Lasse Espeholt

What appears to work is using a custom protoc version in a build file, like: https://github.com/google-research/seed_rl/blob/master/docker/Dockerfile.grpc#L23 Should be fixed in this repo though.

Yes. The local script in the readme runs the actors and learner in separate processes.

server_address can be changed here: https://github.com/google-research/seed_rl/blob/master/common/common_flags.py#L23 it can also be set to e.g. unix:/tmp/agent_grpc for using Unix domain sockets instead of the network stack. Other than that, you see the...

I updated the repository to 2.4.1 and fixed both the Atari and DM Lab docker files. Unfortunately, we don't accept pull requests at the time for a number of reasons.

The training is started. As the description says, try Ctrl+b, [tab id] to switch to learner, actors, etc. We use tmux to show separate logs for every actor and the...

Yes, sorry, I believe you ran into an issue I fixed two hours ago. Please try and pull the latest repo.

You use this network: return networks.ActorCriticMLP(parametric_action_distribution, 1,[32,32]) which isn't really suited for atari. However, if you just do tf.cast([input], tf.float32) on the input in the network I'm guessing it will...

Do you build the grpc library in docker? If no, try and do that. If you do already: I have seen this error before with specific versions of TF, so...

We are using C++ grpc, not Python grpc. Do you right now use the prebuild .so file (i.e. you don't try to build the grpc library?

I'm not sure what goes wrong for you, the following in Docker works: FROM ubuntu:18.04 RUN apt-get update && apt-get install -y tmux libsm6 libxext6 libxrender-dev python3-pip RUN pip3 install...