pytyche¶
Smarter, calibrated multi-round A/B tests for online experimentation. Pytyche tells you which customer segments drive a treatment’s lift and which don’t respond, discovers those segments from your data, and over a series of rounds proposes redirecting more traffic toward responders while keeping controls everywhere so measurement stays honest.
Install pytyche from source, run your first fit in five minutes. The shortest runnable path from install to a posterior you can interpret.
End-to-end walk-throughs. Your first hurdle BCF fit takes you through install, a synthetic adaptive-enrichment dataset, the canonical fit, and per-segment posterior interpretation.
Goal-oriented recipes for specific tasks: apply a shipped calibration correction, run a power sweep, configure a custom DGP for SBC.
Why the library is built the way it is. Start with overview: what pytyche does, who it’s for, and the design lineage. Then drill into the calibration / hurdle BCF / sequential targeting concept docs.
The full API. Public API for typical users (the curated surface the library commits to); Internal for extenders and contributors who need the full module tree.
Live status of the documentation corpus — polish-state breakdown,
drift list, and the per-doc last-human-review audit.
Project status¶
0.2.1 — published on PyPI. The public API is the curated Reference surface and is stabilizing; internal modules may change without notice. MIT licensed; source on GitLab.