\u0627\u0634\u062f.<\/p>\n<blockquote>\n<p>\u0645\u0646 \u06a9\u06cc \u0647\u0633\u062a\u0645 \u06a9\u0647 \u0627\u06cc\u0646 \u0627\u062f\u0639\u0627 \u0631\u0627 \u062f\u0627\u0634\u062a\u0647 \u0628\u0627\u0634\u0645\u061f<\/p>\n<\/blockquote>\n<p>\u0641\u0642\u0637 \u06a9\u0633\u06cc \u06a9\u0647 \u062f\u0631 \u062e\u0627\u0646\u0647 \u0646\u0634\u0633\u062a\u0647 \u0648 \u0628\u0627 \u0634\u06cc\u0641\u062a\u06af\u06cc \u0639\u0645\u06cc\u0642 \u0646\u0633\u0628\u062a \u0628\u0647 \u0641\u0631\u062f\u0627\u060c \u0627\u06cc\u0646 \u0647\u0646\u0631 \u0631\u0627 \u062a\u0645\u0631\u06cc\u0646 \u0645\u06cc \u06a9\u0646\u062f.<\/p>\n<blockquote>\n<p>\u0622\u06cc\u0627 \u0645\u0646 \u062a\u0648\u0627\u0646\u0627\u06cc\u06cc \u0627\u0634\u062a\u0628\u0627\u0647 \u06a9\u0631\u062f\u0646 \u0631\u0627 \u062f\u0627\u0631\u0645\u061f<\/p>\n<\/blockquote>\n<p>\u062e\u06cc\u0644\u06cc \u0632\u06cc\u0627\u062f.<\/p>\n<blockquote>\n<p>\u0686\u06af\u0648\u0646\u0647 \u0628\u0627\u06cc\u062f \u0627\u06cc\u0646 \u0642\u0637\u0639\u0647 \u0631\u0627 \u0628\u06af\u06cc\u0631\u06cc\u062f\u061f<\/p>\n<\/blockquote>\n<p>\u0622\u0646 \u0631\u0627 \u0647\u0645\u0627\u0646\u0637\u0648\u0631 \u06a9\u0647 \u0645\u0645\u06a9\u0646 \u0627\u0633\u062a \u0628\u06af\u06cc\u0631\u06cc\u062f &#8211; \u062e\u0648\u062f\u062a\u0627\u0646 \u0641\u06a9\u0631 \u06a9\u0646\u06cc\u062f \u06a9\u0647 \u0622\u06cc\u0627 \u0645\u0646\u0637\u0642\u06cc \u0627\u0633\u062a \u06cc\u0627 \u0646\u0647.  \u0627\u06af\u0631 \u0634\u0645\u0627 <em>\u0646\u06a9\u0646<\/em> \u0641\u06a9\u0631 \u0645\u06cc \u06a9\u0646\u0645 \u0645\u0646 \u0628\u0631\u0627\u06cc \u062a\u0648\u062c\u0647 \u0628\u0647 \u0627\u06cc\u0646 \u0645\u0648\u0636\u0648\u0639 \u0645\u0646\u0627\u0633\u0628 \u0646\u06cc\u0633\u062a\u0645\u060c \u0628\u0647 \u0645\u0646 \u0627\u0637\u0644\u0627\u0639 \u062f\u0647\u06cc\u062f.  \u0627\u06af\u0631 \u0641\u06a9\u0631 \u0645\u06cc \u06a9\u0646\u06cc\u062f \u0645\u0646 \u0627\u0634\u062a\u0628\u0627\u0647 \u0645\u06cc \u06a9\u0646\u0645 \u0631\u0648\u06cc \u0627\u06cc\u0646 &#8211; \u0628\u0647 \u0647\u0631 \u062d\u0627\u0644\u060c \u0644\u0637\u0641\u0627 \u0627\u062d\u0633\u0627\u0633 \u06a9\u0646\u06cc\u062f \u0628\u0647 \u0645\u0646 \u0627\u0637\u0644\u0627\u0639 \u062f\u0647\u06cc\u062f \u0648 \u06a9\u0644\u0645\u0627\u062a \u062e\u0648\u062f \u0631\u0627 \u06a9\u0648\u0686\u06a9 \u0646\u06a9\u0646\u06cc\u062f.  \ud83d\ude42<\/p>\n<\/div>\n<p><script>\n                        !function(f,b,e,v,n,t,s)\n                        {if(f.fbq)return;n=f.fbq=function(){n.callMethod?\n                        n.callMethod.apply(n,arguments):n.queue.push(arguments)};\n                        if(!f._fbq)f._fbq=n;n.push=n;n.loaded=!0;n.version='2.0';\n                        n.queue=();t=b.createElement(e);t.async=!0;\n                        t.src=v;s=b.getElementsByTagName(e)(0);\n                        s.parentNode.insertBefore(t,s)}(window, document,'script',\n                        'https:\/\/connect.facebook.net\/en_US\/fbevents.js');\n                        fbq('init', '525232124909042');\n                        fbq('track', 'PageView');\n                    <\/script>    (\u0628\u0631\u0686\u0633\u0628\u200c\u0647\u0627 \u0628\u0647 \u062a\u0631\u062c\u0645\u0647)# python<br \/>\n<br \/><br \/>\n<br \/>\u0645\u0646\u062a\u0634\u0631 \u0634\u062f\u0647 \u062f\u0631 1403-01-08 14:01:05<br \/>\n<\/p>\n\n\n<div class=\"kk-star-ratings kksr-auto kksr-align-center kksr-valign-bottom\"\n    data-payload='{&quot;align&quot;:&quot;center&quot;,&quot;id&quot;:&quot;15093&quot;,&quot;slug&quot;:&quot;default&quot;,&quot;valign&quot;:&quot;bottom&quot;,&quot;ignore&quot;:&quot;&quot;,&quot;reference&quot;:&quot;auto&quot;,&quot;class&quot;:&quot;&quot;,&quot;count&quot;:&quot;0&quot;,&quot;legendonly&quot;:&quot;&quot;,&quot;readonly&quot;:&quot;&quot;,&quot;score&quot;:&quot;0&quot;,&quot;starsonly&quot;:&quot;&quot;,&quot;best&quot;:&quot;5&quot;,&quot;gap&quot;:&quot;5&quot;,&quot;greet&quot;:&quot;\u0627\u0645\u062a\u06cc\u0627\u0632 \u0634\u0645\u0627 \u0628\u0647 \u0627\u06cc\u0646 \u0645\u0637\u0644\u0628&quot;,&quot;legend&quot;:&quot;0\\\/5 (0 \u0631\u0627\u06cc)&quot;,&quot;size&quot;:&quot;30&quot;,&quot;title&quot;:&quot;\u06cc\u0627\u062f\u06af\u06cc\u0631\u06cc \u0645\u0627\u0634\u06cc\u0646\u06cc: \u062a\u0637\u0628\u06cc\u0642 \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u062f\u0648\u0633\u062a \u0634\u0645\u0627\u0633\u062a\u060c \u0646\u0647 \u062f\u0634\u0645\u0646 \u0634\u0645\u0627&quot;,&quot;width&quot;:&quot;0&quot;,&quot;_legend&quot;:&quot;{score}\\\/{best} ({count} \u0631\u0627\u06cc)&quot;,&quot;font_factor&quot;:&quot;1.25&quot;}'>\n            \n<div class=\"kksr-stars\">\n    \n<div class=\"kksr-stars-inactive\">\n            <div class=\"kksr-star\" data-star=\"1\" style=\"padding-left: 5px\">\n            \n\n<div class=\"kksr-icon\" style=\"width: 30px; height: 30px;\"><\/div>\n        <\/div>\n            <div class=\"kksr-star\" data-star=\"2\" style=\"padding-left: 5px\">\n            \n\n<div class=\"kksr-icon\" style=\"width: 30px; height: 30px;\"><\/div>\n        <\/div>\n            <div class=\"kksr-star\" data-star=\"3\" style=\"padding-left: 5px\">\n            \n\n<div class=\"kksr-icon\" style=\"width: 30px; height: 30px;\"><\/div>\n        <\/div>\n            <div class=\"kksr-star\" data-star=\"4\" style=\"padding-left: 5px\">\n            \n\n<div class=\"kksr-icon\" style=\"width: 30px; height: 30px;\"><\/div>\n        <\/div>\n            <div class=\"kksr-star\" data-star=\"5\" style=\"padding-left: 5px\">\n            \n\n<div class=\"kksr-icon\" style=\"width: 30px; height: 30px;\"><\/div>\n        <\/div>\n    <\/div>\n    \n<div class=\"kksr-stars-active\" style=\"width: 0px;\">\n            <div class=\"kksr-star\" style=\"padding-left: 5px\">\n            \n\n<div class=\"kksr-icon\" style=\"width: 30px; height: 30px;\"><\/div>\n        <\/div>\n            <div class=\"kksr-star\" style=\"padding-left: 5px\">\n            \n\n<div class=\"kksr-icon\" style=\"width: 30px; height: 30px;\"><\/div>\n        <\/div>\n            <div class=\"kksr-star\" style=\"padding-left: 5px\">\n            \n\n<div class=\"kksr-icon\" style=\"width: 30px; height: 30px;\"><\/div>\n        <\/div>\n            <div class=\"kksr-star\" style=\"padding-left: 5px\">\n            \n\n<div class=\"kksr-icon\" style=\"width: 30px; height: 30px;\"><\/div>\n        <\/div>\n            <div class=\"kksr-star\" style=\"padding-left: 5px\">\n            \n\n<div class=\"kksr-icon\" style=\"width: 30px; height: 30px;\"><\/div>\n        <\/div>\n    <\/div>\n<\/div>\n                \n\n<div class=\"kksr-legend\" style=\"font-size: 24px;\">\n            <span class=\"kksr-muted\">\u0627\u0645\u062a\u06cc\u0627\u0632 \u0634\u0645\u0627 \u0628\u0647 \u0627\u06cc\u0646 \u0645\u0637\u0644\u0628<\/span>\n    <\/div>\n    <\/div>\n","protected":false},"excerpt":{"rendered":"<p><span class=\"span-reading-time rt-reading-time\" style=\"display: block;\"><span class=\"rt-label rt-prefix\">\u0632\u0645\u0627\u0646 \u0644\u0627\u0632\u0645 \u0628\u0631\u0627\u06cc \u0645\u0637\u0627\u0644\u0639\u0647: <\/span> <span class=\"rt-time\"> 18<\/span> <span class=\"rt-label rt-postfix\">\u062f\u0642\u06cc\u0642\u0647<\/span><\/span>\u0627\u062c\u0627\u0632\u0647 \u062f\u0647\u06cc\u062f \u0639\u0646\u0648\u0627\u0646 \u0628\u0627\u0644\u0642\u0648\u0647 \u062a\u062d\u0631\u06cc\u06a9 \u0622\u0645\u06cc\u0632 \u0631\u0627 \u0628\u0627 \u0627\u06cc\u0646 \u0645\u0642\u062f\u0645\u0647 \u0628\u06cc\u0627\u0646 \u06a9\u0646\u0645: \u062f\u0631\u0633\u062a \u0627\u0633\u062a\u060c \u0647\u06cc\u0686 \u06a9\u0633 \u0646\u0645\u06cc \u062e\u0648\u0627\u0647\u062f \u0628\u06cc\u0634 \u0627\u0632 \u062d\u062f \u0645\u062f\u0644\u200c\u0647\u0627\u06cc \u0646\u0647\u0627\u06cc\u06cc\u060c \u062f\u0631\u0633\u062a \u0645\u062b\u0644 \u0686\u06cc\u0632\u06cc \u06a9\u0647 \u0647\u06cc\u0686\u200c\u06a9\u0633 \u0646\u0645\u06cc\u200c\u062e\u0648\u0627\u0647\u062f \u06a9\u0645 \u062a\u0646\u0627\u0633\u0628 \u0645\u062f\u0644 \u0647\u0627\u06cc \u067e\u0627\u06cc\u0627\u0646\u06cc \u0645\u062f\u0644 \u0647\u0627\u06cc Overfit \u0639\u0645\u0644\u06a9\u0631\u062f \u0639\u0627\u0644\u06cc \u0631\u0648\u06cc \u062f\u0627\u062f\u0647 \u0647\u0627\u06cc \u0622\u0645\u0648\u0632\u0634\u06cc\u060c \u0627\u0645\u0627 \u0646\u0645\u06cc \u062a\u0648\u0627\u0646\u0646\u062f \u0628\u0647 \u062e\u0648\u0628\u06cc \u0628\u0647 \u0646\u0645\u0648\u0646\u0647 \u0647\u0627\u06cc \u062c\u062f\u06cc\u062f \u062a\u0639\u0645\u06cc\u0645 \u062f\u0647\u0646\u062f. \u0686\u06cc\u0632\u06cc [&hellip;]<\/p>\n","protected":false},"author":3,"featured_media":15094,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1743,620],"tags":[],"class_list":["post-15093","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-python","category-programming"],"acf":[],"_links":{"self":[{"href":"https:\/\/rasanegaar.com\/blog\/wp-json\/wp\/v2\/posts\/15093","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/rasanegaar.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/rasanegaar.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/rasanegaar.com\/blog\/wp-json\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/rasanegaar.com\/blog\/wp-json\/wp\/v2\/comments?post=15093"}],"version-history":[{"count":0,"href":"https:\/\/rasanegaar.com\/blog\/wp-json\/wp\/v2\/posts\/15093\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/rasanegaar.com\/blog\/wp-json\/wp\/v2\/media\/15094"}],"wp:attachment":[{"href":"https:\/\/rasanegaar.com\/blog\/wp-json\/wp\/v2\/media?parent=15093"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/rasanegaar.com\/blog\/wp-json\/wp\/v2\/categories?post=15093"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/rasanegaar.com\/blog\/wp-json\/wp\/v2\/tags?post=15093"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}