In order to run these scripts, the data files should be in the same folder. Also, the packages used are torch, torchvision, numpy, matplotlib and pandas.
cnn_3layers.py
- Architecture: (Conv layer(K=5, S=1, P=0) -> ReLU -> 2x2 MaxPooling -> Dropout)x3 -> (Linear activ. -> dropout)x2)
- Best validation accuracy: 87%
best_model.py
- Architecture: (Conv layer(K=4, S=1, P=1) -> ReLU -> 2x2 MaxPooling -> Dropout)x2 -> (Conv layer(K=3, S=1, P=1) -> ReLU -> 2x2 MaxPooling -> Dropout)x2 -> (Linear activ. -> dropout)x2)
- Best validation accuracy: 92.7%
- Best test accuracy: 93.4%
script_to_submit.py
- Task: Uses the best_model.py architecture to output a csv file to submit on the Kaggle competition
- Uses the CSV package