Chronic kidney dataset https://www.kaggle.com/mansoordaku/ckdisease
Clipper is a low-latency prediction serving system for machine learning. Clipper makes it simple to integrate machine learning into user-facing serving systems.
This tutorial demonstrates how to train a simple Gradient Boosting Classifier model to predict whether a patient has Chronic Kidney Disease (CKD) or not. A predict function is then deployed using the clipper
DockerContainerManager in the clipper_docker.py script and the KubernetesContainerManager in the clipper_kubernetes.py script.
Basically you can just run each script to see how clipper works with the DockerContainerManager and the KubernetesContainerManager.
Basically these Container Manager creates a docker container with the Gradient Boosting model. The KubernetesContainerManager deploys the docker container on Kubernetes. Since the model is containerized in a Docker container one can post data to a API endpoint and this will return predictions for that data.
For the KubernetesContainerManager make sure you first run a proxy by running
kubectl proxy --port 8080
In addition you must make sure you authenticate Kubernetes to your private docker registry. For example
kubectl create secret docker-registry myregistrykey \
--docker-server=yourregistrydomain.io \
--docker-username=yourusername \
--docker-password=yourpassword \
[email protected]
Then to patch the kubernetes service account with your key run
kubectl patch serviceaccount default -p ‘{“imagePullSecrets”: [{“name”: “myregistrykey”}]}’
Now you should be ready to deploy your kubernetes pod with the model.
To check that all the pods are up and running run
kubectl get pods
To find the clipper query front end you can run
kubectl describe service query-frontend
and look for the line saying Endpoints
For example using Python to post to the api address endpoint looks like this
headers = {"Content-type": "application/json"}
addr = clipper_conn.get_query_addr()
response =requests.post("http://%s/%s/predict" % (addr, 'kddtutorial'), headers=headers,
data=json.dumps({"input": list(inputs)})).json()
print(response)
We then get a prediction response in the response variable.