Errors when ensemble = True about flaml HOT 3 CLOSED

When ensemble=True and monotone_constraints is in self.params, you need to set the monotone_constraints according to **params in the constructor if it is passed. For example:

It's required by scikit-learn.

from flaml.

Got it. It is working now. Thank you!

To confirm, does FLAML fit an ensemble of all the base models created or only the best base model of each learner used? It seems that all the based models are used since the time taken for the ensemble significantly increases with more base models.

from flaml.

It fits an ensemble for the best base model of each learner. You should be able to see some msg like:

[flaml.automl: 02-17 11:42:05] {1161} INFO - ensemble: StackingClassifier(estimators=[('lgbm',
<flaml.model.LGBMEstimator object at 0x0000027F30E05F88>),
('RGF',
<test.test_automl.MyRegularizedGreedyForest object at 0x0000027F30E00C88>)],
final_estimator=<flaml.model.LGBMEstimator object at 0x0000027F30E05F88>,
n_jobs=1, passthrough=True)

It ensembles one lgbm and one RGF model in this example.

The increased ensemble time may be due to larger and better models found later in the search. You can check the config of the best individual models to verify that.

from flaml.