mayankkejriwal/GNOME-p3

rl_agent_background missing

Closed this issue · 2 comments

r3a1d3a1 commented

@minhsueh Is there a reason you didn't check in 'rl_agent_background.py' as part of the 'phase3_simulator_v1' branch?

Furthermore, found this other repo wherein the step function does not match your reward description in the paper & let's the agent assign a reward to its own action which is plain wrong from a design perspective. Not to mention incentivizing skipping above other legal moves (+2 vs +1).

So, what's the deal? Did you fake your paper's results?

minhsueh commented

Hi @r3a1d3a1:

Thank you for the comments. Here are the answers to your questions:

  1. This repo aims to develop the Monopoly simulator. AI agent is not in our scope. But thank you for the reminder; I will clean up the code related to rl_agent_background.
  2. I am unsure which paper you are looking at, but those agents are not yet public. (we did not use repo in the paper.) If you have any questions regarding the RL agent, please get in touch with tbonjour@purdue.edu
r3a1d3a1 commented

Paper: https://arxiv.org/abs/2103.00683
At the bottom of page 6, this repo is referenced as the source.

I hope you decide to release the pre-trained agent mentioned in the paper. Thanks!

(Feel free to close this if no further relevant action will be taken.)

  • 👍2