Danijar Hafner

Results 165 comments of Danijar Hafner

Maybe something like this: https://gist.github.com/danijar/1dde5eb528b61560734c16d4fd0d93a1 We will probably have to modify `agents/scripts/train.py` though.

Cool, I looked at the PR and it looks quite nice. The examples directory of the k8s repo is probably a good place for this code. I'm planning on adding...

Thanks. TensorBoard is getting a 401 Unauthorized return code. Yep, checked the code and saw that it's single-instance training for now. Let me know if you run into problems when...

I think we will have multiple machine that each run a batch of environments and an agent. The agents just need to synchronize their gradients. Since most of the time...

Yes, the model parameters are created within the `agents.scripts.networks.feed_forward_gaussian`. Another variable that me might want to share is the current value of the KL penality and of the normalizers for...

Hmm, I thought that the `tf.local_variables_initializer()` would do exactly that ([utility.py line 114](https://github.com/tensorflow/agents/blob/master/agents/scripts/utility.py#L114)). Maybe these variables are mistakenly placed on the parameter server?

@cwbeitel Are you still working on fixing this?

When using MonitoredTrainingSession, you should disable its logging and checkpointing logic. I was using it in the beginning but it was too restrictive so I switched to a standard session....