Experiments with training NNs and Catboost on sWeights

Setup

1. Setup conda environment conda env create -f environment.yml
2. Get the catboost with the losses implemented and negative weights check disabled from the repository, build wheel according to the instructions.
3. Install the compiled catboost with pip
4. Install the utility library for neural networks: pip install git+https://gitlab.com/mborisyak/craynn.git@5ee1057bbc9bc9a9d2a16d826196cc28044cebb9
5. 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
6. To run boosting experimets, create the folder data and download there HIGGS.csv.gz from UCI

Things in the repository

• Higgs-Boosting.ipynb compares different ways to treat sWeights in catboost
• Higgs-NN.ipynb compares different ways to treat sWeigths in neual networks
• Plot*.ipynb plots the plots as they appear in the paper
• *.pdf are the plots