Skip to content

pashna/GradientBoosting

Repository files navigation

GradientBoosting

My own custom implementation of Gradient Boosting based on my own custom implementation of Decision Tree. On the average quality of this gradient gives in 2-4% higher than sklearn's implementation.

The reason for the quality increasing:

Moreover, the model allows you to use your own:

  • Loss Function (must be inherited from Abstract class). MSE and MAE are built-in.
  • Custom node purity for decision tree (must be inherited from Abstract class). Gini and Entropy are built-in.

File Boosting.ipynb shows that this algorithm works better than Sklearn's on the given datasets with the same parameters.

Since python code works incredibly slow, the implementation uses Cython in the most critical functions which gives a huge acceleration.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published