Implement Brier Score/Brier Skill Score

[samland1116]

• adds brier score and brier skill score to probabilistic metrics

  • requires the user define a quantile threshold to evaluate BS (fractional, between 0-1. default=0.75)

• updates probabilistic metrics routine to:

  • to avoid failure when calculating skill score if encountering divide by 0 in _calculate_metric_skill_score(), updated to use try_divide(), returns null. Provides warning for which configs fail.
  • Updates routine in _calculate_metric_skill_score() to calculate skill score to allow for metrics.query() calls with multiple probabilistic metrics. Was encountering an error when trying to calculate CRPSS/BSS in one call. Corrected to work as intended by updating _post_process_metric_results.
    • since we were assigning the results of _calculate_metric_skill_score() directly to the self.sdf, the method was removing all columns aside from the columns specified in group_by, the column corresponding to the metric in the input arg, and the calculated skill score column. Once the method gets called again for the next metric in include_metrics -- the metric column it is targeting was removed in the first iteration and fails.

[samlamont]

In test_ensemble_metrics() I was getting slightly different results that were causing the CRPS assertions to fail.

I was also not able to run both crsp and brier score in the same call (adding both to include_metrics) with the reference forecasts specified, which may have been a memory error on my system.

Do you see these as well? Seems like my local WSL might have some memory issues

[samland1116]

Merging -- might revise the size of the new test data soon to avoid unnecessary memory issues in a later PR