[x] Fixes a bug in predict_many.py regarding assumptions about the .canvas attribute (part of the work towards issue #100 of having more data structure flexibility
[x] Makes a TODO list in elm.model_selection.scoring regarding improvements needed in scoring when we address issue #99 . Those issues are summarized for now in a TODO in the code in case we have confusion building notebooks before #99 is addressed. That TODO mentions: the elm.model_selection.scoring functions needs refactor and docs explanation. They are brittle and difficult to use:
Unclear how to pass in a custom scorer:
Do I need to call make_scorer() myself on that function
If I need to call make_scorer(), do I use the elm.model_selection.scoring.make_scorer above or the make_scorer from sklearn.metrics?
Should a scoring function expect the model to be passed into it, and if so, do should I expect always a Pipeline instance from the elm stack or the Pipeline's _estimator attribute that is typically an sklearn model like LinearRegression()
Should my scoring function, whether custom or standard like r2_score, expect that X and y are xarray data structures or numpy ones that have been already prepared for use with the final sklearn estimator
predict_many.pyregarding assumptions about the.canvasattribute (part of the work towards issue #100 of having more data structure flexibilityelm.model_selection.scoringregarding improvements needed in scoring when we address issue #99 . Those issues are summarized for now in a TODO in the code in case we have confusion building notebooks before #99 is addressed. That TODO mentions: theelm.model_selection.scoringfunctions needs refactor and docs explanation. They are brittle and difficult to use:make_scorer()myself on that functionmake_scorer(), do I use theelm.model_selection.scoring.make_scorerabove or themake_scorerfromsklearn.metrics?Pipelineinstance from theelmstack or thePipeline's_estimatorattribute that is typically ansklearnmodel likeLinearRegression()scoringfunction, whether custom or standard liker2_score, expect thatXandyarexarraydata structures ornumpyones that have been already prepared for use with the final sklearn estimator