Hi,
Is there a way to leverage the built-in LSTM but use it without the intermediary FCNET? I would like to have the observation space to be passed directly to the LSTM block. If not, is there an easy(ish) way to do so?
Thanks,
Spark
Hi,
Is there a way to leverage the built-in LSTM but use it without the intermediary FCNET? I would like to have the observation space to be passed directly to the LSTM block. If not, is there an easy(ish) way to do so?
Thanks,
Spark
Hi @Denys_Ashikhin,
You can do this with config[“model”][“fcnet_hiddens”] = [].
Here is a colab with that setting and a print out of the models it creates.