cplx-gmm v0.2.0

Added

• Store fitted FFT-domain parameters for covariance_type="circulant":
  • means_fft_
  • covariances_fft_
• Store fitted 2D FFT-domain parameters for covariance_type="block-circulant":
  • means_fft2_
  • covariances_fft2_
• Add copy-returning properties:
  • means_fft
  • covariances_fft
  • means_fft2
  • covariances_fft2

Changed

• Keep original-domain means_ and covariances_ as the primary public fitted representation for circulant and block-circulant models.
• Update public fit_predict to go through the same covariance-structure preprocessing as fit.
• Clear stale Fourier-domain fitted attributes when refitting with a different covariance type.
• Document that warm_start=True is currently supported for full, diag, and spherical covariance types, while structured covariance support is planned.

Tests

• Add coverage for retained Fourier-domain fitted parameters.
• Add fit_predict tests for circulant and block-circulant models.
• Add warm-start behavior tests for supported and structured covariance types.

cplx-gmm v0.1.0

Initial PyPI release of cplx-gmm.

Highlights:

• scikit-learn-style estimator for complex-valued Gaussian mixture models
• support for full, diagonal, spherical, circulant, block-circulant, Toeplitz, and block-Toeplitz covariance structures
• optional zero-mean component constraint
• sampling, scoring, prediction, and posterior probability APIs
• modern Python packaging with pyproject.toml and uv
• test suite covering covariance structures, EM behavior, sampling, validation, real-valued compatibility, and doubled real-valued likelihood equivalence

Install:

pip install cplx-gmm