zencoding

@Alfredvc I don't understand the need to reset the LSTM with every minibatch. I would assume that we want the LSTM cell to learn across minibatch otherwise the LSTM is...

@Alfredvc thanks for the clarification, this helps. I am wondering why is it not done in most implementation that uses LSTM, for example, https://github.com/zencoding/DeepRL-Agents/blob/master/A3C-Doom.ipynb which is an on-policy A3C implementation,...