#560 Limit forecasting window to one month

[mangalap123]

Limit forecasting window to one month #560 Setting "number of days to project" min is 1, max is 30

Link to issue

• Fixes #560

Changelog entry

• Short description of problem & solution (1-3 sentences, depending on scope of change) Setting the Min(1) and Max(30) numbers for "number of days to project" parameter

Additional info

• Anything else reviewers and future readers should know

---

Pull Request Checklist

For Reviewer

• [ ] Submitted under the correct tag
• [x] Runs without error or warning
• [x] Documentation revised / included
• [x] Unit tests included
• [x] Issues referenced
• [x] Does not create new bugs / issues

[DougManuel]

I couldn't find the reason for this change. It will be difficult for us and I expect others who use the app for longer range counterfactual examples of the effectiveness of social distancing.

[BrianThomasRoss]

I couldn't find the reason for this change. It will be difficult for us and I expect others who use the app for longer range counterfactual examples of the effectiveness of social distancing.

Hi there, @DougManuel so I would encourage you to check out the repository here.

My basic understanding, and I'd welcome correction by members of the Penn team if I'm handing out bad information, is that currently they have parameters being utilized by the model that essentially represent a single snapshot in time. Using that snapshot a forecast is then generated by using those limited points to outline generally course that could be expected from the "snapshot". So I think of this like trying to do facial recognition with a very low quality video camera.

The model at the repo linked above makes tremendous strides in improving the accuracy and utility of the forecasts.

• It can accepts as input 2 matrices one consisting of the recorded census data similar to the output of this model, as well as another container more refined approaches to things like calculating reduction in relative contact rate, rates of specific interventions, and LOS data and some hyperparameters to allow for finer tuning.

• It incorporates Bayesian inferencing techniques to update the model as it moves through the timeline which produces a perhaps more accurate reflection of reality but also generates distributions of predictions which could allow for answering questions like "Whats my best case worst case here?"

[DougManuel]

Thanks for the explanation. From a user's perspective, what has been helpful from this repo's model has been for us has been:

• accessible - web-based
• initiated on hospital census
• includes social distancing.

The effect of social distancing dwarfs all other parameters outside the R0, etc. There are many models available that focus on various improvements, but all models come to the same conclusion about the role of social distancing.