Entire dataset, mini batch and epoch

vision
1

Example:
I have a data set size of 801
60% of my data set is used for training so approx 481 size of my training dataset.

dataloaders = {x: torch.utils.data.DataLoader(image_datasets[x], batch_size=15,shuffle=True, num_workers=15)for x in ['train', 'val']}

Above statement will fetch 15 set(mini batch) of data from the training dataset.
so in order to complete training on the entire data set 481/15 approx= 32
so number of epochs required to complete training over the entire dataset = 32 given be the statement below.
model_ft = train_model(net,criterion,optimizer_ft,exp_lr_scheduler,num_epochs=32)

Am i right, correct me if i am wrong?

2

You will need 32 iterations to complete one epoch. If you use your DataLoader in a for loop, a whole epoch will be completed by giving 32 batches of size 15.

I’m not sure how train_model is defined, but it seems you are confusing iterations/batches with epochs.

3

example

def train_model(model, criterion, optimizer, scheduler, num_epochs=20):
    since = time.time()

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

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

        for phase in ['train','val']:

            if phase == ' trian':
                scheduler.step()
                model.train()

            else:
                model.eval()
            running_loss = 0.0
            running_corrects = 0


            for inputs,labels in dataloaders[phase]:
                inputs = inputs.to(device)
                labels = labels.to(device)

                optimizer.zero_grad()

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

                    if phase == 'train':
                        loss.backward()
                        optimizer.step()

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

            epoch_loss = running_loss/dataset_sizes[phase]
            epoch_acc = running_corrects.double()/dataset_sizes[phase]
            print('{} Loss: {:.4f} Acc: {:.4f}'.format(
                    phase, epoch_loss, epoch_acc))

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

        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
4

Based on this code you are indeed using 32 epochs, i.e. 32 iterations over your full dataset.
In each epoch the loop over your DataLoader (for inputs, labels in dataloaders[phase]) will give you ~32 batches of size 15.

5 
for inputs,labels in dataloaders[phase]:
                inputs = inputs.to(device)
                labels = labels.to(device)

                optimizer.zero_grad()

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

                    if phase == 'train':
                        loss.backward()
                        optimizer.step()

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

fetches 15 images at once for 32 iterations over the entire training dataset to complete one epoch?

6

Yes, more or less. It will actually give you 33 batches. 32 batches of size 15 and 1 batch of size 1, as the number of samples is not dividable by 15 without a remainder and you didn’t specify drop_last=True in your DataLoader.