WeightedVarianceSplitFinder: WeightedVarianceSplitFinder

In jatherrien/largeRCRF: Large Random Competing Risks Forests

Description

This split finder is used in regression random forests. When a split is made, this finder computes the sample variance in each group (divided by n, not n-1); it then minimizes the sum of these variances, each of them weighted by their sample size divided by the total sample size of that node.

Usage

WeightedVarianceSplitFinder()

Value

A split finder object to be used in train; not useful on its own.

Note

There are other split finders that are used in regression random forests that are not included in this package. This package is oriented toward the competing risks side of survival analysis; the regression options are provided as an example of how extensible the back-end Java package is. If you are interested in using this package for regression (or other uses), feel free to write your own components. It's not too hard to write these components; the WeightedVarianceSplitFinder Java class is quite short; most of the code is to reuse calculations from previous considered splits. I (the author) am also willing to assist if you have any questions.

References

https://kogalur.github.io/randomForestSRC/theory.html#section8.3

Examples

splitFinder <- WeightedVarianceSplitFinder()
# You would then use it in train()

jatherrien/largeRCRF documentation built on Nov. 15, 2019, 7:16 a.m.