FEB 2023: We are no longer actively developing or supporting this project; it has been archived. Others who may want to use it for their own modelling or experimentation are free to do so. For any questions, please email our team.
This repository holds the codebase for the Wadhwani AI Covid modelling effort. During the epidemic (2020 and early 2021) this codebase was developed and used by the Wadhwani AI team to provide estimates of caseload and resource burden to various local governments around India. This codebase is now freely available for others to use and to build upon.
Our primary approach to forecasting covid-related outcomes was through parameter estimation of compartmental, SEIR-like models from count data (confirmed, active, recovered, deceased) using hyper-parameter optimization (hyperopt). However, the code has been abstracted such that that initial case was an instance of a broader theme. There is support to deal with a variety of data types and sources, forecasting models and uncertainty, and techniques for parameter estimation. For example, in addition to the SEIR-family, curve fit models like those developed at IHME can be used; instead of using hyperopt, Bayesian parameter estimation via Markov chain Monte Carlo (MCMC) is also supported. The codebase design treats these concepts as components that can be swapped in and out of a workflow depending on requirements or taste.
In addition to estimation and forecasting, this codebase includes tools for visualizations; making it an end-to-end resource for public health policy makers. There are several Jupyter Notebooks within this repository that can be run to build a standard set of reports. More advanced users can build custom reports by piecing together other components that this repository provides. Finally, developers can extend the capabilities of this codebase by creating concrete modules that extend our abstractions.
Please find the detailed documentation of this repo in this folder.
git clone --single-branch --branch master --depth 1 https://github.com/WadhwaniAI/covid-modelling.git
If you prefer using SSH, use SSH
pip install -r requirements.txt
It would be highly recommended that you use either conda or virtualenv. We have a very long requirements.txt, so some packages there may be old and unavailable.
If any line fails, install the packages using -
cat requirements.txt | xargs -n 1 pip install
run the cells in notebooks/seir/[STABLE] generate_report.ipynb (here) end to end to get an idea of the end-to-end modelling pipeline.
Detailed function and class level documentation can be found in sphinx. The HTML files are not on git, but the instructions to create them are given in sphinx/README.md
- To read more about the codebase structure, click here
- To read more about the data, click here
- To read more about data smoothing methods, click here
- To read more about SEIR models, click here
- To read more about IHME, click here
- To read more about fitting process, click here
- To read more about uncertainty estimation using ABMA, click here
- To read more about uncertainty estimation using MCMC, click here
- To read more about config file and how everything comes together, click here
- To read more about the oncall process, click here