Function training

MPP_Code/training/execute_classification_explicit.py:180–256

Args: mod : It is the mod we have to train criterion : Loss function, her we have Cross entropy loss optimizer : object of torch.optim class call: call is the COMMAND to be excuted to run th

(mod, criterion, optimizer, call, train_loader, valid_loader, fold, e=500, patience=5, report=False)

def training(mod, criterion, optimizer, call, train_loader, valid_loader, fold, e=500, patience=5, report=False):
    """Args:
            mod :
                It is the mod we have to train
            criterion :
                Loss function, her we have Cross entropy loss
            optimizer :
              object of torch.optim class
            call:
                call is the COMMAND to be excuted to run the forward method of the model
                it changed as per the modality and other possible input
            train_loader:
                It is a instance of train dataloader
            valid_loader:
                It is a instance of validation dataloader, it is given as a input to evaluation class
            fold:
                5 FOLD {0,1,2,3,4}
            e:
                maximum epoch
            patience:
                how many epoch to wait after the early stopping condition in satisfied
            report:
                It True then the classification report for the validation set is printed, it is given as a input to evaluation class
            save:
                If true then best model for each fold is saved

    """

    print('-'*100)
    train_losses = [0]
    valid_losses = [0]
    max_f1 = 0
    patience_flag = 1
    best_epooch = 0
    print(fold, e, patience)

    while e > 0:
        total_loss = []
        seed()
        for batch in train_loader:
            uText = batch[0].float().to(device)
            cText = batch[1].float().to(device)
            uAudio = batch[2].float().to(device)
            cAudio = batch[3].float().to(device)
            uVideo = batch[4].float().to(device)
            cVideo = batch[5].float().to(device)
            speaker = batch[6].float().to(device)
            y_true = batch[7].long().to(device)
            del batch
            output = eval(call)
            loss = criterion(output, y_true)
            del uText, cText, uAudio, cAudio, uVideo, cVideo, speaker
            optimizer.zero_grad()
            total_loss.append(loss.detach().item())
            loss.backward()
            optimizer.step()
        with torch.no_grad():
            valid_f1, valid_loss = evaluation(