LightGBM

LightGBM, short for Light Gradient Boosting Machine, is a gradient boosting framework that uses tree-based learning algorithms. Developed by Microsoft, it's designed to be distributed and efficient with the following advantages: faster training speed and higher efficiency, lower memory usage, better accuracy, support of parallel and GPU learning, capable of handling large-scale data.

LightGBM in practice

LightGBM is based on decision tree algorithms and performs a gradient boosting framework. It splits the tree leaf-wise while other algorithms split the tree level-wise. It will choose the leaf with max delta loss to grow, thus reducing more loss than the level-wise algorithms. This is a key factor contributing to its speed and efficiency.

In addition, it uses a novel technique of Gradient-based One-Side Sampling (GOSS) to filter out instances for finding a split value, while keeping the accuracy of learned decision trees. GOSS keeps all the instances with large gradients and performs random sampling on the instances with small gradients.

Another key technology is the Exclusive Feature Bundling (EFB), which is used to significantly reduce the number of features. In real-world scenarios, many features are exclusive, which could be bundled into one feature, thus the number of features can be largely reduced.

Combined all these, LightGBM is powerful in managing large datasets, with faster training and better accuracy. It supports categorical features by value, not requiring one-hot coding. It also supports both continuous and categorical features as well as missing value-infested datasets. It's also worth noting that LightGBM can be integrated with many different machine learning frameworks, including (but not limited to) SciKit-Learn, R, and Python.