Implementation

These pages describe how barry works internally: how it stores support sets and how it computes exact likelihoods. They are written in Markdown with GitHub-compatible math ($...$ and fenced math blocks), so they render the same on GitHub and on this site.

See Likelihood, gradient, and Hessian for the derivation of the exact log-likelihood and its derivatives, their implementation, and an example.

Efficient memory usage

In pooled-data models, barry avoids computing and storing the same support more than once. It keeps track of which arrays (for example, networks) share a support, and stores each unique support only once.

Adding a page

1. Add a Markdown file under docs/impl/.
2. List it under Implementation in the nav section of mkdocs.yml.
3. Preview it with make preview.