search

BasisResearch / collab-creatures

Analyzing animal collaboration with Bayesian and causal inference.
4 stars 1 forks source link

refactor code needed to make `random-hungry-followers.ipynb` run. #42

Closed rfl-urbaniak closed 7 months ago

rfl-urbaniak commented 7 months ago

The key changes are in random-hungry-followers.ipynb, with minor updates in locust_approximate_pipeline.ipynb (especially the visualization at the end. As some bugs are expected with your central park data I also added tests for the derivation of predictors in test_terivation.py. The test data were generated in the PR where animations can verify the correctness, in the future we want to make sure this doesn't get broken.

rfl-urbaniak commented 7 months ago

I had been thinking of this as more of a meta-parameter, but now it's plotted with a coefficient similar to the derived predictor terms in the value function (proximity and trace). How are you interpreting the coefficient on visibility?

So there is the visibility hyperparameter which is the visibility decay rate, and there is visibility score as assigned to points. As foragers have restricted movement range, visibility has a predictive value.