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

Add indicator of the agent's action at each timestep #59

Closed danbraunai closed 2 years ago

danbraunai commented 2 years ago

Note that we probably don't want to put this in the "Agent Observations" plot, because the agent can't actually observe it.

danbraunai commented 2 years ago

@leesharkey See screenshot below for what the above commit does. Note that for this to work, panel_data.json needs to have the actions in it, which I handle in commit 069cdaa of the train-procgen-pytorch repo. I think something like this or having the arrow in the agent observation (but making a note to the reader that the agent can't see this) is probably best/easiest. Let me know what you think. image

leesharkey commented 2 years ago

This looks great!

I think not having the arrow within the observation is good because it'd just cause avoidable confusion.