epochs

Learn · Module 03 · Neural networks · Lesson 6 of 7

Reading a loss curve

A loss curve is a training run's heartbeat. Five shapes tell you nearly everything about what's going wrong.

4 minute read

In short
  • Always plot training and validation loss together.
  • Training down while validation goes up means overfitting; keep the checkpoint with the best validation loss.
  • Jumpy or NaN loss usually means the learning rate is too high.

This lesson is part of Pro

The first two lessons of every module are free. Pro opens every lesson and every project.

See Pro plansFrom ₹119 a month. No auto-renewal.