Decomposition Base

base

CLASS	DESCRIPTION
Decomposition	Base class for all decomposition and dimensionality reduction methods.

Classes

DecompositionResult

Bases: NamedTuple

Named tuple to hold results of decomposition methods.

ATTRIBUTE	DESCRIPTION
eigenvalues	Tensor of eigenvalues from the decomposition. TYPE: Tensor
eigenvectors	Tensor of eigenvectors from the decomposition. TYPE: Tensor
kernel	Kernel (similarity) matrix used in the decomposition, e.g., affinity, covariance. TYPE: Tensor

Decomposition(kernel_fn: Kernel | str | None = None, kernel_kwargs: dict | None = None, distance_fn: PairwiseDistance | str | None = None, distance_kwargs: dict | None = None, score_fn: Callable | None = None)

Bases: Estimator

Base class for all decomposition and dimensionality reduction methods.

Provides unified interface for spectral decomposition methods (PCA, Diffusion Maps, Laplacian Eigenmaps, etc.) with support for custom kernels, distance metrics, and scoring functions.

PARAMETER	DESCRIPTION
kernel_fn	Kernel function for computing similarity matrices. Options: Kernel instance: Custom kernel object "gaussian": Gaussian (RBF) kernel "t": Student's t-distribution kernel "adaptive_gaussian": Adaptive Gaussian kernel with learnable bandwidth None: Defaults to "gaussian" TYPE: Kernel | str | None, optional, by default None DEFAULT: None
kernel_kwargs	Parameters for kernel initialization (e.g., {"bw_method": "median"}). TYPE: dict | None, optional, by default None DEFAULT: None
distance_fn	Distance metric for computing pairwise distances. Options: PairwiseDistance instance: Custom distance object "euclidean": Euclidean (L2) distance "covariance": Covariance-based distance "mahalanobis": Mahalanobis distance None: Defaults to "euclidean" TYPE: PairwiseDistance | str | None, optional, by default None DEFAULT: None
distance_kwargs	Parameters for distance initialization. TYPE: dict | None, optional, by default None DEFAULT: None
n_components	Number of components to extract. If None, determined automatically by the specific decomposition method. TYPE: int | None, optional, by default None
score_fn	Custom scoring function with signature (X, X_reconstructed) -> float. If None, uses method-specific default scoring. TYPE: Callable | None, optional, by default None DEFAULT: None

ATTRIBUTE	DESCRIPTION
kernel_fn	Validated kernel function instance. TYPE: Kernel
distance_fn	Validated distance function instance. TYPE: PairwiseDistance
score_fn	Validated scoring function. TYPE: Callable | None
in_features	Number of input features. TYPE: int | None
out_features	Number of output features or components. TYPE: int | None

See Also

spectre.core.Estimator : Parent class with fit/predict/score interface spectre.kernel.Kernel : Base class for kernel functions spectre.pairwise_distance.PairwiseDistance : Base class for distance metrics

Source code in spectre/decomposition/base.py

def __init__(
    self,
    kernel_fn: Kernel | str | None = None,
    kernel_kwargs: dict | None = None,
    distance_fn: PairwiseDistance | str | None = None,
    distance_kwargs: dict | None = None,
    score_fn: Callable | None = None,
) -> None:
    super().__init__()

    if kernel_fn is not None:
        self.kernel_fn = initialize_kernel_fn(kernel_fn, kernel_kwargs or {})

    if distance_fn is not None:
        self.distance_fn = initialize_distance_fn(
            distance_fn, distance_kwargs or {}
        )

    if score_fn is not None and not callable(score_fn):
        raise TypeError(
            f"Expected `score_fn` to be callable or None, but got {type(score_fn)}."
        )
    self.score_fn = score_fn

Functions