Classification Engine Template with Custom Attributes and Random Forest Algorithm

This example engine is based on Classification Tempplate version v0.1.1 and is modified to use Random Forest algorithm and demonstrates how to use custom attributes for classification.

Classification template

Please refer to http://docs.prediction.io/templates/classification/quickstart/

Development Notes

Example Use Case

predict the service plan (plan) a user will subscribe to based on his 3 properties: Gender, Age and Education

Input Data:

Prepared below input data in the format “Plan, Gender, Age, Education”

0,Male,30,College 0,Female,29,College 1,Male,30,High School 1,Female,28,High School 2,Male,35,No School 2,Male,40,No School 2,Female,25,No School 0,Male,40,College

Input data is available in data/sample_random_forest_data.txt

Importing Input data

Updated “import_eventserver.py” withe below changes

1. added “sample_random_forest_data.txt” as file

parser.add_argument('--file', default="./data/sample_random_forest_data.txt")

2. Creating events with features.

	client.create_event(
    event="$set",
    entity_type="user",
    entity_id=str(count), # use the count num as user ID
    properties= {
      "gender" : data[1],
      "age" : int(data[2]),
      "education" : data[3],
      "plan" : int(data[0])
    }
  )

Changes to engine.json

In the engine.json, removed “naive” algorithm and added “randomforest” algorithm.

"algorithms": [
    {
      "name": "randomforest",
      "params": {
        "numClasses": 3,
        "numTrees": 5,
        "featureSubsetStrategy": "auto",
        "impurity": "gini",
        "maxDepth": 4,
        "maxBins": 100
      }
    }
  ]

Changes to DataSource.scala

1. In the readTraining method, defined below maps

val gendersMap = Map("Male" -> 0.0, "Female" -> 1.0)
val educationMap = Map("No School" -> 0.0,"High School" -> 1.0,"College" -> 2.0)

Then encoded the categorical features values using map.

 LabeledPoint(properties.get[Double]("plan"),
            Vectors.dense(Array(
              gendersMap(properties.get[String]("gender")),
              properties.get[Double]("age"),
              educationMap(properties.get[String]("education"))
            ))
          )

2. Added gendersMap and educationMap to the TrainingData class

class TrainingData(
  val labeledPoints: RDD[LabeledPoint],
  val gendersMap: Map[String,Double],
  val educationMap: Map[String,Double]
) extends Serializable

readTraining returns below:

	 new TrainingData(labeledPoints,
   	     gendersMap,
        educationMap)

Changes to Engine.scala

In the Engine.scala, replaced “naive” algorithm with “randomforest” algorithm

 Map("randomforest" -> classOf[RandomForestAlgorithm]),

Updated Query.sclaa to include attributes

class Query(
 val  gender: String,
 val  age: Int,
 val  education: String
) extends Serializable

Changes to Preparator.scala

added attributes to PreparedData

	class PreparedData(
 		 val labeledPoints: RDD[LabeledPoint],
  val gendersMap: Map[String,Double],
  val educationMap: Map[String,Double]
) extends Serializable

new PreparedData(trainingData.labeledPoints,trainingData.gendersMap,trainingData.educationMap)

Created RandomForestAlgorithm.scala

created new model class

class PIORandomForestModel(
  val gendersMap: Map[String, Double],
  val educationMap: Map[String, Double],
  val randomForestModel: RandomForestModel
) extends Serializable

train method returns new model class

 new PIORandomForestModel(
    gendersMap = data.gendersMap,
    educationMap = data.educationMap,
    randomForestModel = m
   )

Predict method implementation

 def predict(
    model: PIORandomForestModel, // CHANGED
    query: Query): PredictedResult = {
    val gendersMap = model.gendersMap
    val educationMap = model.educationMap
    val randomForestModel = model.randomForestModel
    val label = randomForestModel.predict(Vectors.dense(Array(gendersMap(query.gender),query.age.toDouble,educationMap(query.education))))
    new PredictedResult(label)
  }

Sample Request

curl -H "Content-Type: application/json" -d '{ "gender":"Male",
"age":30,
"education":"College" }' http://localhost:8000/queries.json  

curl -H "Content-Type: application/json" -d '{ "gender":"Female",
"age":35,
"education":"No School" }' http://localhost:8000/queries.json