Pyspark ML - How to save pipeline and RandomForestClassificationModel

Question

I unable to save random forest model generated using ml package of python/spark.

>>> rf = RandomForestClassifier(labelCol="label", featuresCol="features")
>>> pipeline = Pipeline(stages=early_stages + [rf])
>>> model = pipeline.fit(trainingData)
>>> model.save("fittedpipeline")

Traceback (most recent call last): File "", line 1, in AttributeError: 'PipelineModel' object has no attribute 'save'

>>> rfModel = model.stages[8]
>>> print(rfModel)

RandomForestClassificationModel (uid=rfc_46c07f6d7ac8) with 20 trees

>> rfModel.save("rfmodel")

Traceback (most recent call last): File "", line 1, in AttributeError: 'RandomForestClassificationModel' object has no attribute 'save'**

Also tried by pass 'sc' as first parameter to save method.

Deach

What version of spark are you using ? – Capercaillie

I am using spark 1.6.0. Unfortunately I cannot upgrade to higher version due to certain reasons. Do we have some work around to save model in 1.6.0? – Deach

Nothing out of the box for <2.0.0 for pyspark. – Capercaillie

Answer 1

The main issue with your code is that you are using a version of Apache Spark prior to 2.0.0. Thus, save isn't available yet for the Pipeline API.

Here is a full example compounded from the official documentation. Let's create our pipeline first:

from pyspark.ml import Pipeline
from pyspark.ml.classification import RandomForestClassifier
from pyspark.ml.feature import IndexToString, StringIndexer, VectorIndexer

# Load and parse the data file, converting it to a DataFrame.
data = spark.read.format("libsvm").load("data/mllib/sample_libsvm_data.txt")

# Index labels, adding metadata to the label column.
# Fit on whole dataset to include all labels in index.
label_indexer = StringIndexer(inputCol="label", outputCol="indexedLabel")
labels = label_indexer.fit(data).labels

# Automatically identify categorical features, and index them.
# Set maxCategories so features with > 4 distinct values are treated as continuous.
feature_indexer = VectorIndexer(inputCol="features", outputCol="indexedFeatures", maxCategories=4)

early_stages = [label_indexer, feature_indexer]

# Split the data into training and test sets (30% held out for testing)
(train, test) = data.randomSplit([0.7, 0.3])

# Train a RandomForest model.
rf = RandomForestClassifier(labelCol="indexedLabel", featuresCol="indexedFeatures", numTrees=10)

# Convert indexed labels back to original labels.
label_converter = IndexToString(inputCol="prediction", outputCol="predictedLabel", labels=labels)

# Chain indexers and forest in a Pipeline
pipeline = Pipeline(stages=early_stages + [rf, label_converter])

# Train model. This also runs the indexers.
model = pipeline.fit(train)

You can now save your pipeline:

>>> model.save("/tmp/rf")
SLF4J: Failed to load class "org.slf4j.impl.StaticLoggerBinder".
SLF4J: Defaulting to no-operation (NOP) logger implementation
SLF4J: See http://www.slf4j.org/codes.html#StaticLoggerBinder for further details.

You can also save the RF model :

>>> rf_model = model.stages[2]
>>> print(rf_model)
RandomForestClassificationModel (uid=rfc_b368678f4122) with 10 trees
>>> rf_model.save("/tmp/rf_2")

Capercaillie

Hi, I'm not really understanding how to set path. If I just do .save() why not save it to current working directory? I've also tried adding full path however no luck. – Bifrost

You can save it where ever you want. The path here was not the issue. – Capercaillie

Answer 2

You can save pipelines and models. In case of loading these models, you need to know apriori the kind of model corresponding to each one. For example:

from pyspark.sql import SparkSession
from pyspark.ml import Pipeline
from pyspark.ml.feature import VectorAssembler, VectorIndexer, OneHotEncoder, StringIndexer, OneHotEncoderEstimator
from pyspark.ml.tuning import ParamGridBuilder, CrossValidator, CrossValidatorModel

df = *YOUR DATAFRAME*
categoricalColumns = ["A", "B", "C"]
stages = []

for categoricalCol in categoricalColumns:
    stringIndexer = StringIndexer(inputCol=categoricalCol, outputCol=categoricalCol + "Index")
    encoder = OneHotEncoderEstimator(inputCols=[stringIndexer.getOutputCol()], outputCols=[categoricalCol + "classVec"])
    stages += [stringIndexer, encoder]

label_stringIdx = StringIndexer(inputCol="id_imp", outputCol="label")
stages += [label_stringIdx]

assemblerInputs = [c + "classVec" for c in categoricalColumns]
assembler = VectorAssembler(inputCols=assemblerInputs, outputCol="features")
stages += [assembler]

pipeline = Pipeline(stages=stages)

pipelineModel = pipeline.fit(df)
pipelineModel.save("/path")

In the previous case, I saved a Pipeline with different stages. pipelineModel.save("/path")

Now, if you want to use them:

pipelineModel = Pipeline.load("/path")
df = pipelineModel.transform(df)

You can do the same for other cases, like:

cv = CrossValidator(estimator=lr, estimatorParamMaps=paramGrid, evaluator=evaluator, numFolds=2)

(trainingData, testData) = df.randomSplit([0.7, 0.3], seed=100)
cvModel = cv.fit(trainingData)
cvModel.save("/path")
cvM = CrossValidatorModel.load("/path")
predictions2 = cvM.transform(testData)

predictions = cvModel.transform(testData)

In brief, if you want to load the model you need to use the corresponding object.

Narbada