intel-analytics / analytics-zoo

Distributed Tensorflow, Keras and PyTorch on Apache Spark/Flink & Ray
https://analytics-zoo.readthedocs.io/
Apache License 2.0
11 stars 3 forks source link

support ensemble of the best models #984

Open shane-huang opened 4 years ago

shane-huang commented 4 years ago

We can make use of the best several models in search results instead of just the best one. In general, the models obtained from grid or random search are less correlated than those searched using bayersian methods. We can try ensemble the models find in grid or random search. We can ensemble models from different net topology, or even different model family.

Could consider stacking or bagging. Refer to H2O way of doing it. https://docs.h2o.ai/h2o/latest-stable/h2o-docs/automl.html#

shane-huang commented 4 years ago

Related to issue intel-analytics/analytics-zoo-internal#824