training/test split and confusion matrix

[Xiang-Gu]

Hi, thank you for sharing your code here. I found your code very helpful to me to learn something about FER. But there are two things I think you can improve your code on:
1). The FER2013 dataset (fer2013.csv file) actually comes with training/test labels but in your code, you just randomly select 10% of all the images as your test images. Wouldn't it be nicer to split the dataset as indicated in the csv file?

2). When I run the confmatrix.py to generate the confusion matrix, I get a plot with absolute numbers (like 50, 102, etc) in the cells of the matrix rather than the ratios. Can you modify your code so that it actually computes the ratios and plot it the ratios?

Thanks,
Xiang