How to calculate WOE (weight of evidence) for single class bins? While calculating WOE using formula ln(%of bads/%of goods), I encounter this situation, where one of the bins has 100% goods.  But it will result in formula error for ln(0).  How to calculate WOE in such case? Is it appropriate to set the WOE to 0, in such a case. 
For example. I have a dataset, which has target variable drinkers (1,0) and predictor variable smoker (1,0). Now say all smokers are drinkers.  So that is a strong evidence, even though the formula for WOE leads to ln(0).
How should I calculate WOE in such a case?
 A: Natural answer would be that your sample is too small and you should try to enlarge your sample by providing additional experiments or observations. 
However that is sometimes not possible, in that case I would suggest to increase your sample artifficially. For your example that would mean, add an artifficial observation into the sample (smoker, who is not drinker) and assign a "reasonably small" weight to the observation (for example if you have 100 smokers/drinkers, possible weight of the artifficial smoker/non-drinker would be maybe 0.1 or 0.01). The question what is "reasonably small" weight I can not unfortunately answer, but I suggest to set it according to some theoretical knowledge of the examined topic.
From here on you can use the classical logarithmic formula for WoE.
A: One simple practice would be to set WOE for "pure bins" (or single class bins) to some high value, as they provide you the classification with much higher quality than an average bin. Now one can allocate the same high value to all such bins, or different (e.g. how many points lie in that bin? maybe something proportional to that).
