hep-ml 0.5.0

Main points

• uniform classifiers - the classifiers with low correlation of predictions and mass (or some other variable(s))

• uBoost optimized implementation inside

• UGradientBoosting (with different losses, specially FlatnessLoss is very interesting)

• measures of uniformity (see hep_ml.metrics)

• advanced losses for classification, regression and ranking for UGradientBoosting (see hep_ml.losses).

• hep_ml.nnet - theano-based flexible neural networks

• hep_ml.reweight - reweighting multidimensional distributions (multi here means 2, 3, 5 and more dimensions - see GBReweighter!)

• hep_ml.splot - minimalistic sPlot-ting

• hep_ml.speedup - building models for fast classification (Bonsai BDT)

• sklearn-compatibility of estimators.

Installation

Basic installation:

pip install hep_ml

If you’re new to python and don’t never used pip, first install scikit-learn with these instructions.

To use latest development version, clone it and install with pip:

git clone https://github.com/arogozhnikov/hep_ml.git
cd hep_ml
sudo pip install .

Links

• documentation

• notebooks, code examples

• repository

• issue tracker

Related projects

Libraries you’ll require to make your life easier.

• IPython Notebook — web-shell for python

• scikit-learn — general-purpose library for machine learning in python

• yandex/REP — python wrappers around different machine learning libraries (including TMVA) + goodies, required to plot learning curves and reports after classification. Required to execute howtos from this repository

• numpy — ‘MATLAB in python’, vector operation in python. Use it you need to perform any number crunching.

• theano — optimized vector analytical math engine in python

• ROOT — main data format in high energy physics

• root_numpy — python library to deal with ROOT files (without pain)

License

Apache 2.0, library is open-source.

Platforms

Linux, Mac OS X and Windows are supported.