Causal Reasoning

[apfeffer]

One of the tasks at the hackathon was to reason about causal interventions. We currently don't have good mechanisms to do this. We had to set evidence about the interventions and then do inference with that evidence, but this is (a) incorrect causal reasoning, and (b) a difficult inference task because there's a lot of evidence. It would be better if we could properly treat these as interventions, setting the values of the variables and breaking the connections to their parents. We could then simply forward sample from a known initial state to see the effects of the interventions.

[gtakata]

Is there a detailed description of the task that we can look at?

[apfeffer]

I’ll send around the hackathon instructions.

Actually, a simple solution would be to add a new kind of annotation to elements, which is an intervention, similar to conditions or constraints. Algorithms would be free to do with this what they want, but typically they will not generate values for these variables but use their intervention values instead. From: Glenn Takata notifications@github.com Reply-To: p2t2/figaro reply@reply.github.com Date: Wednesday, January 25, 2017 at 10:34 AM To: p2t2/figaro figaro@noreply.github.com Cc: Avi Pfeffer apfeffer@cra.com, Author author@noreply.github.com Subject: Re: [p2t2/figaro] Causal Reasoning (#672)

Is there a detailed description of the task that we can look at?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHubhttps://github.com/p2t2/figaro/issues/672#issuecomment-275140061, or mute the threadhttps://github.com/notifications/unsubscribe-auth/AFJkdxSiXHNd2gKBGM_jMCRuVlcXSqYUks5rV2uDgaJpZM4LtkD7.