[FIXED] Can sklearn.preprocessing.KBinsDiscretizer with strategy='quantile' drop the duplicated bins?

Issue

I used sklearn.preprocessing.KBinsDiscretizer(n_bins=10, encode='ordinal') to discretize my continuous feature.

The strategy is 'quantile', by defalut. But my data distribution is actually not uniformly, like 70% of rows is 0.

Then I got KBinsDiscretizer.bins_edges=[0.,0.,0.,0.,0.,0.,0.,256.,602., 1306., 18464.].

There're many duplicate bins. So, is there a method to drop the duplicates in KBinsDiscretizer's bins?

KBinsDiscretizer calculates the quantile of input. If the most samples of input are zero, the 10-quantiles will have multiple zeros. The result I expected is a discretizer with unique bins. For the example I mentioned, is [0.,256.,602., 1306., 18464.].

Solution

That will not be possible. Set strategy='uniform' to achieve your goal.

Answered By - Franco Piccolo