Gradient
Descent Visualizer

Slope Walker turns gradient descent into a playful interactive lesson. Watch a stick figure optimize across multiple landscapes with live metrics, adjustable controls, and optional momentum.

The Algorithm

x_{n+1} = x_n - \alpha \nabla f(x_n)

Switch terrains and tune the controls to compare vanilla GD, momentum, local minima, and overshoot.

Function Landscape

Compare the terrain

Current Function

f(x) = x^2

\nabla f(x) = 2x

OptimizerVanilla GD

Step Size (Learning Rate)0.85

Step size scales the gradient into an actual move.

Start Position (x₀)4.0

Iteration

Loss f(x)

16.00

Gradient

8.00

Best Loss

16.00

Run State

READY

What to look for

Pick a function landscape, adjust the sliders, and run to see how gradient descent behaves across different terrains.

Gradient: 8.00. Learning rate 0.85 scales the next step size.

Trail points: 1 · Last step: 0.00

Contact