Use GBM to prune features, but do not use GBM as base model when performing ensemble

[jmren168]

We're trying to train our customized models, and then use GBM to prune features. However, after feature pruning, GBM will be one of base models in the stage of ensemble. Is it possible to not use GBM when performing ensemble?

Settings:
# use gbm for feature_prune custom_model_hyperparameters.update({ 'GBM':{} }) # Define feature_prune_kwargs feature_prune_kwargs = { 'n_train_subsample': 50000, 'n_fi_subsample': 10000, 'prune_ratio': 0.05, 'stopping_round': 10, 'min_improvement': 1e-6, 'force_prune': True, }

Results:

Expcected Results:
Only CustomPipelineRFModel are used as base models in WeightedEnsmeble.

[Innixma]

I created a GitHub issue to track adding support for this: #3890