Skip to content

CHANGELOG.md

Latest commit

 

History

History
325 lines (264 loc) · 13.5 KB

CHANGELOG.md

File metadata and controls

325 lines (264 loc) · 13.5 KB
Raw

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

0.6.0 - 2026-08-17

Added

  • New statistics API (anofox-statistics 0.4.1 → 0.4.2):
    • one_way_anova — one-way ANOVA with Fisher and Welch variants; returns F-statistic, degrees of freedom, p-value, sums of squares, and η² effect size.
    • two_way_anova — two-way ANOVA with A×B interaction term; partitions variance into main effects (A, B) and interaction (A×B).
    • repeated_measures_anova — repeated-measures ANOVA with Mauchly's sphericity test and Greenhouse-Geisser / Huynh-Feldt ε corrections.
    • energy_distance_nd — multivariate energy distance for two multivariate samples (generalization of energy_distance to arbitrary dimension).
    • icc — intraclass correlation with real matrix-input contract; replaced the previous all-NaN stub (see Breaking Changes).
  • New regression model classes (anofox-regression 0.5.4 → 0.5.13):
    • GLM family: Gamma (log link, positive data), GLMM (generalized linear mixed model with random intercepts), PSpline (penalized B-spline smoother).
    • Robust / sparse linear: TheilSen (median-of-slopes, ~29% breakdown point), RANSAC (random sample consensus), BayesianRidge (evidence-maximisation hyperparameter tuning), ARD (automatic relevance determination with per-feature precision), LARS (least angle regression with full path), PassiveAggressive (online PA/PA-I/PA-II regression).
    • Online learning: MomentAccumulator — accumulates sufficient statistics (XᵀX, Xᵀy) from streaming batches; pass to OLS.fit_from_accumulator or Ridge.fit_from_accumulator.
    • HC-robust inference: Ridge.hc_inference(X, y, hc_type) — HC0–HC3 sandwich standard errors for Ridge fits.
    • GLM diagnostic expressions for Gamma GLM fits: gamma_dispersion_deviance, gamma_dispersion_pearson, gamma_pearson_chi_squared, gamma_standardized_pearson_residuals, gamma_standardized_deviance_residuals.

Changed

  • Updated anofox-statistics dependency 0.4.1 → 0.4.2.
  • Updated anofox-regression dependency 0.5.4 → 0.5.13.
  • Column-pivot correctness fix (inherited from anofox-regression 0.5.13): OLS, WLS, and NNLS models trained on differently-scaled designs (e.g. one predictor in units of thousands while another is in units of ones) now return correct coefficients. The previous release could pivot to the wrong column when building the design matrix, causing coefficients to be assigned to the wrong predictor.

Breaking Changes

  • icc new contract: the expression now accepts one column per rater (ps.icc("rater1", "rater2", "rater3", icc_type="icc3")) and computes real ICC values validated against R's irr::icc(). The old single-column stub always returned NaN and is removed. Update call sites to pass all rater columns explicitly.

Known Limitations

  • WLS.hc_inference raises NotImplementedError — heteroscedasticity-consistent inference is not yet implemented for WLS; use OLS.hc_inference or Ridge.hc_inference as alternatives.
  • icc matrix-input contract change: code passing only a single column to ps.icc() must be updated to pass one column per rater.

0.5.0 - 2026-05-28

Added

  • Hybrid crate (Rust + Python) — polars-statistics now builds as both a cdylib (Python plugin) and an rlib (Rust dependency). Every Polars expression has a public <name>_fit Rust entry point in polars_statistics::expressions. New python Cargo feature gates the pyo3 / numpy / pymodels surface; downstream Rust crates use default-features = false. (#13)
  • New regression model wrappers:
    • Huber — M-estimator robust to outliers; class + huber() expression. (#14)
    • LogisticRegression — sklearn-style API with predict_proba, decision_function, score, penalty="l2", C kwarg. Distinct from the existing Logistic wrapper. (#14)
    • PLS — Partial Least Squares with transform() for the latent space; class + pls() expression. (#19)
  • Penalized IRLS lambda_ kwarg on the GLM PyClass models — PyLogistic, PyPoisson, PyNegativeBinomial, PyTweedie, PyProbit, PyCloglog (the expression layer already supported it). (#15)
  • ALM expression parity with PyALM — all 25 distributions reachable from ps.alm(...), plus loss, link, role_trim, and extra_parameter kwargs. (#16)
  • Summary / predict completeness — added the matching expressions for families that previously only had a base fit (#18):
    • quantile_summary, quantile_predict
    • isotonic_predict
    • lm_dynamic_predict
  • Diagnostics toolkit (#17 + #27):
    • Multicollinearity: vif, generalized_vif, high_vif_predictors
    • OLS residual battery: standardized_residuals, studentized_residuals, externally_studentized_residuals, residual_outliers
    • GLM residuals (logistic + Poisson): *_pearson_residuals, *_deviance_residuals, *_working_residuals for each family
    • Influence / leverage: leverage, cooks_distance, dffits, influential_cooks, influential_dffits, high_leverage_points
    • Goodness of fit: pearson_chi_squared_logistic, pearson_chi_squared_poisson
  • Documentation:
    • Sweep of the README and mkdocs site to cover every v0.5.0 addition.
    • New "Use from Rust" section + examples/rust_wls.rs walking through the rlib path.

Changed

  • Updated anofox-regression dependency to v0.5.4 (introduces HuberRegressor and the sklearn-style LogisticRegression)
  • Updated anofox-statistics dependency to v0.4.1
  • Python __version__ caught up from 0.3.0 → 0.5.0 (was lagging two minor versions behind the wheel metadata)

Backwards compatibility

Additive. PyLogistic is unchanged — the new sklearn-style LogisticRegression is a separate class. The alm() expression's input contract grew but all existing keyword-only callers continue to work because the new kwargs default to None / "likelihood".

0.4.0 - 2026-01-17

Added

  • Robust Regression:

    • quantile - Quantile regression (median and arbitrary quantiles)
    • isotonic - Isotonic (monotonic) regression using PAVA algorithm
    • Quantile and Isotonic model classes for scikit-learn-style API

  • Regression Diagnostics:

    • condition_number - Detect multicollinearity via condition number analysis
    • check_binary_separation - Detect complete/quasi-complete separation in logistic regression
    • check_count_sparsity - Detect sparsity issues in Poisson/count regression

  • Conda-forge Support:

    • Added conda-forge recipe for conda install polars-statistics
    • CI workflow for conda package building and testing

  • Documentation:

    • Reorganized API documentation into modular structure
    • Added detailed descriptions for all statistical tests
    • New docs for TOST equivalence tests, correlation, categorical tests, and forecast comparison

Changed

  • Updated anofox-regression dependency to v0.5.1
  • Updated pyo3 to v0.27
  • Updated faer to v0.23.2
  • Improved cross-platform test compatibility for GLM functions

0.3.0 - 2024-12-22

Added

  • TOST Equivalence Tests (10 functions):

    • tost_t_test_one_sample - One-sample equivalence test
    • tost_t_test_two_sample - Two-sample equivalence test
    • tost_t_test_paired - Paired samples equivalence test
    • tost_correlation - Correlation equivalence test (Pearson/Spearman)
    • tost_prop_one - One-proportion equivalence test
    • tost_prop_two - Two-proportion equivalence test
    • tost_wilcoxon_paired - Non-parametric paired equivalence test
    • tost_wilcoxon_two_sample - Non-parametric two-sample equivalence test
    • tost_bootstrap - Bootstrap-based equivalence test
    • tost_yuen - Trimmed means equivalence test

  • Correlation Tests (7 functions):

    • pearson - Pearson correlation with confidence intervals
    • spearman - Spearman rank correlation
    • kendall - Kendall's tau (variants a, b, c)
    • distance_cor - Distance correlation (detects nonlinear relationships)
    • partial_cor - Partial correlation controlling for covariates
    • semi_partial_cor - Semi-partial (part) correlation
    • icc - Intraclass correlation coefficient

  • Categorical Tests (13 functions):

    • binom_test - Exact binomial test
    • prop_test_one - One-sample proportion test
    • prop_test_two - Two-sample proportion test
    • chisq_test - Chi-square test for independence
    • chisq_goodness_of_fit - Chi-square goodness of fit test
    • g_test - G-test (likelihood ratio test)
    • fisher_exact - Fisher's exact test for 2x2 tables
    • mcnemar_test - McNemar's test for paired proportions
    • mcnemar_exact - McNemar's exact test
    • cohen_kappa - Cohen's Kappa for inter-rater agreement
    • cramers_v - Cramer's V for association strength
    • phi_coefficient - Phi coefficient for 2x2 tables
    • contingency_coef - Contingency coefficient (Pearson's C)

  • Comprehensive test suite for all new statistical tests

Changed

  • Updated anofox-statistics dependency to v0.4.0
  • Updated anofox-regression dependency to v0.4.0

0.2.0 - 2024-12-15

Added

  • R-style Formula Syntax:

    • ols_formula, ridge_formula, elastic_net_formula, etc.
    • Support for polynomial terms: poly(x, 2)
    • Support for interactions: x1 * x2 expands to x1 + x2 + x1:x2
    • Support for explicit transforms: I(x^2)

  • Summary Functions for tidy coefficient output:

    • ols_summary, ridge_summary, elastic_net_summary
    • wls_summary, rls_summary, bls_summary
    • logistic_summary, poisson_summary, negative_binomial_summary
    • tweedie_summary, probit_summary, cloglog_summary, alm_summary
    • Returns: term, estimate, std_error, statistic, p_value

  • Prediction Functions with intervals:

    • ols_predict, ridge_predict, elastic_net_predict, etc.
    • Support for confidence and prediction intervals
    • Configurable confidence level

  • Additional Regression Models:

    • WLS - Weighted Least Squares
    • RLS - Recursive Least Squares
    • BLS - Bounded Least Squares
    • NNLS - Non-negative Least Squares

  • GLM Models:

    • Logistic - Logistic regression
    • Poisson - Poisson regression
    • NegativeBinomial - Negative binomial regression
    • Tweedie - Tweedie regression
    • Probit - Probit regression
    • Cloglog - Complementary log-log regression

  • ALM (Augmented Linear Model):

    • Support for 24+ distributions (normal, laplace, cauchy, student-t, etc.)
    • Robust regression alternatives

  • Dynamic Models:

    • LmDynamic - Dynamic linear model with forgetting factor
    • lm_dynamic expression for rolling regression

  • Demand Classification:

    • Aid - Automatic Item-level Demand classification
    • aid and aid_anomalies expressions
    • Based on Kolassa (2025) methodology

  • Model Classes for Statistical Tests:

    • TTestInd, TTestPaired - t-test classes
    • BrownForsythe, YuenTest - variance/robust tests
    • MannWhitneyU, WilcoxonSignedRank, KruskalWallis, BrunnerMunzel
    • ShapiroWilk, DAgostino - normality tests

  • Forecast Comparison Tests:

    • diebold_mariano - Diebold-Mariano test
    • clark_west - Clark-West test
    • spa_test - Superior Predictive Ability test
    • model_confidence_set - Model Confidence Set
    • mspe_adjusted - MSPE-adjusted test
    • permutation_t_test - Permutation t-test

  • Modern Distribution Tests:

    • energy_distance - Energy distance test
    • mmd_test - Maximum Mean Discrepancy test

  • Bootstrap Methods:

    • StationaryBootstrap - Stationary bootstrap for time series
    • CircularBlockBootstrap - Circular block bootstrap

Changed

  • Renamed regress-rs dependency to anofox-regression
  • Improved documentation with API reference

0.1.0 - 2024-11-01

Added

  • Initial release of polars-statistics

  • Core Statistical Tests:

    • ttest_ind - Independent samples t-test
    • ttest_paired - Paired samples t-test
    • brown_forsythe - Brown-Forsythe test for variance equality
    • yuen_test - Yuen's test for trimmed means

  • Non-parametric Tests:

    • mann_whitney_u - Mann-Whitney U test
    • wilcoxon_signed_rank - Wilcoxon signed-rank test
    • kruskal_wallis - Kruskal-Wallis H test
    • brunner_munzel - Brunner-Munzel test

  • Distributional Tests:

    • shapiro_wilk - Shapiro-Wilk normality test
    • dagostino - D'Agostino-Pearson normality test

  • Regression Models:

    • OLS - Ordinary Least Squares
    • Ridge - Ridge regression
    • ElasticNet - Elastic Net regression
    • Expression API: ols, ridge, elastic_net

  • Polars Integration:

    • Full support for group_by aggregations
    • Full support for over window functions
    • Lazy evaluation support
    • Struct output for all statistical results

  • Performance:

    • Rust-powered with zero-copy data transfer
    • SIMD-optimized linear algebra via faer
    • Automatic parallelization for group operations