Summary

This chapter introduced LightGBM as a library to train boosted machines efficiently. We looked at where the complexity of building GBDTs comes from and the features in LightGBM that address them, such as histogram-based sampling, feature bundling, and GOSS. We also reviewed LightGBM’s most important hyperparameters.

We also gave a detailed overview of using LightGBM in Python, covering both the LightGBM Python API and the scikit-learn API. We then built our first tuned models using LightGBM to predict student academic performance, utilizing cross-validation and grid-search-based parameter optimization.

In the next chapter, we compare LightGBM against another popular gradient-boosting library, XGBoost, and DL techniques for tabular data.