A machine learning project using convolutional neural networks and adversarial training for short-term COVID-19 epidemiological forecasting.
This project implements a machine learning approach for predicting COVID-19 epidemiological trends using multivariate time series data. The project includes specialized neural network architectures for processing city-specific features and making joint forecasts of multiple public health indicators across multiple time horizons.
- Data pipeline to transform datasets in the PHES-ODM v1 format into ready-to-use tensors.
- Time series forecasting for COVID-19 indicators.
- City-specific model adaptations.
- Visualization of the training process and modelling results.
- Results analysis and comparison tools.
- Qualitative trend analysis (slope, curvature, monotonicity).
# Clone the repository
git clone https://github.com/modelEAU/covid19-ml.git
cd covid19-ml
# Install the package with dev dependencies
pip install -e ".[dev]"- Python 3.10
- PyTorch >= 2.0.0
- PyTorch Lightning == 2.0.8
- Pandas, NumPy, SciPy
- Comet ML (for experiment tracking)
- Additional dependencies listed in pyproject.toml
The simplest way to train a model with default settings:
python src/covid19_ml/learning_loop.pyThis will train a model using the default "basicph_n1_smooth" recipe and the Quebec datasets specified in the configuration file.
python src/covid19_ml/learning_loop.py --recipename reg_cases_hosp_deaths_smooth --datasets qc1_2021 qc2_2021 montreal_2022python src/covid19_ml/learning_loop.py --debugpython src/covid19_ml/learning_loop.py --use_cometMake sure you have the environment variables COMET_API_KEY, COMET_WORKSPACE, and COMET_PROJECT set.
python src/covid19_ml/learning_loop.py --datasets qc1_2021 qc2_2021 --unseendataset montreal_2022python src/covid19_ml/learning_loop.py --batch_size 8 --dropout_rate 0.5 --leaking_rate 0.01python src/covid19_ml/learning_loop.py --model path/to/model.pt --newheaddataset new_city_2022For more complex scenarios, you can use the hyperparameter table approach:
python src/covid19_ml/learning_loop.py --hp_table hyperparameters.csv --hp_line 7This will load hyperparameter configuration from line 7 of the specified CSV file.
The system uses YAML configuration files to define:
- Model hyperparameters
- Dataset configurations
- Input/output recipes
- Logging settings
Example configuration structure:
models_folder: "models/"
data_folder: "data/"
logs_folder: "logs/"
metrics_folder: "metrics/"
hparams:
# Hyperparameters
batch_size: 32
learning_rate: 0.001
# ... other hyperparameters
datasets:
# Dataset configurations
recipes:
# Input/output recipespytestMIT
Contributions are welcome! Please feel free to submit a Pull Request.
- Jean-David Therrien mail
- Patrick Dallaire
- Simon Halle
- Thomas Maere
- Peter A. Varolleghem
This project uses Comet ML for experiment tracking and visualization.