search

interpreting-rl-behavior / interpreting-rl-behavior.github.io

Code for the site https://interpreting-rl-behavior.github.io/
Creative Commons Attribution 4.0 International
0 stars 0 forks source link

Generate data that we can use to show that the environment simulation functions independent of the agent #25

Closed leesharkey closed 3 years ago

leesharkey commented 3 years ago

This is actually very important and is something I'm not totally confident about.

If we can't provide a convincing demonstration that this works (that the environment simulation function independent of the agent), then it's back to training the gen model in such a way that it works (e.g. by using multiple agents, some of which are trained, some of which are not).

The quality of the generations for the demos for the AISC presentation were okay but not ideal. We haven't yet checked it for this generative model. Assigning this to @danbraunai . IIRC you did this for the presentation too.

Once the demos are produced, then put them in the article.

leesharkey commented 3 years ago

Closing this issue because @danbraunai showed that it the environment did not function properly independent of the agent. This calls for modifications to the gen model and to its training method.

Closing this issue, though we will need to repeat this step later when we have our new model.