ntree_limit is deprecated, use `iteration_range` or model slicing instead.

The parameter ntree_limit is no longer recommended to be used in XGBoost, instead, the recommended alternatives are iteration_range or model slicing.

Iteration range refers to the range of iterations to use for prediction. For example, if you want to predict using the first 100 trees of your model, you can set the iteration_range to (0, 100). Similarly, if you want to predict using the trees from the 100th iteration to the 200th iteration, you can set the iteration_range to (100, 200).

Model slicing refers to creating a new model object that only includes a subset of the trees from the original model. This can be done using the slice() method of the Booster object. For example, to create a new model that includes only the first 100 trees of the original model, you can use the following code:

new_model = original_model.slice(0, 100)

Using iteration_range or model slicing is recommended over using ntree_limit because they provide better control and flexibility over the number of trees used for prediction.