Build LightGBM and set up Virtualenv

git clone --recursive https://github.com/microsoft/LightGBM
cd LightGBM
mkdir build ; cd build
cmake ..
make -j64

cd ../python-package

sudo apt-get install virtualenv python3-venv
python3 -m venv /tmp/xendcg
source /tmp/xendcg/bin/activate
pip install scipy scikit-learn numpy wheel
python setup.py install

Running Experiments

You may reproduce:

• quality experiments:

  python main.py --paths /path/to/data/*libsvm

• experiments with 20% of labels perturbed:

  python main.py --label_noise 0.2 --paths /path/to/data/*libsvm

• experiments where each query is augmented by 40% of its original list size:

  python main.py --negs_from_others 0.4 --paths /path/to/data/*libsvm

• experiments where 10 impressions are generated for each query with click probabilities conditioned on labels [.05, .3, .5, .7, .95] (i.e., documents with label 0 are clicked with 5% probability):

  python main.py --click_impressions 10 --click_probs .05 .3 .5 .7 .95 --paths /path/to/data/*libsvm