shaded

shaded provides tools for linear spectral feature extraction and dimension reduction, with a focus on fast, interpretable projections for machine learning and signal processing.

Features

• PseudoLda: Fast, approximate Linear Discriminant Analysis (LDA) using class centers.
• PseudoPca: Fast, approximate Principal Component Analysis (PCA) using random hyperplanes.
• PairwiseLda: LDA projections for every pair of classes.
• Chained Spectral Projectors: Compose multiple projection methods in sequence.
• Band Projection Matrix: Utilities for frequency band bucketing and projection.
• Linear Algebra Utilities: Null space, projection, and residue computations.

Installation

pip install shaded

Usage Examples

PseudoLda: Fast LDA-like Projection

from shaded.pseudo_lda import PseudoLda
from sklearn.datasets import load_iris

X, y = load_iris(return_X_y=True)
plda = PseudoLda(n_components=2)
plda.fit(X, y)
X_proj = plda.transform(X)
print(X_proj.shape)  # (n_samples, 2)

PseudoPca: Fast PCA-like Projection

from shaded.pseudo_pca import PseudoPca

ppca = PseudoPca(n_components=2)
ppca.fit(X)
X_proj = ppca.transform(X)
print(X_proj.shape)  # (n_samples, 2)

PairwiseLda: LDA for All Class Pairs

from shaded.pair_wise_lda import PairwiseLda

pwlda = PairwiseLda(n_components=2)
pwlda.fit(X, y)
X_proj = pwlda.transform(X)
print(X_proj.shape)

Chained Spectral Projectors

Chain multiple projections (e.g., PCA followed by LDA):

from shaded.chained_spectral_projector import GeneralProjectionLearner

chain = (
    {'type': 'pca', 'args': {'n_components': 3}},
    {'type': 'lda', 'args': {'n_components': 2}},
)
gpl = GeneralProjectionLearner(chain=chain)
X_proj = gpl.fit_transform(X, y)
print(X_proj.shape)  # (n_samples, 5)

Band Projection Matrix

Create frequency band buckets and projection matrices:

from shaded.band_projection_matrix import make_band_proj_matrix

proj_matrix = make_band_proj_matrix(n_buckets=5, n_freq=20)
print(proj_matrix.shape)  # (5, 20)

API Overview

• PseudoLda: Fast LDA-like projection for well-separated clusters.
• PseudoPca: Fast PCA-like projection using random hyperplanes.
• PairwiseLda: LDA projections for all class pairs.
• GeneralProjectionLearner: Chain and compose projections (PCA, LDA, etc.).
• band_projection_matrix: Create frequency band buckets and projection matrices.
• linalg_utils: Linear algebra helpers (null space, projection, residue).

Testing

Unit tests are provided for core linear algebra and projection utilities. To run tests:

pytest shaded/tests/

License

[Specify your license here]

Let me know if you want to add more advanced examples or further API details!