SparkML Decision Tree Regression Script with Cross-Validation and Parameter Sweep

SparkML_DecisionTree_Regression.py

########################################
## Title: Spark MLlib Decision Tree Regression Script, with Cross-Validation and Parameter Sweep
## Language: PySpark
## Author: Colby T. Ford, Ph.D.
########################################

from pyspark.ml.regression import DecisionTreeRegressor
from pyspark.ml.tuning import ParamGridBuilder, CrossValidator
from pyspark.ml.evaluation import RegressionEvaluator

# Create initial Decision Tree Model
dt = DecisionTreeRegressor(labelCol="label", featuresCol="features")

# Create ParamGrid for Cross Validation
dtparamGrid = (ParamGridBuilder()
             .addGrid(dt.maxDepth, [2, 5, 10, 20, 30])
             #.addGrid(dt.maxDepth, [2, 5, 10])
             .addGrid(dt.maxBins, [10, 20, 40, 80, 100])
             #.addGrid(dt.maxBins, [10, 20])
             .build())

# Evaluate model
dtevaluator = RegressionEvaluator(predictionCol="prediction", labelCol="label", metricName="rmse")

# Create 5-fold CrossValidator
dtcv = CrossValidator(estimator = dt,
                      estimatorParamMaps = dtparamGrid,
                      evaluator = dtevaluator,
                      numFolds = 5)

# Run cross validations
dtcvModel = dtcv.fit(train)
print(dtcvModel)

# Use test set here so we can measure the accuracy of our model on new data
dtpredictions = dtcvModel.transform(test)

# cvModel uses the best model found from the Cross Validation
# Evaluate best model
print('RMSE:', dtevaluator.evaluate(dtpredictions))