The model is predicting the majority class (class 2) for nearly all inputs

[wchong5227]

The output I got looks really weird, here is a part of the output I got from one of the training epoch:
Train Epoch: 65 [0/316 (0%)] Loss: 1.467374
Train Epoch: 65 [128/316 (41%)] Loss: 1.446983
Train Epoch: 65 [256/316 (81%)] Loss: 1.493753
Training Confusion Matrix for Epoch 65:
[[ 0 0 7917 0 0]
[ 0 0 2748 0 0]
[ 0 0 17009 0 0]
[ 0 0 5407 0 0]
[ 0 0 7271 0 0]]
epoch : 65
loss : 1.481035739937915
accuracy : 0.4215535996835443
f1 : 0.11835105639593411
val_loss : 1.4931234866380692
val_accuracy : 0.3857421875
val_f1 : 0.1358819961691086
I carefully checked the python files, and I checked that I didn't do any modification. I am really confused why I have this imbalance issue.

[wchong5227]

I solve this issue by creating a matching environment for the gpu to run.