Support for toys in pyhf on master

9 views

Skip to first unread message

pyhf announcements

unread,

Oct 28, 2020, 10:32:03 PM10/28/20

to pyhf announcements

Dear pyhf users,

Some of you are probably aware that a major point on our roadmap to pyhf v0.6.0 was support of pseudo-experiments (or "toys" in our HEP-centric parlance) for setting limits in the regime where the Wald approximation is no longer valid. With the merging to PR #790 to master we have begun to work on bringing this as a full feature to our upcoming minor release and have additionally added a pedagogical example notebook to our documentation to help you familiarize yourself with the upcoming API.

We should stress that we only support "official" releases of pyhf distributed on PyPI, conda-forge, and Docker Hub and do not support branches --- even the master branch. However, we know that some of you have been brave early adopters and are already using pyhf toys for various studies and we are thrilled to have such enthusiastic users. The development branch for the toys has been deleted, so for all those who were using it and still want to be testing this pre-release-candidate feature we recommend following the instructions we have on our website for installing from TestPyPI (where we do test deployments of each PR). The TL;DR is you need pyhf v0.5.4.dev3 or later so make sure you do a `pip list | grep pyhf` afterwards to check.

This feature is still being made "release ready" so please be aware of the following known issues:

* Issue #892 (strange issues with toys)

* Issue #1146 (x0 violates bounds due to known bug in SciPy)

- use minuit optimizer or try a different backend while pyhf team iterates with SciPy team

* Issue #1059 (weakref warnings)

- harmless and fixed in Python 3.8+

If you have any questions or if you find any pain points or issues please contact us through the recommended channels. We're happy to help!