Train Sparkling Water Algorithms with Grid Search¶

Grid Search serves for finding optimal values for hyper-parameters of a given H2O/SW algorithm. Grid Search in Sparkling Water is able to traverse hyper-space for H2OGBM, H2OXGBoost, H2ODRF, H2OGLM, H2OGAM, H2ODeepLearning, H2OKMeans, and H2OIsolationForest. For more details about hyper-parameters for a specific algorithm (see H2O-3 documentation).

Sparkling Water provides API in Scala and Python for Grid Search. The following sections describe how to Apply Grid Search on H2ODRF in both languages. See also Parameters of H2OGridSearch.

Scala

First, let’s start Sparkling Shell as

./bin/sparkling-shell

Start H2O cluster inside the Spark environment

import ai.h2o.sparkling._
import java.net.URI
val hc = H2OContext.getOrCreate()

Parse the data using H2O and convert them to Spark Frame

    import org.apache.spark.SparkFiles
spark.sparkContext.addFile("https://raw.githubusercontent.com/h2oai/sparkling-water/master/examples/smalldata/prostate/prostate.csv")
    val rawSparkDF = spark.read.option("header", "true").option("inferSchema", "true").csv(SparkFiles.get("prostate.csv"))
val sparkDF = rawSparkDF.withColumn("CAPSULE", $"CAPSULE" cast "string")
val Array(trainingDF, testingDF) = sparkDF.randomSplit(Array(0.8, 0.2))

Define the algorithm, which will be a subject of hyper-parameter tuning

import ai.h2o.sparkling.ml.algos.H2ODRF
val algo = new H2ODRF().setLabelCol("CAPSULE")

By default, the H2ODRF algorithm distinguishes between a classification and regression problem based on the type of the label column of the training dataset. If the label column is a string column, a classification model will be trained. If the label column is a numeric column, a regression model will be trained. If you don’t want be worried about column data types, you can explicitly identify the problem by using ai.h2o.sparkling.ml.algos.classification.H2ODRFClassifier or ai.h2o.sparkling.ml.algos.regression.H2ODRFRegressor instead.

Define a hyper-space which will be traversed

import scala.collection.mutable.HashMap
val hyperSpace: HashMap[String, Array[AnyRef]] = HashMap()
hyperSpace += "ntrees" -> Array(1, 10, 30).map(_.asInstanceOf[AnyRef])
hyperSpace += "mtries" -> Array(-1, 5, 10).map(_.asInstanceOf[AnyRef])

Pass the algorithm and hyper-space to the grid search and set properties defining the way how the hyper-space will be traversed.

Sparkling Water supports two strategies for traversing hyperspace:

Cartesian - (Default) This strategy tries out every possible combination of hyper-parameter values and finishes after the whole space is traversed.
RandomDiscrete - In each iteration, the strategy randomly selects the combination of values from the hyper-space and can be terminated before the whole space is traversed. The termination depends on various criteria (consider parameters: maxRuntimeSecs, maxModels, stoppingRounds, stoppingTolerance, stoppingMetric). For details see H2O-3 documentation

import ai.h2o.sparkling.ml.algos.H2OGridSearch
val grid = new H2OGridSearch()
    .setHyperParameters(hyperSpace)
    .setAlgo(algo)
    .setStrategy("Cartesian")

Fit the grid search to get the best DRF model.

val model = grid.fit(trainingDF)

You can also get raw model details by calling the getModelDetails() method available on the model as:

model.getModelDetails()

Run Predictions

model.transform(testingDF).show(false)

Python

First, let’s start PySparkling Shell as

./bin/pysparkling

Start H2O cluster inside the Spark environment

from pysparkling import *
hc = H2OContext.getOrCreate()

Parse the data using H2O and convert them to Spark Frame

import h2o
frame = h2o.import_file("https://raw.githubusercontent.com/h2oai/sparkling-water/master/examples/smalldata/prostate/prostate.csv")
sparkDF = hc.asSparkFrame(frame)
sparkDF = sparkDF.withColumn("CAPSULE", sparkDF.CAPSULE.cast("string"))
[trainingDF, testingDF] = sparkDF.randomSplit([0.8, 0.2])

Train the model. You can configure all the available DRF arguments using provided setters or constructor parameters, such as the label column.

from pysparkling.ml import H2ODRF
algo = H2ODRF(labelCol = "CAPSULE")

By default, the H2ODRF algorithm distinguishes between a classification and regression problem based on the type of the label column of the training dataset. If the label column is a string column, a classification model will be trained. If the label column is a numeric column, a regression model will be trained. If you don’t want to be worried about column data types, you can explicitly identify the problem by using H2ODRFClassifier or H2ODRFRegressor instead.

Define a hyper-space which will be traversed

hyperSpace = {"ntrees": [1, 10, 30], "mtries": [-1, 5, 10]}

Pass the algorithm and hyper-space to the grid search and set properties defining the way how the hyper-space will be traversed.

Sparkling Water supports two strategies for traversing hyperspace:

Cartesian - (Default) This strategy tries out every possible combination of hyper-parameter values and finishes after the whole space is traversed.
RandomDiscrete - In each iteration, the strategy randomly selects the combination of values from the hyper-space and can be terminated before the whole space is traversed. The termination depends on various criteria (consider parameters: maxRuntimeSecs, maxModels, stoppingRounds, stoppingTolerance, stoppingMetric). For details see H2O-3 documentation

from pysparkling.ml import H2OGridSearch
grid = H2OGridSearch(hyperParameters=hyperSpace, algo=algo, strategy="Cartesian")

Fit the grid search to get the best DRF model.

model = grid.fit(trainingDF)

You can also get raw model details by calling the getModelDetails() method available on the model as:

model.getModelDetails()

Run Predictions

model.transform(testingDF).show(truncate = False)