Scenario comparison for deterministic models

[pratikunterwegs]

This PR is WIP #221 and adds basic intervention scenario comparison functionality. The function outcomes_averted() offers options for calculating the differences in outcomes averted (cases, and deaths where applicable) between a baseline scenario (with parameter uncertainty), and multiple comparator scenarios (with identical parameter uncertainty).

• See the vignette on scenario modelling, modelling_scenarios.Rmd for how to use this.
• There's a limitation that this only works for models with parameter uncertainty. This is because the output format of a model with a single parameter set does not preserve model arguments/infection parameters which are used for matching. This in turn is because it was felt that users should be able to run a non-vectorised model and get simplified output that can be passed directly to plotting (see Vector inputs for ODE models #176).
• There's no support for comparing scenarios where the axis of variation is an infection parameter, e.g. $R_0$ is one of 1.3, 2.0, 2.5 etc.
• Support for the Ebola model is forthcoming in a separate PR based on feedback from this one.

Happy to take feedback on the implementation here, and more general discussion on scenario comparison in #221.

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if a6e001e is merged into main:

• 🚀default_ode: 10.6ms -> 10.5ms [-1.79%, -0.01%]
• ✔️default_ode_interventions: 97.2ms -> 95.9ms [-2.98%, +0.47%]
• ✔️default_ode_param_vec: 845ms -> 842ms [-1.4%, +0.76%]
• ✔️default_ode_paramvec_intervs: 6.19s -> 6.17s [-1.27%, +0.7%]
• ✔️ebola: 577ms -> 575ms [-1.44%, +0.83%]
  Further explanation regarding interpretation and methodology can be found in the documentation.

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if a6e001e is merged into main:

• ✔️default_ode: 10.7ms -> 10.6ms [-2.01%, +0.96%]
• ❗🐌default_ode_interventions: 93.4ms -> 94.6ms [+0.02%, +2.4%]
• ✔️default_ode_param_vec: 819ms -> 819ms [-0.94%, +0.92%]
• ✔️default_ode_paramvec_intervs: 6.2s -> 6.2s [-0.65%, +0.42%]
• ✔️ebola: 550ms -> 549ms [-1.44%, +1.23%]
  Further explanation regarding interpretation and methodology can be found in the documentation.

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if a6e001e is merged into main:

• ✔️default_ode: 10.8ms -> 10.8ms [-0.48%, +1.07%]
• ✔️default_ode_interventions: 97.7ms -> 97.7ms [-1.02%, +1.04%]
• ❗🐌default_ode_param_vec: 832ms -> 841ms [+0.14%, +1.9%]
• ✔️default_ode_paramvec_intervs: 6.24s -> 6.24s [-0.62%, +0.54%]
• ✔️ebola: 563ms -> 562ms [-1.07%, +0.56%]
  Further explanation regarding interpretation and methodology can be found in the documentation.

[joshwlambert]

@pratikunterwegs nice work. All looks good to me. The new documentation in the vignettes makes the new functionality very clear. Left a couple of minor comments in the code.

[joshwlambert]

Leaving a comment here for figure 2 below. The x-axis labels are overlapping on the pkgdown site, making it hard to match the column with the intervention.

[pratikunterwegs]

Yep - unsure how to fix this. I tried setting n.dodge = 3 but this still has overlaps.

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if a6e001e is merged into main:

• ✔️default_ode: 10.8ms -> 10.8ms [-0.83%, +1.14%]
• ✔️default_ode_interventions: 100ms -> 100ms [-1%, +1.42%]
• ✔️default_ode_param_vec: 837ms -> 835ms [-0.68%, +0.39%]
• ✔️default_ode_paramvec_intervs: 6.24s -> 6.26s [-0.1%, +0.79%]
• ✔️ebola: 570ms -> 566ms [-1.57%, +0.14%]
  Further explanation regarding interpretation and methodology can be found in the documentation.

[pratikunterwegs]

Thanks @joshwlambert. I haven't found a good fix for the second issue, but I'm merging this now so I can move ahead with more scenario functionality.