AShrink
Adaptive shrinkage uses information shared across estimates to learn a prior distribution and compute posterior summaries. AShrink provides this approach in Python for supplied effect estimates and selected regression models. It also supports Poisson rates, binomial probabilities, and effects measured across conditions. NumPy and SciPy provide the numerical calculations; R is not a runtime dependency.
Start with installation, then follow the worked tutorial. For outcomes and predictors, use the regression guide.
Choose an analysis
| Available input | Fitting function |
|---|---|
| Effect estimates and standard errors | fit |
| Counts and exposure times | fit_poisson |
| Successes and trial counts | fit_binomial |
| Effect vectors and error covariances | fit_multivariate |
| Continuous outcomes and predictors | fit_regression |
| Binary/binomial outcomes and predictors | fit_logistic_regression |
| Counts and predictors | fit_poisson_regression |
| Grouped continuous outcomes and predictors | fit_mixed_regression |
The API guide explains fitting, posterior summaries, tables, saved results and plots. The reference manual (PDF) gives the detailed arguments and methods.
Posterior intervals condition on fitted mixture weights. Regression also uses a variational approximation. Read the numerical limits and reference checks before interpreting uncertainty. Agreement with selected reference cases does not establish interval coverage or full coverage of the R packages.
Methods and attribution
Scalar adaptive shrinkage follows Stephens (2017). Gaussian regression follows Kim et al. (2024). Multivariate fitting uses supplied covariance patterns, following the fixed-pattern part of Urbut et al. (2019). Cite the method used and record the AShrink version or commit.
AShrink is an independent implementation distributed under GPL-3.0-or-later. See the repository notice and license for attribution and terms.
Citation
See the package paper and BibTeX citation.
Acknowledgements
Computations were performed on the Arseven Computing Cluster, Department of Statistics, Texas A&M University.
This work received no grant funding.