Conversation
|
@albrja or @patricktnast any possibility you could help me add tests for this? |
zmbc
requested review from
albrja,
hussain-jafari,
patricktnast,
rmudambi and
stevebachmeier
as code owners
zmbc
requested review from
NathanielBlairStahn,
SylLutze,
aflaxman,
alibow and
pletale
|
|
||
| When a target 95% UI is specified instead of a single target value, | ||
| we use a `normal inverse-gamma distribution <https://en.wikipedia.org/wiki/Beta_distribution>`_ that has approximately that UI | ||
| on the mean (if the variance parameter is marginalized out) and the same :math:`IG(2, s_{ref}^2)` prior on |
There was a problem hiding this comment.
IG or \Gamma^{-1}? Better to be consistent, and I would recommend switching all to \text{InverseGamma} as the most clear option.
| specifically :math:`N\text{-}\Gamma^{-1}(0, 10^{-3}, 2, s_{ref}^2)` | ||
| where :math:`N\text{-}\Gamma^{-1}` refers to `the normal-inverse-gamma distribution <https://en.wikipedia.org/wiki/Normal-inverse-gamma_distribution>`_ | ||
| and :math:`s_{ref}` is set to the target mean or the midpoint of the UI. | ||
| In fact, rather than specifying :math:`10^{-3}` for the :math:`\lambda` parameter, |
There was a problem hiding this comment.
I found this confusing; maybe use a variable instead of 10^{-3} above, and then define it below.
|
Note: while this is marked "ready for review" because I'd like review from researchers on the documentation, engineers should likely not review this yet -- no tests yet! |
| - We calculate how likely your data is under each scenario, and compare them. If your data is much more likely under the bug/issue scenario, we flag it. | ||
|
|
||
| For more details, see: | ||
| https://vivarium-research.readthedocs.io/en/latest/model_design/vivarium_features/automated_v_and_v/index.html#proportions-and-rates |
There was a problem hiding this comment.
I think these links have moved (to here?)
|
As a general status update on this, I think I should be able to find a better mathematical formulation that will not require the user to make such an accurate guess about the variance parameter; I could not find a version of the current formulation that passed tests for a variety of misses on this parameter. I hope to get back to this! |
3 participants
Add fuzzy checking for continuous means
Description
Changes and notes
Added and documented fuzzy checks for means of continuous values.
Testing
Note: Tests are currently failing! I need some help figuring out how to test this, aside from in-situ in the MNCNH V&V.