"How to create and train a custom agent" doesn't actually do any learning

[Rylan Schaeffer]

The Colab notebook showing how to train a custom agent (https://colab.research.google.com/github/google/dopamine/blob/master/dopamine/colab/agents.ipynb) doesn't actually do any learning. Could we have an example of modifying the DQN architecture or its loss function?

[Pablo Samuel Castro]

thanks for pointing this out and the suggestion, which i think is great! we'll look into this.

[Bryan Yuan]

I think the reason was that the example agent was not using the .gin file given in the folder. I tried copy and paste the configs for DQN into the notebook and then it seems to have an identical learning curve.