xenonnt / gofevaluation Goto Github PK

We would like to add the calculation of the p-value for the GoF measures:

• 1D KS 2-sample
• 1D KS 1-sample
• nD binned poisson LLH
• nD binned Chi2
• 2-sample point cloud GOF test

When we used arrays converted from 34 column dataframes as inputs of ge.BinnedPoissonChi2GOF, it returned weird results instead of complaining about undesired input dimension.

We learned that we should open an issue to make things complain about undesired input dimension.

Since we will often be interested in just finding poor fits rather than finding exact p-values (so p<0.001 is an adequate statement, we do not need to know it is actually 1.34e-4) I think it might be useful to throw only a warning here and return 1 or 0 depending.

Since it seems that the ge.get_equiprobable_binning method was taken out, the plotting is way less flexible:
What solutions are there to plot the events + the equiprobable bins and how can one save the plot produced with ge.BinnedPoissonChi2GOF.bin_equiprobable and plot=True?
There are a few workarounds but none very elegant that come to mind for now

Function in GOFevaluation provides a way to calculate binning of equiprobable_histogram, but the reference_sample is equally weighted. I am wondering if it is necessary to add an equal probability binning function with a weighted data sample.

We would like to add CI for pytest and flake 8. This should be run for every PR which is filed.

We might want to have a unified class which has as members the available GOFs that we want to look at. Like this we would only have one object in the users hand an give some wrapper functions. @JosephJHowlett

As discussed with @kdund we would like to add some GOF measures:

• 1D KS 2-sample
• 1D KS 1-sample
• 1D AD 2-sample
• nD binned poisson LLH
• nD binned Chi2
• 2-sample point cloud GOF test