Gbm.step update: gbm3, xgboost

SimonDedman / gbm.auto

Machine-learning Boosted Regression Tree software suite for species distribution modelling in R

https://doi.org/10.1371/journal.pone.0188955

Other

18 stars 6 forks source link

Gbm.step update: gbm3, xgboost #33

Open SimonDedman opened 6 years ago

SimonDedman commented 6 years ago

From Jane Elith: "a while ago I had colleagues look at the gbm.step code and improved it for parallel. They also fixed a bug. But I haven’t had time to test it or pass it on.

gbm.step is a bit outdated and won’t deal with all the families in gbm now, which is a shame. I just don’t have time to work on these things
gbm is now orphaned and gbm3 the new faster relative. I would be working now on gbm3 if it was me doing a lot of development. "

SimonDedman commented 5 years ago

Emailed Paul Metcalfe asking why gbm.fit isn't backwards compatible with gbm.step 2018.10.11, main function params look identical Emails Jane asking again about her/RObert's updated gbm.predict.grids code which is apparently faster.

SimonDedman commented 5 years ago

Discussed with Brandon @ gbm Feb 2019, recent update to gbm was bugfixes, state of gbm3 unknown, could just try it out.

SimonDedman commented 5 years ago

Could overhaul gbm with xgboost - see https://xgboost.readthedocs.io/en/latest/R-package/xgboostPresentation.html & related docs, can restructure gbm.auto to use this instead? Has inbuilt parallel & potentially better performance. Could also interface with CUDA GPU processing https://github.com/gpuRcore/gpuRcuda Doesn't support AUC though? https://xgboost.readthedocs.io/en/latest/gpu/index.html edit: aucpr not supported, auc is: https://xgboost.readthedocs.io/en/latest/gpu/index.html#metric-functions

SimonDedman commented 1 year ago

Use CatBoost instead, is better than all others. https://github.com/catboost/catboost https://catboost.ai/en/docs/installation/r-installation-binary-installation https://catboost.ai/en/docs/concepts/r-quickstart