Getting poor validation accuracy if i split the data using sklearn train_test_split.

Source: Deep Learning on Medium


Getting poor validation accuracy if i split the data using sklearn train_test_split. it works fine if i split the data using subsetting as showen in this article.

Here is the complete code

Please help me why my random spliting is not working. I am doing this because i want to do some image augumentation on train data set that i don’t want to apply on validation data.