sdpython / mlinsights

Extends scikit-learn with new models, transformers, metrics, plotting.
http://www.xavierdupre.fr/app/mlinsights/helpsphinx/index.html
MIT License
69 stars 13 forks source link

Faster split criterion for Piecewise Tree Regressor #129

Open craciuncg opened 8 months ago

craciuncg commented 8 months ago

As far as I have noticed, for splitting a node, it is required to compute the beta coefficients for the linear regression model for the node samples, calculate the MSE score and then decide the split based on that(note that this is my high level understanding of the process).

I believe that a faster criterion for splitting nodes could be to use the Pearson Correlation Coefficient(its absolute value), since the model that needs to be fitted is linear.