This is the code accompanying the following paper:
Shen, Yanting, et al. "AutoNet-Generated Deep Layer-Wise Convex Networks for ECG Classification." IEEE Transactions on Pattern Analysis and Machine Intelligence (2024).
Clone the repository
git clone https://github.com/Arfea/ecg.git
If you don't have virtualenv, install it with
pip install virtualenv
Make and activate a new Python 2.7 environment
virtualenv -p python2.7 ecg_env
source ecg_env/bin/activate
Install the requirements (this may take a few minutes).
For CPU only support run
./setup.sh
To install with GPU support run
env TF=gpu ./setup.sh
For illustration purpose, we demonstrate the LCN benchmarked with ResNet-based model by Hannun et al. on the International Conference on Biomedical Engineering and Biotechnology (ICBEB) 2018 Challenge.
You can also download the datasets by running the relevant cells in the notebook (see below).
Run AutoNet-LCN/ecg/icbeb.ipynb cell by cell. Be careful that the command build_json() should only be run once and the kernel needs to be restarted before running the subsequent codes, to avoid out-of-memory errors.
So far the autonet(params) function cannot be directly called in the notebook due to memory constraints, as the kernel needs to be restarted every time train_hannun_model(params), train_LCN(params), evaluate_hannun(params), or evaluate(params) is called. Instead, one needs to adjust the hyperparameters (repeat, skip, bn) manually according to the autonet(params) algorithm.