Quantile K-nearest neighbors (KNeighborsQuantileRegressor) SampleExtraTreesQuantileRegressor: an approximation, that is much faster than the main implementation.ExtraTreesQuantileRegressor: the main implementation.RandomForestMaximumRegressor: mathematically equivalent to the main implementation but much faster.SampleRandomForestQuantileRegressor: an approximation, that is much faster than the main implementation.RandomForestQuantileRegressor: the main implementation.They include an example that for quantile regression forests in exactly the same template as used for Gradient Boosting Quantile Regression in sklearn for comparability.

This is obviously possible with the implemented models here, but this requires the use of a single quantile during prediction, thus losing the speed advantage described above.įor guidance see docs (through the link in the badge). As can be seen in the example in the documentation: with certain data characteristics different quantiles might require different parameter optimisation for optimal performance. Note that accuracy of doing this depends on the data.

The advantage of this (over for example Gradient Boosting Quantile Regression) is that several quantiles can be predicted at once without the need for retraining the model, which overall leads to a significantly faster workflow. The quantile information is only used in the prediction phase. The models implemented here share the trait that they are trained in exactly the same way as their non-quantile counterpart. This means that practically the only dependency is sklearn and all its functionality is applicable to the here provided models without code changes. This module provides quantile machine learning models for python, in a plug-and-play fashion in the sklearn environment.

Conda install sklearn