AttributeError: 'PipelineModel' object has no attribute 'fitMultiple'

Question

I am trying to tune a random forest model using pyspark, CrossValidator, and BinaryClassificationEvaluator, CrossValidator, but when I do so I get an error. Here is my code.

from pyspark.ml.evaluation import BinaryClassificationEvaluator
from pyspark.ml.classification import RandomForestClassifier
from pyspark.ml.feature import VectorAssembler
from pyspark.ml import Pipeline

# Create a spark RandomForestClassifier using all default parameters.
# Create a training, and testing df
training_df, testing_df = raw_data_df.randomSplit([0.6, 0.4])

# build a pipeline for analysis
va = VectorAssembler().setInputCols(training_df.columns[0:110:]).setOutputCol('features')

# featuresCol="features"
rf = RandomForestClassifier(labelCol="quality")

# Train the model and calculate the AUC using a BinaryClassificationEvaluator
rf_pipeline = Pipeline(stages=[va, rf]).fit(training_df)

bce = BinaryClassificationEvaluator(labelCol="quality")

# Check AUC before tuning
bce.evaluate(rf_pipeline.transform(testing_df))


from pyspark.ml.tuning import CrossValidator, ParamGridBuilder

paramGrid = ParamGridBuilder().build()

crossValidator = CrossValidator(estimator=rf_pipeline, 
                          estimatorParamMaps=paramGrid, 
                          evaluator=bce, 
                          numFolds=3)

model = crossValidator.fit(training_df)

It is throwing this error:

AttributeError: 'PipelineModel' object has no attribute 'fitMultiple'

How do I fix this issue?

Answer 1

CrossValidator estimator takes a object of Pipeline and not the Pipeline model.

Please check this example for reference- https://github.com/apache/spark/blob/master/examples/src/main/python/ml/cross_validator.py

your code should be modified as below

1. create a pipeline

rf_pipe = Pipeline(stages=[va, rf])

2. Use that pipeline as estimator in crossvalidator

crossValidator = CrossValidator(estimator=rf_pipe, 
                          estimatorParamMaps=paramGrid, 
                          evaluator=bce, 
                          numFolds=3)

Oveall-

....

# Train the model and calculate the AUC using a BinaryClassificationEvaluator
rf_pipe = Pipeline(stages=[va, rf])
rf_pipeline = rf_pipe.fit(training_df)

...

crossValidator = CrossValidator(estimator=**rf_pipe**, 
                          estimatorParamMaps=paramGrid, 
                          evaluator=bce, 
                          numFolds=3)

model = crossValidator.fit(training_df)