Hospital Readmission Model

This is a model for predicting hospital readmission among patients with diabetes. Data from UCI: https://archive.ics.uci.edu/ml/datasets/diabetes+130-us+hospitals+for+years+1999-2008

Model is built with Sklearn. Python environment specified in requirements.txt. I also created a Docker image which packages the .pkl as an executable.

Model	AUC
LR+RF+XGB Stack	0.6990824552912449
LR+RF+XGB Avg	0.6981398497127431
XGB	0.6956653497449965
RF	0.6952079165690574
LR	0.684611003872049

Build model (output is model.pkl):

python model.py

Build container for generating predictions with model:

gcloud container builds submit --gcs-source-staging-dir=gs://djr-data/cloudbuild \
    --async --timeout 4h0m0s --config cloudbuild.yaml .

NOTE: Container is built asyncronously with Google Container Builder and stored in Google Container Repository.

Generate predictions with container:

export PROJECT_ID=$(gcloud config get-value project -q)
export IMAGE_ID=gcr.io/${PROJECT_ID}/hospital-readmissions:latest
gcloud docker -- pull ${IMAGE_ID}
docker run -v $(pwd)/data:/usr/share/model/data ${IMAGE_ID} --input diabetic_data.csv

Shelling into container for testing:

docker run -it -v $(pwd)/data:/usr/share/model/data --entrypoint=ash ${IMAGE_ID}