An hybrid recommender system to predict the effectiveness of therapies in curing a medical condition of a patient. Check the report for more details on the implementation.
The system is based on a combination of different methods:
- Latent Factorization Models (SVD/SVD++)
- Collaborative Filtering (user-user/item-item)
- Nearest Neighbors
Clone the repository and install the required dependencies:
git clone https://github.com/materight/therapy-recommender-system.git
cd therapy-recommender-system
pip install -r requirements.txt
To produce recommendations for a patient and a condition, run:
python main.py -d [dataset_path] -p [patient_id] -c [condition_id]
Alternatively, it is possible to specify a test csv file containing the patient and condition ids to predict:
python main.py -d [dataset_path] -t [test_path]
Additional options can be specified (run with --help to see the available ones).
To evaluate multiple configurations for the recommender and perform an ablation study,run:
python benchmark.py -d [dataset_path]
The results will be saved in a csv file under the ./results folder.
The --val_split [fraction] option can be specified to customize the size of the validation split to use for evaluation.
A synthetic dataset can be generated by simply running:
python data_generator.py --n-patients [num_generated_patients] -o [output_path]
The generation can be customized with additional options (use --help to see the available ones).