Some of the important arguments are:
--prep_data: when used, the actual data will be downloaded and extracted to the data folder
--model: specifies the pre-trained model to be used in the unified architecture, e.g., resnet.
--eval: should be used when evaluating the models for drawing ROC curves and computing AUC values.
--epoch: specifies which checkpoint should be used for model evaluation
--lr: the learning rate for training the model (should also be specified for model evaluation to find the correct checkpoint folder).
--freezed: if specified, the parameters of the pre-trained model will be freezed.
Data preparation: python3 main.py --prep_data
Training: python3 main.py --model resnet --lr 2e-6 --freezed
Evluation: python3 main.py --eval --model resnet --freezed --lr 2e-6 --epoch 20
Some of the packages needed to be installed include:
torchsummary
Default parameters like the learning rate are in the params.json file, but could be changed if specified in the program arguments (to be explained mode later).