Bart Maciszewski

Hi assuming you have an action space such as gym.spaces.Discrete(100) and the processed column is part of your observation then you just need to define an observation_and_action_constraint_splitter function that cycles...

Hi Marco, I am working on a similar agent but I am using gym.spaces.Discrete(num_actions) and relying on the TFPyEnvironment wrapper to configure my action_spec. I checked my resulting action_spec and...

Do you know where in TF code this error is coming from? I encountered a similar error with a DQN agent and a scalar action space and using an observation...

Nice!

Hi Sergio, how would specify the observation spec to pass to the qnetwork at creation time so that it is compatible with the true observation spilt out by the observation_and_action_constraint_splitter?...

Hi Marco, I was even thinking it maybe possible to just pass the entire observation together with the action mask to the network. In the worst case the network would...

Hi assuming you got your action space and mask splitter function I think you can try what Marco described or just do as I described [here](https://github.com/tensorflow/agents/issues/255#issuecomment-559870536) and pass the entire...