Please implement the Ranger random forest package

[marco_zara]

Hello,

the randomforest package implementation in Alteryx works fine for smaller datasets but becomes very slow for large datasets with many features.

There is the opensource Ranger package https://arxiv.org/pdf/1508.04409.pdf that could help on this.

Along with XGBoost/LightGMB/Catboost it would be an extremely welcome addition to the predictive package!

[Atabarezz]

To train (400 K rows X 1900 variables). Scoring is always lightning fast no problem there...

[marco_zara]

1900 variables is pretty huge, I thought my 300ish datasets were too big already... I can see why you need faster random forests!

[Atabarezz]

Well I turned it into an algorithm to make faster runs...

1. I run a small sample %5 with full variable set, make sure equal ratio of all labels to classify

2. Select %50 of the meaningful variables from random forest

3. Re run with a bigger sample X2 -> %10 with half the variables

in 2-3 iterations roughly know the prime variables needed and an ideal sub-sample size...

[marco_zara]

Can we hope from some feedback from the devs? It's sad to see most ML algorithms depending on single core speed in 2019, especially as the hardware to make them perform best is not something corporate procurement would approve...

[mojgan1987]

I fully support the proposal of adding LightGBM, XGBoost and CatBoost.