-
Notifications
You must be signed in to change notification settings - Fork 3
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
TZA 2016-20 #452
Comments
Data pull: |
Model run: |
One chained failed to warmup, the other three OK. Suggestion: reject. Need to work on initialization. |
One chain failed initialization and others ok. Opinion: Reject and rerun after investigating initialization and with higher sfrac_thresh_border to remove some high rate cells in N border |
Opinion: Reject and rerun on dev with |
Data pull: |
64G is not enough for data pull. plan: rerun with 100G |
rerunning with 100G (same HASH as |
Model runnng: |
Suspicious border pixel with very high rates/cases in northern border |
2019 modeled case are slightly overestimated. Convergency looks good. |
I think there are sometimes 2 national annual estimates in Table 6 because the obs actually represent Zanzibar and I think we manually changed ZNZ to admin 0 in some places for plotting purposes -- (not sure how the cdr is pulling the admin level now). (Same issue occurs in TZA 2011-15 cdr). There are lots of national annual obs in Table 6 with varying tfracs and I think we could consider removing some of them so that the annual estimates are not artificially lowered. Temp Approve - could consider rerunning without lower tfrac (but above 0.65) national full annual obs |
Sep 2023 Production run: convergence OK. Rhats mostly OK, some have very large Rhats. These seem to be in zanzibar, to investigate further. |
Action Required - investigate large Rhats and rerun cdr to see if there are imputed observations in 2019 and 2020 We don't think imputation gets triggered if at least one of Mainland or ZNZ has an admin 0. Could change it so that ZNZ is no treated as admin 0 as one possible solution |
Second Action Required - In Table 6, it looks like the national observations for Mainland and Zanzibar have the same values in OC 20109 and OC 21025 (You can tell there are different estimates for Mainland and Zanzibar because they have different predicted values). We should audit these two observations -- it is suspicious that they have the same exact number of cases. Investigation: In OCs 21019 and 21025 there is a phantom for AFR::Tanzania because there are subnational cases that are all reported in AFR::TZA::Mainland and presumably if you read through the AFRO reports the subnational smaller tfrac obs are getting reported on the Mainland only Decision: We decided to keep both observations in the OCs |
Decision: Test whether the no-mixture config enables better model fit and reduces Rhat issues. |
After removing the admin 0 assignment for TZA Mainland and ZNZ, the model convergence and w estimate issues were resolved, even when using a mixture sd_w prior. There was one incorrect multi-year observation (censored) that was underestimated but that data has now been fixed in the database. Rerun with updated observation |
The multi-year observation is no longer present and the convergence and estimates look good. Approve |
No description provided.
The text was updated successfully, but these errors were encountered: