Error reduction in GBM

Question

I would like to know how ErrorReduction is calculated in pretty.gbm.tree function: https://github.com/gbm-developers/gbm3/blob/62c8dafd87b16fe1d2079cdd5058169f1f08967b/R/pretty-gbm-tree.r#L32

This post helped me a lot, but did not quite answer my question. Understanding tree structure in R gbm package

Here is an output of pretty.gbm.tree function for the first tree: output table

How this value of 19167.524 was calculated?

Thanks

Answer 1

The ErrorReduction is calculated as following:

number of samples in a current node * mse - number of samples in the left child node * mse - number of samples in the right child node *mse