classification - Feature importances, discretization and criterion in decision trees -


i'm working numerical features , want use decision tree classifier in sklearn find feature importances.

so, if select entropy criterion splitting, information gain used measure of impurity split data. guess, equivalent fayyad & irani binary discretization.

in end, classifier returns attribute called "feature importances".

the feature importances. higher, more important feature. importance of feature computed (normalized) total reduction of criterion brought feature. known gini importance [r195].

my question is, though i'm using information gain find best split, "feature importances" return value of gini importance measured in split found entropy criterion?

yes! there iterative method calculate gini importance of different splitting points , hits termination criteria (minimum description length), optimal splitting points returned. can find more info toy example here: http://clear-lines.com/blog/post/discretizing-a-continuous-variable-using-entropy.aspx


Comments

Popular posts from this blog

angularjs - ADAL JS Angular- WebAPI add a new role claim to the token -

php - CakePHP HttpSockets send array of paramms -

node.js - Using Node without global install -