- Setup conda environment
conda env create -f environment.yml
- Get the catboost with the losses implemented and negative weights check disabled from the repository, build wheel according to the instructions.
- Install the compiled catboost with
pip
- Install the utility library for neural networks:
pip install git+https://gitlab.com/mborisyak/craynn.git@5ee1057bbc9bc9a9d2a16d826196cc28044cebb9
- By default, the precompiled CPU version of tensorflow is installed. If you plan to run the Neural Network experiments and have a GPU, you might want to install a GPU version
- To run boosting experimets, create the folder
data
and download thereHIGGS.csv.gz
from UCI
Higgs-Boosting.ipynb
compares different ways to treat sWeights in catboostHiggs-NN.ipynb
compares different ways to treat sWeigths in neual networksPlot*.ipynb
plots the plots as they appear in the paper*.pdf
are the plots