I think it is important to check that the predicted CDFs are non-decreasing. Without this requirement, the continuous_ranked_probability_score can produce incorrect results.
>>> import numpy as np
>>> from solarforecastarbiter.metrics.probabilistic import continuous_ranked_probability_score
>>> continuous_ranked_probability_score(np.array([0.5]), np.array([[0.5, -1e6, 0.5]]), np.array([[0, 0.5, 1]]))
-9950.004974999989
I think it is important to check that the predicted CDFs are non-decreasing. Without this requirement, the
continuous_ranked_probability_scorecan produce incorrect results.