Parameter | Explanation |
---|---|
Free | |
Learning rate | Shrinkage rate (how much will the weights be adjusted every iteration). |
Number of leaves | Maximum number of leaves in one tree. |
Lambda L1 | L1 regularization. |
Lambda L2 | L2 regularization. |
Feature fraction | Randomly select part of the predictors on each iteration. |
Fixed | |
Early stopping | The cross-validation score needs to improve at least every n round to continue with the next boosting iteration. |
Maximum depth | Maximum tree depth (note that it is less relevant here since the tree grows leaf-wise). |
Minimum data | Minimal number of records in one leaf. A higher number prevents overfitting. |
Bagging fraction | Randomly select part of the data without resampling. |
Bagging frequency | Per how many rounds should bagging be applied. |
Unbalanced data | Does data need to be balanced or not. |