REFACTO: in split setting, remove checking NaNs and irrelevant aggregation to avoid triggering unwanted warnings

[Valentin-Laurent]

Currently, during calibration, the same logic is used in the split setting and in the cross setting.

Specifically, at some point we call check_nan_in_aposteriori_prediction and aggregate_all in both settings.

It works in the split setting because check_nan_in_aposteriori_prediction does basically nothing except checking NaNs, and aggregate_all simply flattens the prediction matrix to a prediction array, from shape (n_samples, 1) to shape (n_samples,).

However, calling those 2 functions brings 2 issues:

1. check_nan_in_aposteriori_prediction will always trigger a warning because by definition the train samples are not used for calibration.
2. aggregate_all also triggers warning in the split setting. Moreover, aggregating is not needed anyways in the split setting during calibration, and the dependency on agg_function could be removed entirely in further refactoring

In this PR, we check if we are in a split setting, and if yes simplify the code by simply flattening the array. It is not an ideal solution because we add an extra condition to the existing logic, but it fixes the first issue in a pragmatic way, and prepares the code for further refactoring.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 100.00%. Comparing base (614293e) to head (aebfe7d).
Report is 835 commits behind head on master.

Additional details and impacted files

@@             Coverage Diff             @@
##            master      #586     +/-   ##
===========================================
  Coverage   100.00%   100.00%             
===========================================
  Files           39        61     +22     
  Lines         4616      6003   +1387     
  Branches       487       352    -135     
===========================================
+ Hits          4616      6003   +1387     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[vincentblot28]

All good to me, it improves the readability of the code !