不同scale方法对结果的影响

standardize
lgbm score 0.9310618224253633
lr score 0.8508692055869721
rf score 0.9074903045815534
et score 0.8914293518321599

none
lgbm score 0.930806442906943
lr score 0.7266447624392611
rf score 0.9075442389649911
et score 0.8914293518321599

robust (0.25, 0.75)
lgbm score 0.9309728221789764
lr score 0.8308836857131522
rf score 0.9074126275758883
et score 0.8914293518321599

robust (0.1, 0.9)
lgbm score 0.9309728221789764
lr score 0.8396153729253992
rf score 0.9075545086335028
et score 0.8914293518321599

robust (0.01, 0.99)
lgbm score 0.9309728221789764
lr score 0.8447463547252905
rf score 0.9075758674155785
et score 0.8914293518321599

quantile 1000
lgbm score 0.9308608987554141
lr score 0.8457393518542955
rf score 0.9074221948629267
et score 0.8916683318073919

quantile 2000
lgbm score 0.9308780326065063
lr score 0.8457344351839823
rf score 0.9073322581079085
et score 0.891797027249166

quantile 100
lgbm score 0.9309025521052118
lr score 0.8458876607666055
rf score 0.9071914625489348
et score 0.8914761453372853

quantile 10
lgbm score 0.9309761106013288
lr score 0.8475597756431813
rf score 0.9072265816226018
et score 0.8891021278838294

minmax
lgbm score 0.9310023754115744
lr score 0.8358462236651645
rf score 0.9073546917464813
et score 0.8914293518321599

normalize
lgbm score 0.9108568189832683
lr score 0.7518646631795626
rf score 0.8906031383766424
et score 0.8796279383171144

power
lgbm score 0.9309224954821973
lr score 0.8450210923028015
rf score 0.9022971407369846
et score 0.8864163594776462