chenhongge / robusttrees Goto Github PK

I am trying to follow the README to get started with this project. So far I ran

make

in the xgboost directory. I successfully created the project. Then I downloaded the conf files by running

./download_data.sh

Then I went up a directory and I gave permissions to xgboost like so

chmod +x ./xgboost

I followed the example

./xgboost data/ori_mnist.conf

However, I get back

zsh: permission denied: ./xgboost

I manually installed xgboost with pip. It works for some cases, but not for the one shown below:

xgboost data/ori_mnist.conf       
Error running xgboost:

Invalid Input: 'robust_exact', valid values are: {'approx', 'auto', 'exact', 'gpu_hist', 'hist'}
Use xgboost -h for showing help information.

Environnment :

Python 3.6 with ubuntu

Bug :

The monotone_constraints parameters have unexpected behavior with the robust tree method.

Condition :

XGBRegressor(monotone_constraints = monotone_constraints, tree_method = 'exact') works fine, but XGBRegressor(monotone_constraints = monotone_constraints, tree_method = 'robust_exact', robust_eps = 0) have different behavior.

(First time posting an issue, let me know if you want more information, and thank you anyway for this great code that you've provided !)

Thank you again for your paper and presentation. I absolutely love the idea! I was wondering if there is any effort underway to have this idea on scikit-learn or xgboost libraries? A "robustness" parameter for the model classes.