JOSS submission release

Features

• Two-stage covariance optimization pipeline

  • Stage 1: Tensor basis optimization with L×D parameters
  • Stage 2: Outer loop refinement with L scalar parameters per round
  • Levenberg-Marquardt optimization with adaptive damping

• Divergence-free basis functions

  • Gaussian RBF-based tensor basis construction
  • Volume-preserving transformations guaranteeing entropy conservation
  • Configurable σ parameter (optimal: σ=4.0 for typical 2D uniform distributions)

• High-level APIs

  • DataFrameTransformer for pandas DataFrame transformations
  • transform_csv() one-liner for CSV file processing
  • VectorSampler for generating uniform/Gaussian test distributions

• Hydra configuration support

  • Typed dataclass configs with IDE autocompletion
  • Parameter sweep configurations for σ optimization
  • Configurable stage iterations and tolerances

• Visualization tools

  • Distribution plots at each optimization round
  • Optimization history tracking (determinant, entropy, gap)
  • Sigma sweep summary plots

Documentation

• Comprehensive algorithm flowcharts in README
• Full optimization example with outer loop refinement
• JOSS paper submission

Dependencies

• numpy, scipy, pandas, matplotlib
• hydra-core, omegaconf (for configuration)
• tqdm (for progress bars)