tth-bdt-hyperparameter-optimization

Evolutionary algorithms for hyperparameter optimization for BDT (XGBoost) and NN (with MNIST numbers dataset for testing).

Installation

If running with CMSSW:

git clone https://github.com/HEP-KBFI/tth-bdt-hyperparameter-optimization.git $CMSSW_BASE/src/tthAnalysis/bdtHyperparameterOptimization
cd $CMSSW_BASE/src
scram b -j 8
cd $CMSSW_BASE/src/tthAnalysis/bdtHyperparameterOptimization
pip install -r requirements.txt --user

• attrs==19.1.0 version needed due to 19.2.0 breaking with older version of pytest. update pytest to pytest==5.2.0

Also in order for feature importances to work with NN, eli5 package is needed:

pip install --user eli5

Wiki

For more detailed information visit the wiki

MNIST numbers dataset

Available here: Mnist numbers dataset

Tests

After installation please run the unittests (in $CMSSW_BASE/src/tthAnalysis/bdtHyperparameterOptimization) with:

pytest test/