JHU-Cornell_hierarchSIR

A hybrid multi-strain SIR - Bayesian hierarchical discrepancy model for infectious disease forecasting.

Installation (local)

Available platforms: macOS and Linux.

Note: Will not work on Windows because the OS-dependent path to the C++ Boost libraries is not included in setup.py.

Setup and activate a conda environment

Update conda to make sure your version is up-to-date,

```bash
conda update conda
```

Setup/update the environment: All dependencies needed to run the scripts are collected in the conda hierarchSIR_env.yml file. To set up the environment,

```bash
conda env create -f hierarchSIR_env.yml
conda activate HIERARCHSIR
```

or alternatively, to update the environment (needed after adding a dependency),

```bash
conda activate HIERARCHSIR
conda env update -f hierarchSIR_env.yml --prune
```

Install the Boost libraries

Install the C++ Boost libraries needed to integrate the multi-strain SIR model, for Linux users,

```bash
sudo apt-get update && sudo apt-get install -y libboost-all-dev
```

Mac users must install Boost through Homebrew,

```bash
brew install boost
```

Note: Boost is a C++ library and is not installed "inside" the conda environment but rather on your local machine. In setup.py the software is pointed to the location of the Boost library.

Install the hierarchSIR package

Install the hierarchSIR Python package inside the conda environment using,

```bash
conda activate HIERARCHSIR
pip install -e . --force-reinstall
```

Note: The installation script requires the use of pybind11 to "bind" the multi-strain SIR model in C++ to a Python function. This is the purpose of pyproject.toml. Note: If you make any changes to the C++ files you need to reinstall hierarchSIR.

Model training and forecasting

See scripts/code/hierarchical_training.py and scripts/code/incremental_calibration.py.

Running on the JHU Rockfish cluster

See JHU-ROCKFISH_README.md.