This project has retired. For details please refer to its Attic page.

Overview

This Recommendation Engine Template has integrated Apache Spark MLlib's Collaborative Filtering algorithm by default. You can customize it easily to fit your specific needs.

We are going to show you how to create your own recommendation engine for production use based on this template.

Usage

Event Data Requirements

By default, the template requires the following events to be collected:

  • user 'rate' item events
  • user 'buy' item events

You can customize this engine to use other events.

Input Query

  • user ID
  • number of recommended items

Output PredictedResult

  • a ranked list of recommended itemIDs

1. Install and Run PredictionIO

First you need to install PredictionIO 0.14.0 (if you haven't done it).

Let's say you have installed PredictionIO at /home/yourname/PredictionIO/. For convenience, add PredictionIO's binary command path to your PATH, i.e. /home/yourname/PredictionIO/bin:

1
$ PATH=$PATH:/home/yourname/PredictionIO/bin; export PATH

If you launched PredictionIO AWS instance, the path is located at /opt/PredictionIO/bin.

Once you have completed the installation process, please make sure all the components (PredictionIO Event Server, Elasticsearch, and HBase) are up and running.

If you launched PredictionIO AWS instance, you can skip pio-start-all. All components should have been started automatically.

If you are using PostgreSQL or MySQL, run the following to start PredictionIO Event Server:

1
$ pio eventserver &

If instead you are running HBase and Elasticsearch, run the following to start all PredictionIO Event Server, HBase, and Elasticsearch:

1
$ pio-start-all

You can check the status by running:

1
$ pio status

If everything is OK, you should see the following outputs:

1
2
3
4
...

(sleeping 5 seconds for all messages to show up...)
Your system is all ready to go.

To further troubleshoot, please see FAQ - Using PredictionIO.

2. Create a new Engine from an Engine Template

Now let's create a new engine called MyRecommendation by downloading the Recommendation Engine Template. Go to a directory where you want to put your engine and run the following:

1
2
$ git clone https://github.com/apache/predictionio-template-recommender.git MyRecommendation
$ cd MyRecommendation

A new directory MyRecommendation is created, where you can find the downloaded engine template.

3. Generate an App ID and Access Key

You will need to create a new App in PredictionIO to store all the data of your app. The data collected will be used for machine learning modeling.

Let's assume you want to use this engine in an application named "MyApp1". Run the following to create a new app "MyApp1":

1
$ pio app new MyApp1

You should find the following in the console output:

1
2
3
4
5
6
...
[INFO] [App$] Initialized Event Store for this app ID: 1.
[INFO] [App$] Created new app:
[INFO] [App$]       Name: MyApp1
[INFO] [App$]         ID: 1
[INFO] [App$] Access Key: 3mZWDzci2D5YsqAnqNnXH9SB6Rg3dsTBs8iHkK6X2i54IQsIZI1eEeQQyMfs7b3F

Note that App ID, **Access Key* are created for this App "MyApp1". You will need the Access Key when you collect data with EventServer for this App.

You can list all of the apps created its corresponding ID and Access Key by running the following command:

1
$ pio app list

You should see a list of apps created. For example:

1
2
3
4
[INFO] [App$]                 Name |   ID |                                                       Access Key | Allowed Event(s)
[INFO] [App$]               MyApp1 |    1 | 3mZWDzci2D5YsqAnqNnXH9SB6Rg3dsTBs8iHkK6X2i54IQsIZI1eEeQQyMfs7b3F | (all)
[INFO] [App$]               MyApp2 |    2 | io5lz6Eg4m3Xe4JZTBFE13GMAf1dhFl6ZteuJfrO84XpdOz9wRCrDU44EUaYuXq5 | (all)
[INFO] [App$] Finished listing 2 app(s).

4. Collecting Data

Next, let's collect some training data. By default, the Recommendation Engine Template supports 2 types of events: rate and buy. A user can give a rating score to an item or buy an item. This template requires user-view-item and user-buy-item events.

This template can easily be customized to consider more user events such as like, dislike, etc.

You can send these events to PredictionIO Event Server in real-time easily by making a HTTP request or through the provided SDK. Please see App Integration Overview for more details how to integrate your app with SDK.

Let's try sending events to EventServer with the following curl commands (The corresponding SDK code is showed in other tabs).

Replace <ACCCESS_KEY> by the Access Key generated in above steps. Note that localhost:7070 is the default URL of the Event Server.

For convenience, set your access key to the shell variable, run:

$ ACCESS_KEY=<ACCESS_KEY>

Example rate event

A user (ID "u0") gives an item (ID "i0") a rating of 5 at 2014-11-02T09:39:45.618-08:00 (current time will be used if eventTime is not specified)

Run the following curl command to send the rate event:

1
2
3
4
5
6
7
8
9
10
11
12
13
$ curl -i -X POST http://localhost:7070/events.json?accessKey=$ACCESS_KEY \
-H "Content-Type: application/json" \
-d '{
  "event" : "rate",
  "entityType" : "user",
  "entityId" : "u0",
  "targetEntityType" : "item",
  "targetEntityId" : "i0",
  "properties" : {
    "rating" : 5
  }
  "eventTime" : "2014-11-02T09:39:45.618-08:00"
}'
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
import predictionio

client = predictionio.EventClient(
    access_key=<ACCESS KEY>,
    url=<URL OF EVENTSERVER>,
    threads=5,
    qsize=500
)

# A user rates an item
client.create_event(
    event="rate",
    entity_type="user",
    entity_id=<USER ID>,
    target_entity_type="item",
    target_entity_id=<ITEM ID>,
    properties= { "rating" : float(<RATING>) }
)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
<?php
require_once("vendor/autoload.php");
use predictionio\EventClient;

$client = new EventClient(<ACCESS KEY>, <URL OF EVENTSERVER>);

// A user rates an item
$client->createEvent(array(
   'event' => 'rate',
   'entityType' => 'user',
   'entityId' => <USER ID>,
   'targetEntityType' => 'item',
   'targetEntityId' => <ITEM ID>,
   'properties' => array('rating'=> <RATING>)
));
?>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
# Create a client object.
client = PredictionIO::EventClient.new(<ACCESS KEY>, <URL OF EVENTSERVER>)

# A user rates an item.
client.create_event(
  'rate',
  'user',
  <USER ID>, {
    'targetEntityType' => 'item',
    'targetEntityId' => <ITEM ID>,
    'properties' => { 'rating' => <RATING (float)> }
  }
)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
import org.apache.predictionio.Event;
import org.apache.predictionio.EventClient;

EventClient client = new EventClient(<ACCESS KEY>, <URL OF EVENTSERVER>);

// A user rates an item
Event rateEvent = new Event()
    .event("rate")
    .entityType("user")
    .entityId(<USER_ID>)
    .targetEntityType("item")
    .targetEntityId(<ITEM_ID>)
    .property("rating", new Float(<RATING>));
client.createEvent(rateEvent);

Example buy event

A user (ID "u1") buys an item (ID "i2") at 2014-11-10T12:34:56.123-08:00 (current time will be used if eventTime is not specified)

Run the following curl command to send the buy event:

1
2
3
4
5
6
7
8
9
10
$ curl -i -X POST http://localhost:7070/events.json?accessKey=$ACCESS_KEY \
-H "Content-Type: application/json" \
-d '{
  "event" : "buy",
  "entityType" : "user",
  "entityId" : "u1",
  "targetEntityType" : "item",
  "targetEntityId" : "i2",
  "eventTime" : "2014-11-10T12:34:56.123-08:00"
}'
1
2
3
4
5
6
7
8
# A user buys an item
client.create_event(
    event="buy",
    entity_type="user",
    entity_id=<USER ID>,
    target_entity_type="item",
    target_entity_id=<ITEM ID>
)
1
2
3
4
5
6
7
8
9
10
<?php
// A user buys an item
$client->createEvent(array(
   'event' => 'buy',
   'entityType' => 'user',
   'entityId' => <USER ID>,
   'targetEntityType' => 'item',
   'targetEntityId' => <ITEM ID>
));
?>
1
2
3
4
5
6
7
8
9
# A user buys an item.
client.create_event(
  'buy',
  'user',
  <USER ID>, {
    'targetEntityType' => 'item',
    'targetEntityId' => <ITEM ID>
  }
)
1
2
3
4
5
6
7
8
// A user buys an item
Event buyEvent = new Event()
    .event("buy")
    .entityType("user")
    .entityId(<USER_ID>)
    .targetEntityType("item")
    .targetEntityId(<ITEM_ID>);
client.createEvent(buyEvent);

Query Event Server

Now let's query the EventServer and see if these events are imported successfully.

Go to following URL with your browser:

http://localhost:7070/events.json?accessKey=&lt;YOUR_ACCESS_KEY>

or run the following command in terminal:

1
$ curl -i -X GET "http://localhost:7070/events.json?accessKey=$ACCESS_KEY"

Note that you should quote the entire URL by using single or double quotes when you run the curl command.

It should return the imported events in JSON format. You can refer to Event Server Debugging Recipes for more different ways to query Event Server.

Import More Sample Data

This engine requires more data in order to train a useful model. Instead of sending more events one by one in real time, for quickstart demonstration purpose, we are going to use a script to import more events in batch.

A Python import script import_eventserver.py is provided in the template to import the data to the Event Server using the Python SDK. Please upgrade to the latest Python SDK.

First, you will need to install Python SDK in order to run the sample data import script. To install Python SDK, run:

1
$ pip install predictionio

or

1
$ easy_install predictionio

You may need sudo access if you have permission issue. (ie. sudo pip install predictionio)

Execute the following to import the data:

These commands must be executed in the Engine directory, for example: MyRecomendation.

1
2
3
$ cd MyRecommendation
$ curl https://raw.githubusercontent.com/apache/spark/master/data/mllib/sample_movielens_data.txt --create-dirs -o data/sample_movielens_data.txt
$ python data/import_eventserver.py --access_key $ACCESS_KEY

You should see the following output:

1
2
Importing data...
1501 events are imported.

Now the movie ratings data is stored as events inside the Event Store.

If you see error TypeError: init() got an unexpected keyword argument 'access_key', please update the Python SDK to the latest version.

You can query the event server again as described previously to check the imported events.

By default, the template trains the model with "rate" events (explicit rating). You can customize the engine to read other custom events and handle events of implicit preference (such as, view, buy)

5. Deploy the Engine as a Service

Now you can build, train, and deploy the engine. First, make sure you are under the MyRecommendation directory.

1
$ cd MyRecommendation

Engine.json

Under the directory, you should find an engine.json file; this is where you specify parameters for the engine.

Modify this file to make sure the appName parameter match your App Name you created earlier (e.g. "MyApp1" if you follow the quickstart).

1
2
3
4
5
6
7
  ...
  "datasource": {
    "params" : {
      "appName": "MyApp1"
    }
  },
  ...

You may see appId in engine.json instead, which means you are using old template. In this case, make sure the appId defined in the file match your App ID. Alternatively, you can download the latest version of the template or follow our upgrade instructions to modify the template to use appName as parameter.

Building

Start with building your MyRecommendation engine. Run the following command:

1
$ pio build --verbose

This command should take few minutes for the first time; all subsequent builds should be less than a minute. You can also run it without --verbose if you don't want to see all the log messages.

Upon successful build, you should see a console message similar to the following.

1
[INFO] [Console$] Your engine is ready for training.

Training the Predictive Model

To train your engine, run the following command:

1
$ pio train

When your engine is trained successfully, you should see a console message similar to the following.

1
[INFO] [CoreWorkflow$] Training completed successfully.

Deploying the Engine

Now your engine is ready to deploy. Run:

1
$ pio deploy

When the engine is deployed successfully and running, you should see a console message similar to the following:

1
2
[INFO] [HttpListener] Bound to /0.0.0.0:8000
[INFO] [MasterActor] Bind successful. Ready to serve.

Do not kill the deployed engine process.

By default, the deployed engine binds to http://localhost:8000. You can visit that page in your web browser to check its status.

Engine Status

6. Use the Engine

Now, you can try to retrieve predicted results. To recommend 4 movies to a user whose id is 1, you send this JSON { "user": "1", "num": 4 } to the deployed engine and it will return a JSON result of the recommended movies. Simply send a query by making an HTTP request or through the EngineClient of an SDK.

With the deployed engine running, open another terminal and run the following curl command or use an SDK to send the query:

1
2
3
$ curl -H "Content-Type: application/json" \
-d '{ "user": "1", "num": 4 }' http://localhost:8000/queries.json

1
2
3
import predictionio
engine_client = predictionio.EngineClient(url="http://localhost:8000")
print engine_client.send_query({"user": "1", "num": 4})
1
2
3
4
5
6
7
8
9
10
<?php
require_once("vendor/autoload.php");
use predictionio\EngineClient;

$client = new EngineClient('http://localhost:8000');

$response = $client->sendQuery(array('user'=> 1, 'num'=> 4));
print_r($response);

?>
1
2
3
4
5
6
7
# Create client object.
client = PredictionIO::EngineClient.new(<ENGINE DEPLOY URL>)

# Query PredictionIO.
response = client.send_query('user' => <USER ID>, 'num' => <NUMBER (integer)>)

puts response
1
2
3
4
5
6
7
8
9
10
11
12
13
import com.google.common.collect.ImmutableMap;
import com.google.gson.JsonObject;

import org.apache.predictionio.EngineClient;

// create client object
EngineClient engineClient = new EngineClient(<ENGINE DEPLOY URL>);

// query
JsonObject response = engineClient.sendQuery(ImmutableMap.<String, Object>of(
        "user", "1",
        "num",  4
    ));

The following is sample JSON response:

1
2
3
4
5
6
7
8
{
  "itemScores":[
    {"item":"22","score":4.072304374729956},
    {"item":"62","score":4.058482414005789},
    {"item":"75","score":4.046063009943821},
    {"item":"68","score":3.8153661512945325}
  ]
}

Congratulations, MyRecommendation is now running!

To update the model periodically with new data, simply set up a cron job to call pio train and pio deploy. The engine will continue to serve prediction results during the re-train process. After the training is completed, pio deploy will automatically shutdown the existing engine server and bring up a new process on the same port.

Note that if you import a large data set and the training seems to be taking forever or getting stuck, it's likely that there is not enough executor memory. It's recommended to setup a Spark standalone cluster, you'll need to specify more driver and executor memory when training with a large data set. Please see FAQ here for instructions.

Next: DASE Components Explained