logistic¶

Logistic-regression projection axis.

Use method="logistic" to fit a linear classifier and project neural data onto its learned decision direction(s).

When to choose this method¶

Choose logistic reduction when predictive decoding performance is the primary goal.

Optimizes class separation with a discriminative objective.
Supports regularization, which can improve robustness in high-dimensional neural feature spaces.
Useful for linking low-dimensional projections to classifier decision boundaries.

Compared with coding direction, logistic can fit more flexible discriminative axes, but those axes may be less directly interpretable as condition means.

ds = da.ephys.reduce(
    method="logistic",
    labels="choice",
)

Notes¶

labels is required.
regularization and cv can be used to control model fitting.