sklearn

Creating your own model and testing the SKLearn Server locally.

To test the Scikit-Learn server, first we need to generate a simple scikit-learn model using Python.

from sklearn import svm
from sklearn import datasets
from joblib import dump
clf = svm.SVC(gamma='scale')
iris = datasets.load_iris()
X, y = iris.data, iris.target
clf.fit(X, y)
dump(clf, 'model.joblib')

Then, we can install and run the SKLearn Server using the generated model and test for prediction. Models can be on local filesystem, S3 compatible object storage, Azure Blob Storage, or Google Cloud Storage.

# we should indicate the directory containing the model file (model.joblib) by --model_dir
python -m sklearnserver --model_dir ./  --model_name svm

We can also use the inbuilt sklearn support for sample datasets and do some simple predictions

from sklearn import datasets
import requests
iris = datasets.load_iris()
X, y = iris.data, iris.target
formData = {
    'instances': X[0:1].tolist()
}
res = requests.post('http://localhost:8080/v1/models/svm:predict', json=formData)
print(res)
print(res.text)

Predict on a InferenceService using SKLearnServer

Setup

1. Your ~/.kube/config should point to a cluster with KFServing installed.
2. Your cluster's Istio Ingress gateway must be network accessible.

Create the InferenceService

Apply the CRD

kubectl apply -f sklearn.yaml

Expected Output

$ inferenceservice.serving.kubeflow.org/sklearn-iris created

Run a prediction

The first step is to determine the ingress IP and ports and set INGRESS_HOST and INGRESS_PORT

MODEL_NAME=sklearn-iris
INPUT_PATH=@./iris-input.json
SERVICE_HOSTNAME=$(kubectl get inferenceservice sklearn-iris -o jsonpath='{.status.url}' | cut -d "/" -f 3)
curl -v -H "Host: ${SERVICE_HOSTNAME}" http://${INGRESS_HOST}:${INGRESS_PORT}/v1/models/$MODEL_NAME:predict -d $INPUT_PATH

Expected Output

*   Trying 169.63.251.68...
* TCP_NODELAY set
* Connected to 169.63.251.68 (169.63.251.68) port 80 (#0)
> POST /models/sklearn-iris:predict HTTP/1.1
> Host: sklearn-iris.default.svc.cluster.local
> User-Agent: curl/7.60.0
> Accept: */*
> Content-Length: 76
> Content-Type: application/x-www-form-urlencoded
>
* upload completely sent off: 76 out of 76 bytes
< HTTP/1.1 200 OK
< content-length: 23
< content-type: application/json; charset=UTF-8
< date: Mon, 20 May 2019 20:49:02 GMT
< server: istio-envoy
< x-envoy-upstream-service-time: 1943
<
* Connection #0 to host 169.63.251.68 left intact
{"predictions": [1, 1]}

Run SKLearn InferenceService with your own image

Since the KFServing SKLearnServer image is built from a specific version of scikit-learn pip package, sometimes it might not be compatible with the pickled model you saved from your training environment, however you can build your own SKLearnServer image following this instruction.

To use your SKLearnServer image:

• Add the image to the KFServing configmap

        "sklearn": {
            "image": "<your-dockerhub-id>/kfserving/sklearnserver",
        },

• Specify the runtimeVersion on InferenceService spec

apiVersion: "serving.kubeflow.org/v1alpha2"
kind: "InferenceService"
metadata:
  name: "sklearn-iris"
spec:
  default:
    predictor:
      sklearn:
        storageUri: "gs://kfserving-samples/models/sklearn/iris"
        runtimeVersion: X.X.X