K-Histogram algorithm in ranger?

[Mamba413]

Would you like to add K-histogram algorithm used in lightgbm (https://github.com/Microsoft/LightGBM/blob/master/docs/Experiments.rst#comparison-experiment) or xgboost? I think this algorithm is an efficient algorithm which balances bias and variance.

[mnwright]

From the link above

bucketing continuous feature(attribute) values into discrete bins

So, the idea is to reduce the number of split points to evaluate, as in randomised splitting. However, the candidate split points are based on histograms instead of sampling them randomly. Are the bins of equal width? Or should they contain the same amount of training observations?

[Mamba413]

Generally, these methods try to keep each bin with the same number of observation. There also exist some advanced methods which try to retain the distribution of variables, such as Weighted Quantile Sketch in the paper "XGBoost: A Scalable Tree Boosting System"(http://www.kdd.org/kdd2016/papers/files/rfp0697-chenAemb.pdf).