search

IntelLabs / coach

Reinforcement Learning Coach by Intel AI Lab enables easy experimentation with state of the art Reinforcement Learning algorithms
https://intellabs.github.io/coach/
Apache License 2.0
2.32k stars 461 forks source link

Weight update during Evaluate (Testing) #323

Closed nitsanluke closed 5 years ago

nitsanluke commented 5 years ago

Is there a way to switch between doing updates during the evaluation phase (testing).

gal-leibovich commented 5 years ago

Hi @nitsanluke, the evaluation phase (as the name suggests) is used for evaluating the last learned policy, resulting from the last training period. If wanted, you can either decrease the number of episodes or environment steps that are being executed at each evaluation phase in the ScheduleParameters defined in the preset.

nitsanluke commented 5 years ago

@galleibo-intel Thanks for the information. So in the evaluation phase, there aren't any online updates to the model is what you are confirming. Because online learning models generally are updating in the test as well (since they are always predicting future). If I want to make online updates in the test at this point I'm assuming using train_act is the best option is it? Thanks!