Playground: Adaptive Ridge Forecasting

Adaptive HP Playground

Pre-fit ridge weights at the searched optimum for each (series, horizon). Move the L or α slider to refit and see how the prediction reacts.

Test MSE

—

lower is better

Baseline MSE

—

global baseline (L=720, α=10⁻⁵)

vs global baseline
—
global mean+std scaler, L=720, α=10⁻⁵

Fit time

—

idle

Lookback (L)

Long lookbacks slow each fit; α-only changes are cached and fast.

Regularization (log α)

α = 1.0

Other HPs (scaler, augmentation, local-norm fraction) are pinned to the searched optimum for the current horizon.