Example has problem of data leakage

[nguyenbim]

In your example, you do the iv filter, woe binning for the whole data before splitting train, test. If I understand correctly. this action causes data leakage because the trainset created after splitting will contain woe data of the whole dataset.

[ShichenXie]

You are right. You can split your real dataset at the very beginning of model training. In this example, the binning result might be unstable when based on such a small train dataset only.