LightBGM Algorithm
The LGBM algorithm is a fast, distributed, high performance gradient boosted framework based on decision tree algorithms. It is a relatively new algorithm that is challenging and outperforming existing algorithms.
Pros
- Faster training speed and higher efficiency
- Lower memory usage
- Better accuracy
- Support of GPU learning
- Handle large-scale data
Cons
- Sensitive to overfitting
- Not recommended for small data sets
Tree Growth
Most decision tree algorithms grow by level-wise. That is, they grow horizontally. However LGBM is unique because it grows leaf-wise or vertically. It will choose the leaf with the max delta loss to grow which can reduce more loss. The follow visual explains the growth:
