The Smart Alarm 2.0 project aims to suppress false, invalid SpO2 low alarms emitted by pulse oximetry machines. Large SpO2 alarm datasets with labels of whether an alarm should be suppressed or not do not exist and are expensive to create. Hence we are using Snorkel to programmatically label datasets using several rules based on domain knowledge.

Clone this repository.

git clone https://github.com/sfpugh/smart-alarm-learning-labels.git

Create a virtual environment using Python3 virtualenv and add it to Jupyter Notebook.

pip3 install --upgrade virtualenv
virtualenv -p python3 <env_name>
source <env_name>/bin/activate

pip install -r requirements.txt

python -m ipykernel install --user --name=<env_name>

deactivate

Unzip the data to avoid having to re-build it from scratch.

cd data/
unzip data.zip

Jupyter notebook "learn_labels_STABLE.ipynb" is the notebook from which you should call all other experiments in "./Experiments". You can open the notebook with the following command.

jupyter notebook learn_labels_STABLE.ipynb

Then activate the virtual environment created in the installation steps by navigating to the 'Kernel' tab in the toolbar, then hover on 'Change kernel', then select the proper environment. Edit the last cell in the notebook to run the experiment of interest as follows, then run all cells.

%run ./Experiments/<E#-description>.ipynb

• Snorkel https://www.snorkel.org
• Matrix profile https://github.com/target/matrixprofile-ts