LightGBM

LightGBM is a tree-based gradient boosting library designed to be distributed and efficient. This algorithm provides fast training speed, low memory usage, good accuracy and is capable of handling large scale data.

For more information on gradient tree boosting, see the "Gradient tree boosting" algorithm.

Boosting type

Try Gradient Boosting Decision Tree Try Gradient One-Side sampling At least one option must be selected.

"Gradient-based One-Side Sampling" (GOSS) is a variant of "Gradient Boosting Decision Tree" (GBDT) where, at each step of the training phase, all of the data with a large gradient is kept while random sampling is performed on the data with a small gradient.

Maximum depth of trees

Maximum depth of each tree. High values can increase the quality of the prediction, but can also lead to over-fitting. Typical values: 3 - 10. Set to -1 to unconstrain.

Use bagging

Bagging can be used to speed up training and/or deal with over-fitting. Please note that Gradient-based One-Side Sampling does not support bagging. As a consequence, bagging parameters will be ignored for this boosting type.

Subsample ratio

Subsample ratio of the training instance.

Subsample frequency

Frequency of subsample. <= 0 means no enable

Early stopping

Use LightGBM built-in early stop mechanism so the exact number of trees will be optimized.
The cross-validation scheme defined in the "Hyperparameters" tab will be used.

Early stopping is not available for causal prediction.

Early stopping rounds

The optimizer stops if the loss never decreases for this consecutive number of iterations. Typical values: 1 - 10

Random state

Using a fixed random seed allows for reproducible result

Parallelism

Number of cores used for parallel training. Using more cores leads to faster training but at the expense of more memory consumption, especially for large training datasets.

Allow sparse matrices

Allow DSS to use sparse matrices to train the model
This may help reduce RAM and CPU usage