2017.08.16 review of PSI about woe non-rejection inference effect

Continue thinking Woe IV psi refusal to infer this is a problem that I find difficult to control, but today we finally have some of my own conclusions and future directions

Variable each group is good or bad than the same, the distribution change, woe with increase and decrease, the impact of less psi no eggs, but also should pay attention to the variables of each group of the trend of the bad or not to infer that I currently do not need, but such a low pass rate, should later encounter this problem can try to use bad Rate in the overall calculation of the woe or with bad rate code, this can eliminate the modeling set to the overall distribution changes to woe impact