search

causalincentives / pycid

Library for graphical models of decision making, based on pgmpy and networkx
Apache License 2.0
96 stars 13 forks source link

VoI admission test for MACIDs #59

Open sbenthall opened 2 years ago

sbenthall commented 2 years ago

Is your feature request related to a problem? Please describe. A clear and concise description of what the problem is. Ex. I'm always frustrated when [...]

Describe the solution you'd like The test for VoI admission currently does not work for multi-agent games.

https://github.com/causalincentives/pycid/blob/5f78703de2fe1ac4ba6864e68f5af02ad923f38b/pycid/analyze/value_of_information.py#L17-L20

It would be nice if this test also worked for models with more than one agent.

RyanCarey commented 2 years ago

Yes, this is a very desirable feature. However, I don't know any good algorithm to compute that quantity [edit: this should've said "to evaluate admissibility"] in a multi-agent setting. Unless there's a paper I don't know about, this will have to wait for us to do more theoretical research before we can build it.

sbenthall commented 2 years ago

Is the admissibility criterion the same?

RyanCarey commented 2 years ago

I think that might depend 1) how we define VoI, 2) whether you include the ways that your policy might affect others policy, and 3) what rationality assumptions we place on the other agents. The answer will be yes for some answers to (1-3), but I'm not sure how we should answer such questions. Sorry if that's not a clear answer.

sbenthall commented 2 years ago

Thank you -- I see now that the question is not entirely well-formed.