EMSA: Epidemic Modeling, Sensitivity Analysis

The purpose of this project is to provide a general framework for the sensitivity analysis of deterministic compartmental epidemic models, as well as a tool for predictive modeling. Sensitivity analysis is performed using Latin Hypercube Sampling (LHS) for parameter sampling and Partial Rank Correlation Coefficient (PRCC) as a metric for sensitivity.

The pipeline is fully compatible with CUDA and provides an efficient representation of the epidemic model used, enabling parallel evaluation based on different parameter combinations and initial values. For evaluation of the model we use the package torchode, and we represent the ODE system corresponding to the model with matrix operations.

Installation Guide

This guide provides the essential steps to install EMSA on your system.

System Requirements

• Python: Version 3.11 (should work from 3.9, but it may require adjusting the dependencies)

• Operating System: Windows, macOS, or Linux.

• (Optional) CUDA: 11.8+

Basic Installation

Install EMSA using pip:

pip install emsa

To experiment with the latest code and explore implemented examples, you can install the development version by cloning the repository:

git clone https://github.com/KKol21/epimodel-sensitivity

Optional: verify CUDA compatibility

After installing PyTorch, you can check if CUDA is available by running the following:

import torch
print(torch.cuda.is_available())

For further instructions on setting up CUDA, please see the official PyTorch guide. This is only relevant if you’d like to use a GPU for evaluation.

Getting started

Contents:

• Use cases for EMSA
  • Preparing Model Inputs for EMSA
  • Model evaluation
  • Sensitivity analysis
• Advanced topics
  • Extending the basic model format
  • Sampling form non-uniform distributions
  • Using different numerical methods
  • Running simulations with varying parameters
• API reference
  • Model module
  • Sensitivity module
  • Utilities
  • Target calculator module