pipeline_guide.txt

# get args from cmd
# args <= shadow_dataset_name, shadow_model, attack_model, [eval/test]_dataset

config:
    - shadow_model
    - attack_model
    - target_model
    - shadow_dataset
    - shadow_dataset_num_classes
    - split_ratio
    - train_attack_epochs,
    - train_shadow_epochs,

########
---> incorporated this step into "train_shadow_models" function <---
# split shadow dataset into : train/test
# def get_dataloaders(dataset_name, dataset_path) -> shadow_[train/test]_dataloaders
########

# train shadow models
# def train_shadow_models(shadow_model: str, shadow_dataset: str, target_model: str, split_raito: float) -> /saved_shadow_models/{model}_{datset}.pth

# create attack dataset
# def create_attack_dataset(saved_shadow_model_path, train_dataset_path, test_dataset_path) -> attack_dataset_{shadow_model}.p

# train attack model
# def train_attack_models(model, attack_dataset_path) -> /saved_attack_models/{model}_{dataset}.pth

# run membership attack
# def membership_inference(inference_dataset_path, attack_model_path, [optional] (str) out_suffix) -> task{i}_{dataset}.npy ,