Joost van Doorn

@Kaixhin I will have a look later.

I think this is in the `rnn` branch. This may or may not be a bug when using FastLSTM with the nngraph version. Setting `nn.FastLSTM.usenngraph = false` changed the error...

@lake4790k Do you have the flickering catch version somewhere?

@Kaixhin Great thanks. Have you tried storing the state instead of calling forget for every time step? I am doing this now, however it takes longer to train but it...

@Kaixhin Yes that line is enough, I will change that in my pull request. I missed `memSampleFreq`, so I assumed it was calling forget every time. I guess `memSampleFreq >=...

I guess like this; forget is called at the first time step so the LSTM will not have accumulated any information at this point, once [here](https://github.com/Kaixhin/Atari/blob/master/Agent.lua#L202) it will start accumulating...

The PR should be ready for review now. I ran into a typing issue after some further testing which I solved in the vendored library https://github.com/manicminer/hamilton/pull/172

I'll be able to work on this, we use this to automatically provision the Azure Databricks SCIM Provisioning Connector. I'll pick this up in a few days.

@bittnt Thanks for the response, in the end I ran it on Linux instead

@manicminer I refactored this over the weekend (and squashed the history). This should be good to go, let me know if there is still some issues.