Step 5: The Training Loop

Epochs, batches, and loss curves

1 ExplorePlay below

›

2 ReadUnderstand

›

3 BuildHands-on lab

›

4 CompareSolution

›

💡 ReflectThink deeper

The training loop

Training repeats: forward pass → compute loss → backward pass (adjust weights) → repeat. Each full pass through all data is one epoch.

Epoch

Epoch 1

Loss

—

Train Acc

—

Val Acc

—

Think Deeper

Try this:

Watch the loss curve. Does it ever go UP during training? Why?

Yes — individual batches can increase loss (noise from random batch selection). But the trend should be downward. If the overall trend goes up, the learning rate is too high.

Cybersecurity tie-in: Each epoch is one pass through all your training logs. With 100,000 firewall entries and batch_size=32, each epoch makes ~3,125 weight updates. More epochs = more chances to learn, but also more risk of memorising noise.

← Previous ← → to navigate Next →