search

ciemss / pyciemss

Causal and probabilistic reasoning with continuous time dynamical systems
Other
12 stars 4 forks source link

Missing values in Optimize results #498

Closed liunelson closed 4 months ago

liunelson commented 4 months ago

Optimize currently returns only policy, OptResults. However, it used to also return risk, samples, qoi. I think we need samples to plot what the optimal intervention policy looks like as a function of time (top of right panel), qoi and risk to plot the kernel density of the QoI (bottom right of right panel) and the risk threshold value (thick horizontal black line).

image

liunelson commented 4 months ago

@anirban-chaudhuri Is there a plan to bring these results back?

anirban-chaudhuri commented 4 months ago

I believe the plan is to just chain sample after running optimize. I am working towards bringing some of the original statistics back to the output format from sample. This is similar to the change with calibrate. You should be able to get samples using the optimal policy by using sample with the optimal intervention. See cell 12 in https://github.com/ciemss/pyciemss/blob/main/docs/source/interfaces.ipynb as an example for chaining optimize-sample. Let me know if this resolves the issue.

anirban-chaudhuri commented 4 months ago

I have added risk estimation back into the sample interface. See PR #504