Loss curve
Also known as: loss plot · training curve
A loss curve plots the model’s loss over training. It’s the fastest way to tell whether a run is healthy.
- Falling smoothly: learning well.
- Flat early: learning rate too low or data issue.
- Spiking / exploding: rate too high or bad data.
- Train drops, validation rises: overfitting — stop.
Habit: before anything else, look at the curve. It tells you more about a training run than any single number can.