Accuracy experiment with data assimilation for paper results

tapios / risk-networks

Code for risk networks: a blend of compartmental models, graphs, data assimilation and semi-supervised learning

Other

2 stars 2 forks source link

Accuracy experiment with data assimilation for paper results #131

Closed odunbar closed 4 years ago

odunbar commented 4 years ago

Co-authored with all coders!

They will add details below as appropriate.

In this new examples/da-paper folder we would like the following list of examples:

[x] An example of an epidemic model run da-sandbox/saving_contact_networks_epidemic.py
[x] An example of loading an epidemic to run with a simple forward filter assimilation scheme and full (although changeable) user base da-sandbox/forward_filter_epidemic.py
[x] An example of the above with a forward - backward filter assimilation scheme da-sandbox/backward_forward_filter_epidemic.py
[ ] An example of a data assimilation scheme (filter or otherwise) coupled with interventions

This branch is purely to keep the running set of examples.

We have also Resolved #132 by using a 2-step data assimilation scheme.

dburov190 commented 4 years ago

hm, somehow the pull request stopped updating after I reset the 2f97dff commit and force-pushed; it's weird, should we create a new pull request? the branch itself is updated

(there are really no conflicts anymore, they were due to that 2f97dff commit; now everything should be fine)

UPD: ah, nevermind, it was probably Github maintenance late last night

dburov190 commented 4 years ago

@odunbar @jinlong83 I think I'm okay with merging this now... didn't rename "accuracy" -> "performance" yet because Ollie's performance stuff is in the other branch.

The only files left are:

underscore modules which will be without underscores in a new repository (the one with paper experiments)
accuracy_estimation.py, the main one for estimating accuracy/performance
accuracy_estimation_script.py, which is used for running the main one in parallel on desktops
accuracy_estimation_sbatch.sh, which is used for running the main one in parallel on the cluster

The latter two do not quite belong here, but since this is called sandbox, it's okay for now. Once we have a repository for experiments, we can make a separate scripts directory there, for example.

Is there anything else left?

dburov190 commented 4 years ago

Also, does one of you want to create the new repository? something like risk-networks-paper-experiments, or just risk-networks-experiments? It's probably about time (:

odunbar commented 4 years ago

I don't know what you mean by repository here. like a full copy of the current setup? Because currently we are still developing stuff. I feel like the repository should be created when the development is finished for the paper-examples. I.e once we have produced them we make a copy.

dburov190 commented 4 years ago

the new repository -- I mean, just the files from examples/paper and examples/da-sandbox

the underscore modules go into something like covidrisk/ (which will be a package there), and the experiments go into experiments/ -- something like that?

It's true that we're still developing, but we might as well continue developing there -- it'll also be a git repository after all.