outbreak-probabilities

DTC Sandpit Challenge: methods for estimating the probability of a major outbreak

Table of Contents

1. To-Do
2. Set-up
   • Continuous Integration
   • Testing
   • Simulation of Trajectories
3. Methods
   • Method 1: Analytic Solution
   • Method 2: Trajectory Matching
   • Method 3: Machine Learning

To-Do

• Simulate – done
• Analytic – upload the cell in rough-work collab to GitHub without the sliders (input params)
• ML – write code that uses the simulated data and include plots (train 4 separate classifiers and save the models)
• Write unit tests (ask Matthew) to cover as many lines as you can

Set-up

Continuous Integration

• Create CI workflow in .github/workflows/ci.yml
  • GitHub Actions
  • Code Coverage

Testing

• Create tests in tests/test_name.py

  • Test all .py files (test each Method separately + IO)
  • Use PyTest

• Add Read the Docs documentation: https://docs.readthedocs.com/platform/stable/intro/add-project.html#manually-import-your-docs

Simulation of Trajectories

Input: first k weeks of infectious cases, e.g. k[0:3] of k = [1,2,6,8,...].
Output: a CSV file simulated_cases.csv with case number entries; columns are days, e.g. day_1, day_2, day_3, ...

• Consider using the tempfile module rather than saving to the user directory every time.

Methods

Method 1: Analytic Solution

Input:

• the first k days worth of simulated infection data from simulated_cases.csv
• estimated range for the reproduction number

Output:

• The conditional probability P([I1,I2,I3] | R)
• Outbreak probability given first three cases P(PMO | [I1,I2,I3])
• Outbreak probability given reproduction number P(PMO | R)
• Overall outbreak probability: (conditional probability) × (outbreak probability given reproduction number)

What to do:

1. Numerically compute the integral for the serial interval distribution
2. Compute the expected number of new cases

Method 2: Trajectory Matching

Input:

• Sequence of case counts

Output:

• All trajectories of cases where the first k days of simulated data match the observed sequence
• Outbreak probability: fraction of those trajectories classified as major outbreaks

Method 3: Machine Learning

Input:

• an observed input sequence of early case counts, e.g. data = [1,2,6] = k[0:3]
• ML model(s) trained on simulated trajectories

Output:

• predicted outbreak probability (and model metrics); saved model files (TBD)