Personalised Short-Term Glucose Prediction via Recurrent Self-Attention Network

This is the official repository of the paper "Personalised Short-Term Glucose Prediction via Recurrent Self-Attention Network".

Dependencies

This repository has been tested on the following configuration of dependencies.

• Python 3.8.8
• torch 1.8.0
• numpy 1.19, pandas 1.2

Proprocessing

First, acquire the raw OhioT1DM data from http://smarthealth.cs.ohio.edu/OhioT1DM-dataset.html.

The raw data should have the following structure

ohiot1dm
|-- OhioT1DM-training
|-- OhioT1DM-testing
|-- OhioT1DM-2-training
|-- OhioT1DM-2-testing

Then run the following command to preprocess the data using our scripts.

python3 ./preprocess/linker.py --data_folder_path path/to/ohiot1dm --extract_folder_path ./data

Demo

For a fast demo of our results, run

python3 eval.py --ckpts_dir ./pretrained/set1_30

Train

An example of replicating the setting 1 on subject 540 for 30 minutes prediciton horizon.

python3 train.py --patient 540 --missing_len 6 --single_pred --transfer_learning

An example of replicating the setting 2 on subject 540 for 30 minutes prediciton horizon.

python3 train.py --patient 540 --missing_len 6 --transfer_learning

An example of replicating the ablation study 1 on subject 540 for 30 minutes prediciton horizon.

python3 train.py --patient 540 --missing_len 6 --single_pred --unimodal --transfer_learning

An example of replicating the ablation study 2 on subject 540 for 30 minutes prediciton horizon.

python3 train.py --patient 540 --missing_len 6 --single_pred
python3 train.py --patient 540 --missing_len 6

Citing

@inproceedings{cui2021personalised,
      author       = {Cui, Ran and Hettiarachchi, Chirath and Nolan, Christopher J and Daskalaki, Elena and Suominen, Hanna},
      title        = {Personalised Short-Term Glucose Prediction via Recurrent Self-Attention Network},
      booktitle    = {2021 IEEE 34th International Symposium on Computer-Based Medical Systems (CBMS)},
      year         = {2021},
      organization = {IEEE}
}

License

MIT