search

uace-azmet / azmet-forecast-qa

Developing QA/QC routines for AZMet
0 stars 1 forks source link

New plan for wrapping up #45

Closed Aariq closed 1 year ago

Aariq commented 1 year ago

@jeremylweiss and I talked just now and I think the new plan for wrapping up the incubator phase of this project is as follows:

  1. Minimal usable product:

    • [x] An interactive document on viz.datascience.arizona.edu
    • [x] Date range selector using Shiny
    • [x] Pulls data from API using azmetr (possibly with only daily data or using caching to keep load time manageable)
    • [x] Rule-based validations only

  2. Use targets pipeline only to create forecasts:

    • [x] Runs on GitHub actions
    • [x] Automated in a way that saves _targets/ data store between runs
    • [x] Combines "legacy" data with data from API
    • [x] Fits models once per year
    • [x] refit()s models daily
    • [x] Builds dataset of observed value, forecasted value, and predictive intervals for all stations and variables
    • [x] Saves output to viz.datascience.arizona.edu as a pin
    • Most of this is already done and hopefully the rest is relatively painless

  3. Add forecast-based validations to report:

    • [x] Add forecast-based validations back to report by pulling in the forecast data saved on the pinboard
    • [x] Same date filter will apply
Aariq commented 1 year ago

Could be time to re-evaluate using Posit Connect to schedule the runs of the pipeline again. Might be easier than doing it with GH Actions because publishing the data to a pin will "just work" I think (https://pins.rstudio.com/articles/posit-connect.html#automating)

Instructions: https://solutions.posit.co/gallery/targets/ Example repo: https://github.com/sol-eng/targets-deployment-rsc

Aariq commented 1 year ago

Report published here: https://viz.datascience.arizona.edu/content/cc035e39-00b7-488b-983c-618c3a331ce4