Inconsistent results between runs

[robynstuart]

s1 = ss.Sim(n_agents=2e3, pars=dict(networks='randomnet', diseases='sir'))
s2 = ss.Sim(n_agents=2e3, pars=dict(networks='randomnet', diseases='sir'))
s1.run()
s2.run()
s2.summarize()
s1.summarize()

returns:

s2.summarize()
Out[10]: 
#0. 'n_alive':            3903.8888888888887
#1. 'new_deaths':         133.88888888888889
#2. 'sir_n_susceptible':  239.30555555555554
#3. 'sir_n_infected':     3649.8611111111113
#4. 'sir_n_recovered':    14.722222222222221
#5. 'sir_prevalence':     0.9664213405874297
#6. 'sir_new_infections': 277.77777777777777
s1.summarize()
Out[11]: 
#0. 'n_alive':            3848.6111111111113
#1. 'new_deaths':         135.0
#2. 'sir_n_susceptible':  240.83333333333334
#3. 'sir_n_infected':     3583.3333333333335
#4. 'sir_n_recovered':    24.444444444444443
#5. 'sir_prevalence':     0.9634548621557708
#6. 'sir_new_infections': 277.77777777777777

these should be identical

[robynstuart]

Updates:

• All results are the same (as expected) if I run two identical sims with if I use syphilis as the disease instead of sir
• Likewise, results are identical if I run a modified version of the SIR model without deaths
• So the difference between sims seems to be happening because of disease-related deaths...

[robynstuart]

... or maybe not - changing dead_uids to an empty array in the sir model doesn't fix the problem, but commenting out

self.ti_dead[rec_uids] = sim.ti + p.dur_inf.rvs(dead_uids)

on line 68 of sir.py does 🤔

[daniel-klein]

Oh yikes, and this is just using the single rng! I am able to reproduce the bug, and have tested reproducibility extensively, so there's something strange going on here. One initial idea is that I saw a warning that ti_dead was repeated in a ndict, maybe something related to that?

[daniel-klein]

Perhaps as expected, renaming ti_dead in sir didn't change anything. Hmm!

[daniel-klein]

Found the bug! self.ti_dead[rec_uids] = sim.ti + p.dur_inf.rvs(dead_uids)

That should be rec_uids on the left, not dead_uids!

[daniel-klein]

Fixed