Merge pull request kubeflow#75 from animeshsingh/sklearn-server

Sklearn server

docs/samples/sklearn/README.md

@@ -0,0 +1,35 @@
+## Creating your own model and testing the SKLearn server.
+
+To test the Scikit-learn Server, first we need to generate a simple scikit-learn model using Python. Sample model is also added in this directory to reuse if needed.
+
+```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 run the Scikit-learn Server using the generated model and test for prediction. Models can be on local filesystem, S3 compatible object storage or Google Cloud Storage.
+
+```shell
+python -m sklearnserver --model_dir model.joblib --model_name svm
+```
+
+We can also use the inbuilt sklearn support for sample datasets and do some simple predictions
+
+```python
+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/models/svm:predict', json=formData)
+print(res)
+print(res.text)
+```

python/sklearnserver/README.md

@@ -0,0 +1,38 @@
+# Scikit-Learn Server
+
+[Scikit-Learn](https://scikit-learn.org/stable/) server is an implementation of KFServing for serving Scikit-learn models, and provides a Scikit-learn model implementation for prediction, pre and post processing. 
+
+To start the server locally for development needs, run the following command under this folder in your github repository. 
+
+```
+pip install -e .
+```
+
+The following output indicates a successful install.
+
+```
+Obtaining file://kfserving/python/sklearn
+Requirement already satisfied: kfserver==0.1.0 in /Users/username/Desktop/kfserving/python/kfserving (from sklearnserver==0.1.0) (0.1.0)
+Requirement already satisfied: scikit-learn==0.20.3 in /anaconda3/lib/python3.6/site-packages (from sklearnserver==0.1.0) (0.20.3)
+Requirement already satisfied: argparse>=1.4.0 in /anaconda3/lib/python3.6/site-packages (from sklearnserver==0.1.0) (1.4.0)
+Requirement already satisfied: numpy>=1.8.2 in /anaconda3/lib/python3.6/site-packages (from sklearnserver==0.1.0) (1.16.2)
+Requirement already satisfied: tornado>=1.4.1 in /anaconda3/lib/python3.6/site-packages (from kfserver==0.1.0->sklearnserver==0.1.0) (5.0.2)
+Requirement already satisfied: scipy>=0.13.3 in /anaconda3/lib/python3.6/site-packages (from scikit-learn==0.20.3->sklearnserver==0.1.0) (1.1.0)
+Installing collected packages: sklearnserver
+  Found existing installation: sklearnserver 0.1.0
+    Uninstalling sklearnserver-0.1.0:
+      Successfully uninstalled sklearnserver-0.1.0
+  Running setup.py develop for sklearnserver
+Successfully installed sklearnserver
+```
+
+Once Scikit-learn server is up and running, you can check for successful installation by running the following command
+
+```
+python3 -m sklearnserver
+usage: __main__.py [-h] [--http_port HTTP_PORT] [--grpc_port GRPC_PORT]
+                   --model_dir MODEL_DIR [--model_name MODEL_NAME]
+__main__.py: error: the following arguments are required: --model_dir
+```
+
+You can now point to your `joblib` model file and use the server to load the model and test for prediction. Models can be on local filesystem, S3 compatible object storage or Google Cloud Storage. Please follow [this sample](https://github.com/kubeflow/kfserving/tree/master/docs/samples/sklearn) to test your server by generating your own model. 

python/sklearnserver/setup.py

@@ -0,0 +1,20 @@
+from setuptools import setup, find_packages
+
+setup(
+    name='sklearnserver',
+    version='0.1.0',
+    author_email='singhan@us.ibm.com',
+    license='https://github.com/kubeflow/kfserving/LICENSE',
+    url='https://github.com/kubeflow/kfserving/python/sklearnserver',
+    description='Model Server implementation for scikit-learn. Not intended for use outside KFServing Frameworks Images',
+    long_description=open('README.md').read(),
+    python_requires='>3.4',
+    packages=find_packages("sklearnserver"),
+    install_requires=[
+        "kfserver==0.1.0",
+        "scikit-learn == 0.20.3",
+        "argparse >= 1.4.0",
+        "numpy >= 1.8.2",
+        "joblib >= 0.13.0"
+    ],
+)

python/sklearnserver/sklearnserver/__init__.py

@@ -0,0 +1 @@
+from .model import SKLearnModel

python/sklearnserver/sklearnserver/__main__.py

@@ -0,0 +1,20 @@
+import kfserving
+import argparse
+
+
+from sklearnserver import SKLearnModel
+
+DEFAULT_MODEL_NAME = "model"
+DEFAULT_LOCAL_MODEL_DIR = "/tmp/model"
+
+parser = argparse.ArgumentParser(parents=[kfserving.server.parser])
+parser.add_argument('--model_dir', required=True,
+                    help='A URI pointer to the model binary')
+parser.add_argument('--model_name', default=DEFAULT_MODEL_NAME,
+                    help='The name that the model is served under.')
+args, _ = parser.parse_known_args()
+
+if __name__ == "__main__":
+    model = SKLearnModel(args.model_name, args.model_dir)
+    model.load()
+    kfserving.KFServer().start([model])

python/sklearnserver/sklearnserver/model.py

@@ -0,0 +1,27 @@
+import kfserving
+import joblib 
+import numpy as np
+import os
+
+class SKLearnModel(kfserving.KFModel):
+    def __init__(self, name, model_dir):
+        self.name = name
+        self.model_dir = model_dir
+        self.ready = False
+
+    def load(self):
+        model_file = kfserving.Storage.download(self.model_dir)
+        self._joblib = joblib.load(model_file)
+        self.ready = True
+
+    def predict(self, inputs):
+        try:
+            inputs = np.array(inputs)
+        except Exception as e:
+            raise Exception(
+                "Failed to initialize NumPy array from inputs: %s, %s" % (e, inputs))
+        try:
+            result = self._joblib.predict(inputs).tolist()
+            return result
+        except Exception as e:
+            raise Exception("Failed to predict %s" % e)