Introducing LightGBM
LightGBM is an open source, gradient-boosting framework for tree-based ensembles (https://github.com/microsoft/LightGBM). LightGBM focuses on efficiency in speed, memory usage, and improved accuracy, especially for problems with high dimensionality and large data sizes.
LightGBM was first introduced in the paper LightGBM: A Highly Efficient Gradient Boosting Decision Tree [1].
The efficiency and accuracy of LightGBM are achieved via several technical and theoretical optimizations to the standard ensemble learning methods, particularly GBDTs. Additionally, LightGBM supports distributed training of ensembles with optimizations in network communication and support for GPU-based training of tree ensembles.
LightGBM supports many machine learning (ML) applications: regression, binary and multiclass classification, cross-entropy loss functions, and ranking via LambdaRank.
The LightGBM algorithm is also very customizable via its hyperparameters. It supports...
