zalandoresearch / crisp Goto Github PK

Wanted to let you know of this agent based epidemic simulator: https://github.com/google-research/agent-based-epidemic-sim

reported by @rherbrich74:

There must be a one-off mistake in terms of the time indices between Gibbs/fwd-sampling and LBP. See here for a marginal plot of the test_gibbs_sampler.py for only forward sampling and LBP (forward mode) with no contacts and no tests. The curves are shifted by one (this is for patient-zero=TRUE)

https://github.com/zalandoresearch/CRISP/blob/master/code/experiments/exp_5.1/exp51.py#L19
The code is currently:

contacts_3 = init_contacts(S=S, T=T, R0=R0_3, seed=args.seed )
fig_3 = make_figure(**contacts_2**, t_branch=120, contacts_branch=contacts_3)

However, the first argument should be contacts_3.

From Kevin Murphy: "I am looking for a good simulator, that can generate (in batch mode) a set of 'realistic' contacts for each user, plus the test diagnostic results for a subset of the population. I have looked into the Oxford OpenABM code and the MPI code (from Bernhard's team), but it seems that it might be easier to use CRISP?"