Wearable based monitoring and self-supervised contrastive learning detect clinical complications during treatment of Hematologic malignancies

1. Data Preparation

Let's start with the data preparation.

By runing data_alignment.py, we perform a crucial data alignment process that synchronizes the time series with real daytime and efficiently divides them into one-hour intervals. This data alignment step ensures that the time series data is organized consistently, enabling easier and more accurate analysis and interpretation.

To read the data and split them to training and test sets, run the following code, train_test_split.ipynb, which generates following files in .pickle format.

train_set.pickle
test_set.pickle
ood_set.pickle

2. Training

To train the self-supervised model, run the following command

python train.py --epochs 500 --batch_size 128 --lr 0.001 --hdim 128 --input_channel 512 --temperature 0.07 --window_size 1000

3. Evaluation

To evaluate the trained model, run the following command

python evaluation.py --model_path <path to checkpoint> --data_path <path to data>

Citation

@article{jacobsen2023wearable,
  title={Wearable based monitoring and self-supervised contrastive learning detect clinical complications during treatment of Hematologic malignancies},
  author={Jacobsen, Malte and Gholamipoor, Rahil and Dembek, Till A and Rottmann, Pauline and Verket, Marlo and Brandts, Julia and J{\"a}ger, Paul and Baermann, Ben-Niklas and Kondakci, Mustafa and Heinemann, Lutz and others},
  journal={npj Digital Medicine},
  volume={6},
  number={1},
  pages={105},
  year={2023},
  publisher={Nature Publishing Group UK London}
}