Wanted to suggest parallelizing the oob importance calculation in order to speed it up, since the importances can be calculated independently for each feature.
In my use case, I saw a >8x reduction in runtime after parallelizing and would be happy to contribute the code needed to implement this.
Wanted to suggest parallelizing the oob importance calculation in order to speed it up, since the importances can be calculated independently for each feature.
In my use case, I saw a >8x reduction in runtime after parallelizing and would be happy to contribute the code needed to implement this.