trainer.py

import time
import copy
import torch


def train_model(model, dataloaders, criterion, optimizer, num_epochs=25):
    """
    Function to train a model for a number of epochs.

    Args:
        model: The neural network model to train
        dataloaders: A dictionary of data loaders of the shape {'train': train_loader, 'val': val_loader}. 
        criterion: The loss function. Takes model outputs and labels as input and produces a loss value
        optimizer: The optimizer object to use in order to train the model
        num_epochs: The number of epochs to train the model for

    Returns:
        (model, val_acc_history): The trained model and the history of validation accuracy
        
    """
    since = time.time()

    val_acc_history = []

    best_model_wts = copy.deepcopy(model.state_dict())
    best_acc = 0.0

    for epoch in range(1, num_epochs+1):
        print('Epoch {}/{}'.format(epoch, num_epochs))
        print('-' * 10)

        # Each epoch has a training and validation phase
        for phase in ['train', 'val']:
            if phase == 'train':
                model.train()  # Set model to training mode
            else:
                model.eval()   # Set model to evaluate mode

            running_loss = 0.0
            running_corrects = 0
            all_preds = []
            all_labels = []
            # Iterate over data.
            for inputs, labels in dataloaders[phase]:
                inputs = inputs.to(device)
                labels = labels.to(device)

                # zero the parameter gradients
                optimizer.zero_grad()

                # forward
                # track history if only in train
                with torch.set_grad_enabled(phase == 'train'):
                    # Get model outputs and calculate loss
                    outputs = model(inputs)
                    loss = criterion(outputs, labels)
                    # Get model predictions
                    _, preds = torch.max(outputs, 1)

                    # backward + optimize only if in training phase
                    if phase == 'train':
                        loss.backward()
                        optimizer.step()

                # statistics
                running_loss += loss.item() * inputs.size(0)
                running_corrects += torch.sum(preds == labels.data)
                all_preds.append(preds)
                all_labels.append(labels)

            epoch_loss = running_loss / len(dataloaders[phase].sampler.indices)
            epoch_acc = running_corrects.double() / len(dataloaders[phase].sampler.indices)
            all_labels = torch.cat(all_labels, 0)
            all_preds = torch.cat(all_preds, 0)
            epoch_weighted_acc = accuracy_score(all_labels.cpu().numpy(), all_preds.cpu().numpy(), sample_weight=compute_sample_weight(class_weights, all_labels.cpu().numpy()))
            

            print('{} Loss: {:.4f} - Acc: {:.4f} - Weighted Acc: {:.4f}'.format(phase, epoch_loss, epoch_acc, epoch_weighted_acc))

            # deep copy the model
            if phase == 'val' and epoch_weighted_acc > best_acc:
                best_acc = epoch_weighted_acc
                best_model_wts = copy.deepcopy(model.state_dict())
            if phase == 'val':
                val_acc_history.append(epoch_weighted_acc)

        print()

    time_elapsed = time.time() - since
    print('Training complete in {:.0f}m {:.0f}s'.format(time_elapsed // 60, time_elapsed % 60))
    print('Best val Acc: {:4f}'.format(best_acc))

    # load best model weights
    model.load_state_dict(best_model_wts)
    return model, val_acc_history