algorithm - More Stastistical Way to Average N Predictions -
i've run randomforestregressor (scikit ensemble) on n loops, each time changing random seed , therefore changing train test split. way i've n set of predictions (m predictions each loop). i've captured following data:
r2_score train dataset on n loops (1 value per loop) r2_score test dataset on n loops (1 value per loop)
currently, average out predictions which:
r2_score_train > median(r2_score_train) && r2_score_test > median(r2_score_test)
i want know better way make out final predictions, using r2_score both train , test dataset. 1 way thinking it, give more weight loop r2_score_train ~ r2_score_test (i.e. difference between them smallest) , lowering weights value high.
i using python this. scikit-learn. wanted know if there inbuilt function or 3rd party library?