Fix hyper model caching, adds data check

Caching method now checks whether input data is the same along with the input parameters when evaluating the probability.

Closes #524 (closed)

changed milestone to %1.0.2

added Bug label

added 1 commit

• 6f279168 - fix flake8

Compare with previous version

marked as a Work In Progress

changed milestone to %1.0.3

added Hyper label

added Testing label

approved this merge request

approved this merge request

unapproved this merge request

removed milestone %1.0.3

unapproved this merge request

Ping @shanika.galaudage

added 139 commits

• 6f279168...da290af1 - 138 commits from branch lscsoft:master
• f02e7a42 - Merge branch 'master' into fix_hyper_model_module

Compare with previous version

added 161 commits

• f02e7a42...4b9c75ba - 158 commits from branch lscsoft:master
• b6162fa7 - Merge branch 'master' into shanika.galaudage/bilby-fix_hyper_model_module
• 9d655607 - Enable method to work with dataframes and dict
• 6d64103e - Update docs

Compare with previous version

marked this merge request as ready

changed milestone to %1.1.4

I tested a version of this locally and got errors with dictionaries of numpy arrays (related SO), the elements of the dictionary need to be tested explicitly.

This would potentially break things downstream for GWPopulation and others using cupy for hyper pe.

I think a dedicated method to test the equality of the data would be worthwhile and make the code a bit cleaner.

Also, for speed, it might be better to test A and then if that fails move on rather than doing the potentially much more expensive call to the second comparison. Very quick benchmarks I did put dictionary comparisons at O(100ns), pandas comparisons O(100 mus), dictionary of numpy arrays O(1ms).

@colm.talbot it is a little opaque to me what types of data are expected. Would it be cleaner to enforce a DataFrame datatype?

I'll push this to the next release as I don't think we'll get these issues fixed soon.

changed milestone to %1.2.0