Common questions

Does AShrink require R?

No. Fitting and prediction use NumPy and SciPy. R is needed only to regenerate the comparison fixtures described in reference checks.

Which fitting function should I use?

Use a summary-input function for effect estimates, counts or successes. Use a regression function when outcomes and predictors are available directly. The API table lists the supported methods and parameter scales.

Are credible intervals confidence intervals?

The reported intervals are posterior credible intervals conditional on fitted mixture weights and the selected component grid. Regression also uses an independent-factor variational approximation. These calculations do not by themselves establish repeated-sample coverage. See numerical methods and limits.

What do null and sign probabilities mean?

lfdr reports posterior point-null probability. lfsr reports the smaller posterior probability of a nonpositive or nonnegative effect, including any zero atom in both events. Count models compare the rate or probability with the specified null value. The API guide explains the related ranked summaries and prior-dependent limits.

Does it implement all of ashr, mashr or mr.ash?

AShrink supports the methods listed in the API guide. The reference checks compare selected matched cases. Multivariate fitting uses supplied covariance patterns; it does not learn those patterns or residual correlations from the effects.

What should I check when a fit does not converge?

Inspect the diagnostics and objective history before using the result. For scalar models, check the component grid and numerical integration warnings. For regression, inspect the variational objective and initialization. A small numerical residual or a converged objective does not establish statistical calibration. The numerical guide explains these distinctions.

Are saved results anonymous?

No. Summary-input results retain the input estimates or counts. Regression JSON retains training outcomes directly or through fitted values and residuals. See tables and saved results.

Where are the detailed arguments and formulas?

Use the online reference manual or download the reference PDF. The development guide explains how to preview this searchable site.