Solvers
FK — Friedman's Discriminative Whitening
FK whitening with eigenvalue flooring and diagonal shrinkage.
How It Works
- FK Projection — computes discriminative whitening with eigenvalue flooring for numerical stability.
- Ridge Regression — fits a linear classifier in the projected space using closed-form solve.
FK is the default adaptation method. It works with continual learning (update()).
Parameters
| Parameter | Default | Description |
|---|---|---|
| shrinkage_alpha | 0.5 | Diagonal shrinkage intensity |
| regularization | 1e-4 | Ridge regularization strength |
from vyntri import Vyntri
from vyntri.data import split
s = split("./my_dataset", train=0.7, test=0.2, seed=42)
# FK is the default adaptation method
model = Vyntri()
model.fit(train=s.train, val=s.val)
# Or explicitly set adaptation="fk"
model = Vyntri(adaptation="fk", shrinkage_alpha=0.5)
model.fit(train=s.train, val=s.val)