LightGbmDotNet

A .NET wrapper for the LightGBM machine learning library, including GPU support.

LightGBM is one of the leading Gradient Boosting solutions, provided open source by Microsoft.

Usage (C#):

using(var lightGbm = new LightGbm())
{
   lightGbm.Train(trainingData); //optionally pass parameters
   var predictions = lightGbm.Predict(predictionData); //returns an array of predictions
   lightGbm.SaveModel("somefile.txt"); //for later reuse of the trained machine learning model without retraining
}

Where both trainingData and predictionData are of type IEnumerable<IEnumerable<double>> representing rows of double values.

First column in trainingData is the label column (the value that you want to predict). Omit this column in predictionData.

License

MIT

Notes

• You might need to install the Visual Studio 2015 C++ Redistributable on the machine you want to run this on.
• Please be aware this wrapper uses files to transfer data to LightGBM and may write a considerable amount of data on your disk
• Don't forget to dispose the LightGbm instance after use to make sure the created files are cleaned up
• To use GPU acceleration (only provided for NVIDIA cards) pass true to the constructor of LightGbm class. If it crashes you might need to install the OpenCL implementation from NVIDIA (CUDA).
• Works only on Windows, 64 Bit, .NET Framework 4.6.1 or higher (should also work with lower .NET Framework versions if compiled with it)

Alternatives

Not tested myself so far:

• XGBoost wrapper in SharpLearning => highly recommending you to check this out
• LightGBMSharp => looks a bit basic so far

Both libraries work using in process / in memory data transfer to / from the native machine learning libraries, which should lead to superior performance compared to LightGbmDotNet