Epoch

An epoch is one complete pass through the entire training dataset. Training a model typically requires many epochs — the model sees the same data multiple times to learn patterns.

Example

num_epochs = 10

for epoch in range(num_epochs):
    for batch in train_loader:
        inputs, labels = batch
        outputs = model(inputs.cuda())
        loss = criterion(outputs, labels.cuda())
        loss.backward()
        optimizer.step()
        optimizer.zero_grad()

    print(f"Epoch {epoch + 1}/{num_epochs} complete")

Key Points

• 1 epoch = the model has seen every training sample once
• More epochs generally improve performance up to a point, then overfitting occurs
• Common to train for 3-100+ epochs depending on the task

See Also

• Batch Size
• Backpropagation