Improved dockerfile and added trajectory collection code to evaluate_dt_agent

jbloomAus / DecisionTransformerInterpretability

Interpreting how transformers simulate agents performing RL tasks

https://jbloomaus-decisiontransformerinterpretability-app-4edcnc.streamlit.app/

MIT License

61 stars 15 forks source link

Improved dockerfile and added trajectory collection code to evaluate_dt_agent #85

Closed JayBaileyCS closed 1 year ago

JayBaileyCS commented 1 year ago

Also covers PR https://github.com/jbloomAus/DecisionTransformerInterpretability/pull/84 while I'm at it, which failed due to issues that were fixed in https://github.com/jbloomAus/DecisionTransformerInterpretability/pull/83

Added code to accumulate trajectories when evaluating a dt agent with a TrajectoryWriter passed to it, in order to collect actions, observations, and rtgs for probing techniques.

jbloomAus commented 1 year ago

Thanks Jay!