Gradient Descent Playground

Click anywhere on the contour plot to set a starting point, then watch gradient descent roll downhill. Play with the learning rate and see the optimizers behave very differently.

Loss Function

Hyperparameters

Learning rate 0.050

Momentum 0.00

Max steps 200

This run

Step0

Position—

Loss—

|Gradient|—

Click on the contour plot to set a starting point. Try high learning rates — you'll see oscillation or divergence. Try multiple starting points on the multi-minima function — you'll see that where you start decides which minimum you reach.

⛰️ Gradient Descent Playground