original paper : https://arxiv.org/abs/1703.04046
original implementation by the authors : https://github.com/akaraspt/deepsleepnet
I followed implementation of the deepsleepnet in tensorflow into keras (actually keras in tensorflow).
conda create -n deepsleepnet_keras environment --file requirements.txt
Download and getting NPZ file is totally same with original implementation from https://github.com/akaraspt/deepsleepnet .
To get 20 fold cross validation data:
python data_preparation.py
python trainer.py
The trained model (both featurenet and deepsleepnet) will be stored at the ./weights as weights.
cat performance.txt
- Revise 'data preparation' to split patients. For example, in test set, patients 0 - 10, valid set patients 11 - 12, test set 13 - 14.