hypors

Hypothesis testing for Python, backed by the hypors Rust crate.

Every statistic is computed by the Rust library — these bindings only convert values across the boundary, so Python and Rust always agree.

Install

pip install hypors

Use

from hypors import TailType
from hypors.t import t_test

result = t_test([1.2, 2.3, 1.9, 2.5, 2.8], pop_mean=2.0, tail=TailType.Two, alpha=0.05)

print(result.test_statistic)
print(result.p_value)
print(result.reject_null)
print(result.to_dict())

Any iterable of numbers works — a list, tuple, generator, numpy array, pandas.Series or polars.Series. The package itself depends on none of them:

import numpy as np
import polars as pl

t_test(np.array([1.2, 2.3, 1.9]), 2.0, TailType.Two, 0.05)
t_test(pl.Series([1.2, 2.3, 1.9]), 2.0, TailType.Two, 0.05)

Tests available

TailType and TestResult are also importable from the top level.

Results

Every test returns a TestResult with test_statistic, p_value, confidence_interval, null_hypothesis, alt_hypothesis and reject_null, plus to_dict() for serialising. Attributes are read-only.

Errors surface as ordinary Python exceptions: ValueError for empty or insufficient data, RuntimeError for a computation that could not be completed.

Types

The package ships py.typed and inline stubs, so editors and type checkers see the full API without any extra install:

from hypors import TailType, TestResult
from hypors.t import t_test

result: TestResult = t_test([1.0, 2.0, 3.0], 2.0, TailType.Two, 0.05)
p: float = result.p_value

Any iterable of numbers is accepted wherever data is taken — lists, tuples, generators, numpy arrays, polars series.

Notes

• The Mann-Whitney U p-value matches scipy.stats.mannwhitneyu with method="asymptotic". Its test_statistic is min(U1, U2) while the p-value derives from U1, so compare p-values rather than statistics when checking against scipy.
• Versions track the Rust crate: hypors on PyPI and hypors on crates.io share a version number.

License

MIT