Getting Started with Predictive Services

In this section we will walk through an end-to-end example of deploying and using a Predictive Service. For more information about each aspect see the subsequent chapters of the User Guide.

Configuration

In order to launch a Predictive Service in EC2 we first need to configure the graphlab.deploy.Ec2Config object, which contains required configuration parameters.

import graphlab

ec2 = graphlab.deploy.Ec2Config(region='us-west-2',
                                instance_type='m3.xlarge',
                                aws_access_key_id='YOUR_ACCESS_KEY',
                                aws_secret_access_key='YOUR_SECRET_KEY')

The configuration object is client-side only; its purpose is merely to encapsulate a set of parameters and pass them to the create command.

Launch the Predictive Service

Having configured the Predictive Service environment, we are ready to launch a Predictive Service Deployment using graphlab.deploy.predictive_service.create. Required parameters include

  1. a name for the deployment
  2. a EC2 configuration we defined in the previous step
  3. an state path for the root 'working directory' for this Predictive Service

When the create() command is executed, the EC2 instances will be launched immediately, followed by a load balancer which adds the instances into the cluster as they pass health checks.

deployment = graphlab.deploy.predictive_service.create(
    'first', ec2, 's3://sample-testing/first')

Additional parameters include the number of EC2 nodes to use for the service, SSL credentials to secure the data flow, and more. See graphlab.deploy.predictive_service.create for a complete list.

Upload a Model

Any model that you create and use locally can back a Predictive Service. We will use a simple recommender model in this walk-through:

data_url = 'https://s3.amazonaws.com/dato-datasets/movie_ratings/sample.small'
data = graphlab.SFrame.read_csv(data_url,delimiter='\t',column_type_hints={'rating':int})
model = graphlab.popularity_recommender.create(data, 'user', 'movie', 'rating')

The PredictiveService.add method stages a model for deployment to the cluster. It also requires us to provide a name which we will later use to query the model.

deployment.add('recs', model)

The change needs to be applied in order for the model to be actually uploaded to the service:

deployment.apply_changes()

Printing the deployment object displays its properties and deployed Predictive Objects:

print deployment

Name                  : first
State Path            : s3://roman-workspace/roman-ugtest
Description           : None
API Key               : 7a99ccbf-3f51-4c5a-bf32-c03a6f07ecd2
CORS origin           :
Global Cache State    : enabled
Load Balancer DNS Name: first-1793598482.us-west-2.elb.amazonaws.com

Deployed predictive objects:
 name: recs, version: 1, cache: enabled, description:
No Pending changes.

At this point the model is ready to be queried.

Query the Model

Each model (or Predictive Object) in a Predictive Service exposes a REST endpoint to query it. GraphLab Create provides wrapper methods to submit queries within its Python API.

recs = deployment.query('recs',
                        method='recommend',
                        data={ 'users': [ 'Jacob Smith' ] })

Outside of the Python client, an application can use the endpoint directly, for instance through curl (replace the api_key and URL with your parameters as displayed by the print deployment command above):

curl -X POST -d '{"api_key": "7a99ccbf-3f51-4c5a-bf32-c03a6f07ecd2",
                  "data": {
                    "method": "recommend",
                    "data": { "users": [ "Jacob Smith" ] }
                    }
                  }'
     http://first-1793598482.us-west-2.elb.amazonaws.com/query/recs

Shutdown the Predictive Service

As long as the Predictive Service is up and running, it incurs AWS charges. You can shut down a service as follows:

deployment.terminate_service()

This will terminate all nodes used by the Predictive Service.