COMP5328Assignment2

Use shell script to run the code. For training a model, use script such as:

python main.py --name faishion_05_symmetric \
    --dataset_name FashionMNIST0.5.npz  --lr 0.0003 \
    --input_nc 1 --loss_name ce --num_classes 3;

For testing a model, use script such as:

python main.py --name faishion_06_symmetric \
    --dataset_name FashionMNIST0.6.npz --is_testing \
    --input_nc 1 --loss_name ce --num_classes 3 \
    --pretrained_model_path checkpoints/faishion_06_symmetric/faishion_06_symmetric_20_model.pth

The meanings of availble parameters are:

1. --name name of the experiment. It decides where to store samples and models
2. --dataset_name the name of the dataset
3. --dataset_root the root folder of the dataset
4. --checkpoints_dir the root folder of checkpoints
5. --input_nc # of input image channels
6. --nf # of gen filters in first conv layer
7. --num_classes # of classes
8. --is_testing if it is testing phase
9. --epoch training epochs
10. --batch_size training batch size
11. --loss_name name of the loss function used to train the model bootstrap|ce|mpe|symmetric.
12. --sigma sigma value in the loss function.
13. --lr initial learning rate for adam
14. --beta1 momentum term of adam
15. --beta2 momentum term of adam
16. --val_split_rate the split rate of data for validating
17. --optimizer the optimizer method sgd|adam
18. --print_loss If loss should be printed during the training phase
19. --save_model If model need to be saved after training
20. --num_trained_model How many model do we need to train
21. --trans_matrix transition matrix if known
22. --pretrained_model_path the path of the pretrained model
23. --pretrained_model_path_format the path format of the pretrained model
24. --log_file_path the path of the log file