This folder has all the necessary code for experiments reported in the paper. We have also added the configurations for experiments and set the default values such that the results will either exactly reproduce the experiments and/or produce very similar results.
The main file is regression.py and it is used to perform experiments to predict Heart Rate(HR), Respiratory Rate (RR) and Oxygen Saturation level (SpO2). Below is a sample command to run one such experiment. Note that chaning the configs will run the different variants of RR, HR and SpO2.
python regression.py --config_path ./configs/regression/RR.json
The main file for performing classification of Eigenworms is present in classification.py. The two different configs provided in this case will make use of BCR_DE and noisy version of BCR_DE introduced in the paper. Sample command:
python classification.py --config_path ./configs/classification/eigenworm.json
Main file for autoencoding time series sequence of ECG/PPG is present in autoencode.py
python autoencode.py --dataset_type ECG
For denoising autoencoding the main file is denoise_AE.py. One can control the noise sensitivity from the parameters for both ECG and PPG sequence.
python denoise_AE.py --dataset_type ECG
In order to perform masked reconstruction, the main file is masked_AE.py. As for other sequence to sequence tasks presented here, there is a choice for ECG/PPG data.
python masked_AE.py --dataset_type ECG
One of the key capabilities of BCR_DE as highlighted in the paper is that of modelling the dynamics in a coupled differntial equation system. The main file for this case is coupled.py. We present several such scenarios, the configurations of which are provided and can be used as follows:
python coupled.py --config_path ./configs/coupled_diffeq/HH.json
Please note that use of GPU is advised to achieve significant speedup in computation. However, the code also can be run on CPU.