Avoid divide by zero error when there are no non-index infections #52
Conversation
ringvax/app.py
Outdated
| @@ -268,19 +282,11 @@ def app(): | |||
| ) | |||
|
|
|||
| st.write( | |||
| "The following table provides summaries of marginal probabilities regarding detection. Aside from the marginal probability of active detection, these are the observed probabilities that any individual is detected in this manner. The marginal probability of active detection excludes index cases, which are not eligible for active detection." | |||
| "The following table provides summaries of marginal probabilities regarding detection. Aside from the marginal probability of active detection, these are the observed probabilities that any individual is detected in this manner. The marginal probability of active detection excludes index cases, which are not eligible for active detection, including the index case." | |||
There was a problem hiding this comment.
"The marginal probability of active detection excludes index cases, which are not eligible for active detection, including the index case."
I think this is just a typo? Index case is mentioned twice, once excluded and once inclusively ineligible
There was a problem hiding this comment.
I think this also highlights that perhaps what we really want, with respect to some of the discussion in #33, is to have 5 summaries, or maybe 6:
- Pr(index detected) (this is just the passive detection probability, accounting for the delay)
- Pr(non-index detected)
- Pr(non-index detected actively)
- Pr(non-index detected pasively)
- Pr(any case detected before it's infectious)
- If 6 summaries, this is two: Pr(index case detected before it's infectious) and Pr(non-index case detected before it's infectious)
ringvax/summary.py
Outdated
| @@ -26,6 +26,17 @@ | |||
| assert set(infection_schema.keys()) == Simulation.PROPERTIES | |||
|
|
|||
|
|
|||
| def frac(num, denom) -> float: | |||
There was a problem hiding this comment.
This is the behavior of numpy.divide, so we could use that?
It throws a user warning, but you can toggle that off for divide by zero warnings with https://numpy.org/doc/2.1/reference/generated/numpy.seterr.html#numpy.seterr
There was a problem hiding this comment.
It's also funny to me how much I would prefer
if num == 0.0 and denom == 0.0:
return nan
elif num > 0.0 and denom == 0.0:
etc.What/when/where I developed that preference?
There was a problem hiding this comment.
Looks like we can locally turn off the warnings with @np.errstate (otherwise we'd risk other things, like logs of negative numbers not warning elsewhere) so I'm deprecating frac in favor of divide plus a decorator on summarize_detections()
Probably overkill, but now we won't have problems even if there are no simulations at all.