[Feature request]: new configuration file specification for parameters

[pearsonca]

Label

documentation, enhancement, meta/workflow

Priority Label

low priority

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

Users identified several issues with configuration file specification, and regularizing the approach to parameters seems necessary to support those changes.

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

No response

Describe the solution you'd like

There are currently several ways various analysis parameters are approached in the examples:

• files, either read in or run a file (initial_conditions)
• as parameters in seir
• as modifiers in outcomes / scenarios

Ideally, these would be under a unified interface. Thinking roughly from https://github.com/HopkinsIDD/flepiMoP/blob/main/examples/tutorials/config_sample_2pop_vaccine_scenarios.yml:

parameters:
    X0:
      file: # or treating this with a more functional representation, like file(name, opt=, opt=)
        read: model_input/ic_2pop.csv
        allow_missing_subpops: TRUE
        allow_missing_compartments: TRUE
    sigma: 1 / 4
    gamma: 1 / 5
    Ro: ~trunc(norm(mean=2.5, sd=0.1), min=1.1, max=3) # or some unbundling
    omega_pess: 0.02
    omega_opt: 0.01
    nu_pess: 0.01
    nu_opt: 0.03
    detection_probability: ~trunc(norm(mean=0.5, sd=0.05), min=0.3, max=0.7)
    detect_delay: 5
    hosp_probability: 0.05
    hosp_delay: 7
    hosp_duration: 10
    death_prob: 0.2
    death_delay: 14

seir:
  # definition basically identical, but no parameters section
  scenarios: # base = some reference scenario; if not present (including if there is no scenario key at all), automatically defined by no modifiers
    use: [base, no_vax, pess_vax, opt_vax] # default to all scenarios, use this to specify which; base only automatically included when unspecified
    pess_vax_nu:
      description: turn off nu_opt, only nu_pess left
      nu_opt: 0
    pess_vax_wane:
      description: turn off omega_opt, only omega_pess left
      omega_opt: 0
    pess_vax:
      description: turn off all vaccination
      stack: [pess_vax_nu, pess_vax_wane]
    opt_vax_nu:
      description: turn off nu_pess, only nu_opt left
      nu_pess: 0
    opt_vax_wane:
      description: turn off omega_pess, only omega_opt left
      omega_pess: 0
    opt_vax:
      description: turn off all vaccination
      stack: [opt_vax_nu, opt_vax_wane]
    no_vax:
      description: turn off all vaccination
      modifiers: [pess_vax, opt_vax]


outcomes:
  incidCase: delay(sample(incidence(I), detection_probability), detect_delay)
  incidHosp: delay(sample(incidence(I), hosp_probability), hosp_delay)
  prevHosp: accumulate(incidence(incidHosp), window = hosp_duration)
  incidDeath: delay(sample(incidence(incidHosp), death_prob), death_delay)
  scenarios: # base = some reference scenario; otherwise a "base" scenario is automatically defined in terms of no mods, and scenarios is unnecessary
    test_limits:
      incidCase: trunc(incidCase, max = SOMENUMBER) # slightly different model spec than example, but communicates idea
  

[pearsonca]

(probably comes after #436 - but possibly do this first, without the initial_conditions change, then build that off this)

[pearsonca]

@TimothyWillard this might be productively broken up into some stages.

e.g. being able to parse direct specification

# replace
parameter:
  a:
    value: 5
# with
parameter:
  a: 5

open to thoughts on how to make these a bit more bite-size

[pearsonca]

Hmm, also a tough part of this: the general case for distinct populations is that they are different collections of parameter values (e.g. population size). Need to do a bit of thinking about how to approach that.