pySHT

pySHT is a Python toolbox for literature-traceable statistical hypothesis testing, with particular emphasis on high-dimensional, joint-parameter, and structured-domain problems.

Documentation: https://kisungyou.com/pysht/

The distribution and import name is pysht. The project is an independently validated successor to the R package SHT; it does not treat legacy numerical output as a correctness oracle.

Status: version 0.1.0 is the initial public release. The scientific surface is correctness-gated, but APIs may still evolve before version 1.0.

Install the release from PyPI:

python -m pip install pysht

Statistical scope

pySHT's public API currently contains 51 canonical Python functions covering 52 of the 54 public statistical routine identities in SHT 0.1.9. Two algebraically identical R entry points share one Python implementation. The withheld identities are mean2.2014CLX, whose practical-size Gumbel calibration failed, and cov1.2012Fisher, whose former audit is not reproducible through the public path. The public functions are organized by scientific question rather than re-exported from the top-level package:

• pysht.mean: univariate, multivariate, high-dimensional, randomized, sparse, and multi-group mean tests;
• pysht.variance: one-, two-, and multi-sample variance tests;
• pysht.covariance: one-, two-, and multi-sample covariance tests;
• pysht.mean_variance and pysht.mean_covariance: joint-parameter tests;
• pysht.equaldist: equality-of-distributions tests;
• pysht.normality and pysht.uniformity: goodness-of-fit tests; and
• pysht.simplex: tests for compositional data on the probability simplex.

The API catalog follows SHT's original [0]--[10] category order and records every public name.

For example:

from pysht.mean import ttest_1samp

result = ttest_1samp([2.1, 2.4, 1.9, 2.2, 2.5], popmean=2.0)
print(result)

Results are immutable Python objects, but their text display follows the human-readable style of R's htest objects: method title, data, statistic, p-value, alternative, degrees of freedom, confidence interval, estimates, and calibration details when applicable.

Procedure-specific results can additionally report resampling diagnostics, numerical metadata, or component log Bayes factors. The user guide explains how to read and use this output.

Correctness policy

Every shipped method must have a formula ledger, an appropriate literal reference or trusted independent comparator, validation evidence, invariance checks, numerical stress tests, and documented assumptions. Known differences from SHT are recorded in the validation documentation.

The current suite also builds the documentation with warnings treated as errors, checks strict typing and formatting, and verifies installed stable-ABI wheels across Python 3.12--3.14 in CI.

License

MIT. See the license and third-party notices.