2019-01-22 Check-in

Date

22 Jan 2019 

Attendees / Stakeholders

• Clay Crosby

• Swetha Srinath

• Timoteo Chaluco

• Felimone Amone Junior

• Dércio Duvane

• Christine Lenihan

• Vidya Sampath

• Ben Leibert

Notes

• In order to allow us to test recent changes to the materialized view, take screenshots of the dashboard, and ultimately demo the system this Friday, Timoteo Chaluco (Unlicensed) will send the team data (population and related coefficients) today. Clay Crosby (Unlicensed)'s team will load these values tomorrow. If the dataset is missing any values, Clay Crosby (Unlicensed)'s team will use made-up (fake) ones for now.
  
  
• Ben is under the impression that population and target coefficients are the only configured coefficients which change from one year to the next. Timoteo Chaluco (Unlicensed) thinks this is accurate but will double-check.
• Ben will delete the District dashboard. It's apparently possible to use the Flask app-builder can create a custom-permission which grants dashboard-level access to users. We don't have experience creating this permission, though, and figuring out how to use it will likely exceed the amount of time necessary to recreate the District-level dashboard if we were to need it in the future. It thus makes sense to simply delete the dashboard.
  
  
• There's not much momentum around creation of a NiFi registry, and it's fair to say that we have higher priorities. We'll thus plan to forgo use of the registry.
  
  
• The devision-by-zero problem occurs when a district reports that a positive number of vaccines were given but simultaneously reports that no vials were opened. Because this isn't logically possible, it represents a data-input error. Because our indicator requires that we treat vials-opened as a denominator, we run into division by zero.
  
  Although this happens at the facility and district levels, we haven't seen the problem at the provincial level.
  
  There following are two ways in which we can address the problem:
  
  A) Ignore the obviously-erroneous data. In this case, we wouldn't graph the problematic records. Our metrics would be technically accurate, but would be based on an incomplete dataset and would differ from those within DHIS-2 which incorporate faulty data. 
  
  B) Use a bar chart with dual-bars. Whereas one bar would show the number of vaccines given, the other would depict the number of vials opened. This option obviates the need for any division and thereby allows us to display the raw data as-is. We would not, however, show a percentage (as per the actual definition of the indicator.)
  
  We decided against option B because there's a lot of variation between usage. The fact that one district may have a values significantly higher than others would make the chart visually awkward. Also, because option B wouldn't yield the indicator (a percentage) desired, it would make direct comparisons related to the indicator difficult.
  
  We therefore decided to:
  
  1. Go with option A.
  2. Display a list of all districts which have bad data. (We'd ideally show a list of such facilities, but are unable to because we don't collect data from DHIS-2 at facility-level granularity.)
  3. Add text to the top of the Utilization tabs which describes the fact that we ignore data within DHIS-2 which is apparently nonsensical.
  
  
  
• Clay Crosby (Unlicensed) can fix the detail-table links prior to this Friday.
  
  
• Clay Crosby (Unlicensed) will be traveling from 2/6 - 2/24. Because he won't be on hand to field questions and issues which may arise as part of a production release, it wouldn't be wise to offer to release the software as-is during this period. We can, however, offer to be ready to offer a training session based on the current version of Superset by 2/6. Because a training exercise is a targeted, curated, one-time event, it's a lot less risky than a general invitation to use and rely on the software for production.
  

Action Items

• Ben Leibert will delete the District-level dashboard.
• Ben Leibert will email Clay about errors in NiFi's log related to its ability to backup the database.
• Ben Leibert will add text to the top of the Utilization tabs which describes the fact that we ignore data within DHIS-2 which is apparently nonsensical.
• Clay Crosby (Unlicensed) will create a table within Superset which shows a list of districts which have positive number of vaccines given but simultaneously report that no vials were opened.
• Clay Crosby (Unlicensed) will update the detail-table links.
• Clay Crosby (Unlicensed) will help ensure that Scalyr is setup to alert when RAM/CPU use exceeds a predefined threshold.