Yegor Tkachenko

http://yegortkachenko.com/

New York, NY

Results 4 issues of


                                            Yegor Tkachenko

Simulator should produce observation array of shape (1,x,...) instead of (x,...) - to be explicit about the fact that it is a single observation. Otherwise, we get an error inside the learner.

Add actor-critic learner module that uses separate action and value nets

Add a special simulator module to allow for training of a DQN from historical observations only, when simulator is not available

LSTM/RNN

need to add a learner to support recurrent neural nets