xgbviz
This interactive application illustrates the mechanics of the most important parameters in tree-based gradient boosting. You can try out the app online. For lower latency, clone the repository, open server.R in RStudio, and click on Run App.
Input parameters
You can modify the most important parameters in XGBOOST models
- Gamma
- Minimum child weight
- Sample size
- Maximum depth
- Maximum number of boosting iterations
- Early stopping
- Learning rate
Partial dependency plot
The partial dependency plot shows how well the learned function approximates the underlying data generating process. This is particularly interesting as you change the parameters Gamma, Minimum child weight, as well as the Sample size of the training data.
Loss function
This tab plots the performance (RMSE) on the training set as well as the test set against the number of boosting iterations. This is particularly interesting as you change the Learning rate, Max. number of boosting iterations, and Early stopping.
Check out the online version: https://maximilianeber.shinyapps.io/xgbviz.