Gradient descent visualizer

Loss function

Function

Range

Colour scale

Log spreads the low-loss valley across more of the ramp, which is what makes Rosenbrock readable.

Contrast (γ)1.20

Grid

Contours12

Mode

Type

Learning rate0.050

Momentum β / β₁0.900

Adam β₂0.999

Gradient noise σ0.000

These apply from the very next step, so you can retune a run while it moves.

Or click anywhere on the landscape. A new start clears the path and the optimizer's memory.

Press Play to let the optimizer roll downhill, or click the landscape to choose where it starts.

Gradient check: pending

Step0

x0.00

y0.00

f(x, y)0.00

|∇f|0.00

Log scale Smooth

Zoom1.0×