search space refactor and ray tune integration

intel-analytics / analytics-zoo

Distributed Tensorflow, Keras and PyTorch on Apache Spark/Flink & Ray

https://analytics-zoo.readthedocs.io/

Apache License 2.0

17 stars 4 forks source link

search space refactor and ray tune integration #899

Closed shane-huang closed 3 years ago

shane-huang commented 4 years ago

ConfigSpace is a very good library for search space specification. Which conditional settings for effective pruning. And it is used in SMAC (which is used in auto-sklearn, and also AutoPyTorch), has proven capability of describing search space as a graph.

shane-huang commented 4 years ago

new version of ray tune supports using config_space for BOHB search algo, while original grid and search still uses old style config. Consider using BOHB as default.