`shrinkage.PathFit`

Shrinkage for Python · python · 0.2.0

Bases: typing.Generic[shrinkage.path.FitT]

PathFit(
    _native: shrinkage._shrinkage._ModelPath,
    _wrap: collections.abc.Callable[[_shrinkage._ModelFit], shrinkage.path.FitT]
)

Owned path in fitting order, with a complete result at every point.

fits[i] exposes every single-fit diagnostic and prediction method. penalties[i] retains both fusion strengths and all edge or rank weights. Finite iteration-limit points remain in the path and warm-start the next point. A fitting error raises without returning a partial path. Results retain no training arrays. Returned arrays are independent copies.

`family`

Source: packages/python/src/shrinkage/path.py

family(self) -> Family

Gaussian means or binomial logits, consistent across all points.

`solver`

Source: packages/python/src/shrinkage/path.py

solver(self) -> typing.Literal["hybrid", "proximal_gradient"]

Explicit numerical method used for every point, including empty paths.

`n_features`

Source: packages/python/src/shrinkage/path.py

n_features(self) -> int

Training column count, retained even for an empty path.

`fits`

Source: packages/python/src/shrinkage/path.py

fits(self) -> tuple[shrinkage.path.FitT, ...]

Per-point owned fits, aligned with penalties and parameter rows.

`penalties`

Source: packages/python/src/shrinkage/path.py

penalties(self) -> tuple[ClusteredLasso | ChainFusedLasso | Slope, ...]

Complete immutable penalty settings for each fitted point.

`coefficients`

Source: packages/python/src/shrinkage/path.py

coefficients(self) -> FloatArray

Original-scale float64 coefficients, shape (n_points, n_features).

`intercepts`

Source: packages/python/src/shrinkage/path.py

intercepts(self) -> FloatArray

Original-scale float64 intercepts, shape (n_points,).

`lambdas`

Source: packages/python/src/shrinkage/path.py

lambdas(self) -> FloatArray | None

Automatic scalar multipliers, shape (n_points,); explicit: None.

`maximum_penalty`

Source: packages/python/src/shrinkage/path.py

maximum_penalty(self) -> float | None

True zero threshold: finite or inf for automatic paths; else None.

`lambda_start`

Source: packages/python/src/shrinkage/path.py

lambda_start(self) -> float | None

Used automatic anchor, including an override; explicit paths: None.

`termination`

Source: packages/python/src/shrinkage/path.py

termination(self) -> shrinkage.PathTermination

Path completion or early stopping, independent of point convergence.

`null_deviance`

Source: packages/python/src/shrinkage/path.py

null_deviance(self) -> float | None

Averaged automatic-path null-model deviance; explicit paths: None.

`deviances`

Source: packages/python/src/shrinkage/path.py

deviances(self) -> FloatArray | None

Automatic fitted deviances, shape (n_points,); explicit: None.

`predict`

Source: packages/python/src/shrinkage/path.py

predict(x: object, *, copy: bool = True) -> FloatArray

Predict float64 means or logits, shape (n_samples, n_points).

Columns follow path order, including duplicates. Empty batches and empty paths retain both axes. The dense input and copying rules match single fits. Inputs are copied once before Rust releases the GIL, and must remain unchanged until return. Prediction never refits transforms.