Code for (Yoo, Jaesung, et al. "Residual one-dimensional convolutional neural network for neuromuscular disorder classification from needle electromyography signals with explainability." Computer Methods and Programs in Biomedicine 226 (2022): 107079.)
Powered by hydra, hideandseek, and tools-jsyoo61
Recommend generating toy data to check format
python preprocess.py toy=true
To use your own data, put the data in data/raw/ directory.
Data should be:
- Signal files
cohort.csvindicating labels for each patientcohort_signal.csvindicating labels for each signal
Training is done in 2 steps:
- nEMGNet training
python train1.py
To specify specific run directory:
python train1.py hydra.run.dir=your_wanted_dir
For a sweep run
python train1.py -m "random.seed=range(0,5)" train.update.lr=1e-3,1e-4 train.update.batch_size=64,128 hydra.sweep.dir=your_sweep_dir
- Classifier training
Use patient feature without muscle type info
python train2.py feature_type=all dir.train1=your_wanted_dir
Use patient feature with muscle type info
python train2.py feature_type=PD dir.train1=your_wanted_dir
Sweep over train1 sweep directories
python train2.py feature_type=all "dir.train1=your_wanted_dir/${subdir}" "subdir=range(0,20)"
Tbd