Use weighted min_bin_size and max_bin_size when sample_weight provided

[YC-1412]

Fix #323

Hi Guillermo,

Thanks for developing this amazing tool. I found it extremely useful and easy to use. I hope this PR resolves an existing issue with sample_weight.

1. Added a property self._n_samples_weighted to calculate weighted sample size - if sample_weight is provided it will be the weighted sample size (sum(sample_weight), if not, if will be the same as self._n_samples.

   • self._n_samples_weighted is used when calculating the min_bin_size and max_bin_size in _fit_optimizer().

2. When sample_weight is provided, use min_weight_fraction_leaf instead of min_samples_leaf for DecisionTreeClassifier. It was suggested in the sklearn documentation:

   Note that min_samples_split considers samples directly and independent of sample_weight, if provided (e.g. a node with m weighted samples is still treated as having exactly m samples). Consider min_weight_fraction_leaf or min_impurity_decrease if accounting for sample weights is required at splits.

   Class balancing can be done by sampling an equal number of samples from each class, or preferably by normalizing the sum of the sample weights (sample_weight) for each class to the same value. Also note that weight-based pre-pruning criteria, such as min_weight_fraction_leaf, will then be less biased toward dominant classes than criteria that are not aware of the sample weights, like min_samples_leaf.

   If the samples are weighted, it will be easier to optimize the tree structure using weight-based pre-pruning criterion such as min_weight_fraction_leaf, which ensure that leaf nodes contain at least a fraction of the overall sum of the sample weights.

I used the test code in #323 to test the fix:

Before:

After:

Todo

• I think it is also worth mentioning in the documentation that min_bin_size and max_bin_size include the mix of missing and special groups. I can add that if you think the change makes sense.

Thanks again for creating and maintaining this tool!

[guillermo-navas-palencia]

Great PR. Thanks a lot for your contribution. Regarding the clarification about the min/max_bin_size, if think it is a good idea.