Matt Gallivan

I found the solution to this, of course, minutes after posting. I'm missing the Boolean done in my structure. Thanks!

Just to update with a similar issue. Everything is running fine (training a-okay) but if I use custom parameters for A2C (or any of the algorithms), the `max_grad_norm` key is...

It would appear this might be a SocketIO issue in U5 Preview 1. See [this thread](https://github.com/getnamo/socketio-client-ue4/issues/268#issuecomment-875088572).

If this is still of interest I'd love to help out. I'm trying to build on Windows 10 with little luck. Any leads you may have found?

It should start from where it's left off. Are you reading the shuffle buffer as it restarting? Because it must do that each time you train (resume or begin anew).