DDW Interface - Logistic Cost Landscape

Gradient Descent Setup

Select model case

Learning rate: 0.100 Learning rate input

If you use z-normalization, you can generally use a larger learning rate than with unscaled features.

Iterations: 80 Start parameter 1: -4.00 Start parameter 2: 4.00 Automatic run speed (ms): 90

The height represents the cost function J(theta). Lower points correspond to parameter values that better fit class labels.

Each trajectory point is one gradient-descent update: horizontal movement = parameter changes, vertical movement = cost-function change.

Landscape View

Surface Contours Trajectory Current Point

Surface clipping: none

Lowest point in shown region: --

Stability: normal

Drag to rotate · Scroll to zoom · Hover to inspect · Use Fit / +/- / Fullscreen to resize

Gradient Descent Step 0 / 0