[Feature request]: Subset Statistics By Subpopulation/Date in gempyor.statistics.Statistic

[TimothyWillard]

Label

enhancement, gempyor, inference, config, quick issue

Priority Label

medium priority

Is your feature request related to a problem? Please describe.

In some cases ground truth data may be missing, such as not reported, but would like to fit to said data anyways. An example would be maybe data is missing for some date range or is missing for a particular subpopulation.

See GH-481 for an example of NaNs in the ground truth that could've been ignored.

Is your feature request related to a new application, scenario round, pathogen? Please describe.

No response

Describe the solution you'd like

The addition of new options to a statistic configuration for subsetting by date/subpopulation. An example would be:

death_incidence:
  name: sum_death_incidence
  sim_var: incidD
  data_var: deaths
  resample:
    aggregator: sum
    freq: W-SAT
    skipna: False
  zero_to_one: True
  likelihood:
    dist: pois
  subpop:
    - "37000"
    - "06000"
  period_start_date: 2024-01-01
  period_end_date: 2025-01-01

But this style of configuration assumes that if you want to make modifications to a statistic it's that data is limited to some period, not that it is missing from some period so maybe some thought should be given to how to best specify "all dates except December 2023" or similar.

In practice I think we could replace the corresponding entries in likelihood with 0s as that will not change the log-likelihood optimization.