- High: It blocks me to complete my task.
Hi. I’m trying to get one of the offline algos to work with the trajectories created with SampleBatch. I’ve gotten no errors when using the default model, but I get several when using rnn. Both with just activating “use_lstm”: True and with custom rnn. The errors are KeyError for state_in_0, state_in_1, eps_id, weights… when using custom rnn I also get error on agent_indicies. All of which I assumed where supposed to be generated in rllib. I’m getting the samme errors with BC/MARWIL and R2D2. Is there an example somewhere of how to use rnn with these algorithms or is it an error in the docs that they can use rnn?