EHR DREAM Challenge. ohdsiDream Team

More information is available here: https://forums.ohdsi.org/t/ehr-dream-challenge-patient-mortality/8028/27

Installation

Download Synthetic Data

Install Python3 and pip (in case you don't have it)

https://wiki.python.org/moin/BeginnersGuide/Download

https://pip.pypa.io/en/stable/installing/

Install python dependencies:

cd <path-to-repo-directory>/data
pip install -r requirements.txt

Download and uncompress synthetic data:

cd <path-to-repo-directory>/data
python download-synthetic-data.py \
       --synapse_username  <your-synapse-login>  \
       --synapse_password <your-synapse-password>

download-synthetic-data.py downloads data from synapse and uncompresses it and creates directories for Docker volumes. You should run it only once on your machine.

Data description is available here https://www.synapse.org/#!Synapse:syn18405991/wiki/595493

Install docker on your machine

For Mac: https://docs.docker.com/docker-for-mac/install/

For Windows https://docs.docker.com/docker-for-windows/install/

Train and evaluate model on your machine using synthetic data

Note: Currenty we are training dummy model to make sure everything works correctly and we can do submissions for the challenge.

Information about requiered docker container and submission is avalable here: https://www.synapse.org/#!Synapse:syn18405991/wiki/595495

Build and run docker container on your machine

You can build docker container with this command:

cd <path-to-repo-directory>
docker build -t docker.synapse.org/syn20822908/dummy-model app

You can train the model with this command:

cd <path-to-repo-directory>
docker run -v $(pwd)/data/syn20685954/training:/train:ro \
           -v $(pwd)/data/scratch:/scratch:rw \
           -v $(pwd)/data/model:/model:rw \
           docker.synapse.org/syn20822908/dummy-model \
           bash /app/train.sh

You can evaluate the model and create file with predictions after train it with this command:

cd <path-to-repo-directory>
docker run -v $(pwd)/data/syn20685954/evaluation:/infer:ro \
           -v $(pwd)/data/scratch:/scratch:rw \
           -v $(pwd)/data/output:/output:rw \
           -v $(pwd)/data/model:/model:rw \
           docker.synapse.org/syn20822908/dummy-model \
           bash /app/infer.sh

Note: Whenever you make any changes in the code in "app" directory you should re-build docker container again. You can use a script local_run.sh to run "build", "train" and "evaluate" as a single command.

All docker volumes are available in "data" directory after run

Upload Docker Image to synapse

Make sure everything works correctly on your machine.

Answer the questions after the following command:

docker login docker.synapse.org

You may now push your docker image to synapse

docker push docker.synapse.org/syn20822908/dummy-model

Verify the Docker image was successfully pushed(optional):

https://www.synapse.org/#!Synapse:syn20822908/docker

Submit docker image:

1. Open https://www.synapse.org/#!Synapse:syn20822908/docker/
2. Select docker.synapse.org/syn20822908/dummy-model
3. In right upper conner select "Docker repository tools"
4. Select "Submit Docker Repository to Challenge"
5. Select tag
6. Select "EHR DREAM Challenge Fast Lane"

Synapse Limitations

• Up to 5 successful submissions / team / day
• Max 1h runtime / submission
• 30 GB of memory per model
• 4 vCPU cores @ 2400 MHz