DM test with score instead of forecasts

[simon-hirsch]

Hi,

would you accept a PR for a DM-test version that accepts scores_1, scores_2 as inputs instead of the original prices and point forecasts? That way, the DM test could as well be used for probabilistic forecasts if one passes two (n, ) or (n, d) arrays of CRPS (or Energy Scores, Variogram Scores, ...). Currently the scope is somewhat limited to the point forecasting case.

Cheers, Simon

[enertel]

Check out https://github.com/ogrnz/feval - you can pass either point or probabilistic loss scores and it will do a more general case of the DM test, it even has built in procedures for identifying model confidence sets. Highly recommend :)