Alexander Held
Alexander Held
### Summary As far as I can see, the current `pyhf.Workspace.prune` approach is to prune anything that matches any of its arguments. This means e.g. that ```python prune(modifier_types=["normsys"], modifiers=["Modeling"]) ```...
For models with some physically motivated bounds (like $\mu\geq0$) the $\tilde{q}_\mu$ test statistic is defined in https://arxiv.org/abs/1007.1727 at equation 16. The use of this test statistics in `pyhf` seems to...
### Summary It would be great to add analytic staterror parameter optimization, as available in Combine: https://cms-analysis.github.io/HiggsAnalysis-CombinedLimit/part2/bin-wise-stats/#analytic-minimisation. Technically this would imply hiding such parameters from the minimizer and doing the...
### Summary I frequently am dealing with RooFit workspaces stored in .root format[^1], which still contain the HistFactory measurement object. That object can be used to `->PrintXML()` to then arrive...
### Summary There is currently no validation happening for the shape of the observations per channel when reading a workspace and extracting data from it. Making that part of the...
### Summary When adding a modifier twice to a sample, only one instance is used. For `normfactor` modifiers this could in principle be something a user would want to do...
### Summary It might be useful to catch failed fits in `pyhf` and investigate them for possible reasons for the crash. Negative yields, which result in NaN likelihoods, are presumably...
### Summary The return types for numbers that are read from the workspace specification depends on how the number is written down in the workspace. I believe it would be...
### Summary A batch size of != 1 can cause issues in places where `batch_size=1` is presumably assumed. It might be useful to catch this (not necessarily a bug, could...
### Summary I was trying to find whether `return_result_obj` is documented on the website. This is a bit tricky to find, as it can be passed to e.g. `pyhf.infer.mle.fit` but...