Getting started#

This guide takes you from installation to a result you can inspect, print, interpret, and use programmatically. If you are unsure which procedure matches your question and sampling design, start with the test chooser.

Installation#

pySHT requires Python 3.12 or newer. Install the latest stable release, currently 0.1.0, from PyPI:

python -m pip install pysht

This website follows the master branch, including unreleased numerical corrections beyond the draft 0.5.0rc1 candidate. Installing the stable release does not provide all methods and corrections described here. See the changelog for the distinction.

To use the development version, clone the repository and install from its current source checkout. A C++17 compiler and CMake are also required:

git clone https://github.com/kisungyou/pySHT.git
cd pySHT
python -m pip install -e ".[test]"

Record the checkout’s commit hash alongside the package version when reproducing a development result; the candidate version string alone does not identify these unreleased changes.

Run a first test#

from pysht.mean import ttest_1samp

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

print(result)

The printed summary contains the method, data label, statistic, degrees of freedom, p-value, alternative hypothesis, calibration, confidence interval, and estimate when those quantities apply. pySHT uses the same display contract across its scientific modules; the exact fields depend on the procedure.

Before interpreting the p-value, confirm that the procedure, alternative, and assumptions match the scientific question. See the data and assumptions guide.

Result objects#

Every test returns an immutable StatisticalTestResult subtype rather than printing as a side effect. A frequentist result supports both programmatic access and a human-readable display:

from pysht.mean import ttest_1samp

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

print(result)  # R htest-style summary
result.statistic  # test statistic
result.pvalue  # p-value
result.alternative  # alternative hypothesis
result.method  # method name

Additional fields, such as degrees of freedom, confidence intervals, estimates, named diagnostics, and calibration metadata, are present when the procedure computes them. Exact and Monte Carlo results also record simulation counts and tail uncertainty. The two Lee–You–Lin procedures return maximum and component log Bayes factors with no pvalue; they do not fabricate a frequentist decision. Result instances are frozen data classes, so attempts to mutate a field raise an error.

Use repr(result) for a compact developer-facing representation and print(result) for the statistical report.

Interpret the result#

Treat the printed output as a compact report, not as a decision made by the library. A p-value is calculated under the stated null model and calibration; it is not the probability that the null hypothesis is true. pySHT therefore does not add a reject field. Set the significance level before analysis and report an estimate and confidence interval when the procedure provides them.

The test-results guide explains each printed section and the common structured fields. For exact signatures and parameter defaults, use the category pages in the API reference, beginning with univariate mean tests for the function above.

Preserve the analysis#

Record the data provenance, preprocessing, fully qualified function name, all non-default arguments, software versions, and returned result. If a procedure uses resampling, random projections, random subspaces, or numerical optimization, also record its calibration mode, budget, random-number policy, and convergence controls. See reproducible inference for a complete checklist.

Where to go next#