'JavaPackage' object is not callable - MLeap

Question

When I try to serialize a model using MLeap using the following code:

import mleap.pyspark
from mleap.pyspark.spark_support import SimpleSparkSerializer

# Import standard PySpark Transformers and packages
from pyspark.ml.feature import VectorAssembler, StandardScaler, OneHotEncoder, StringIndexer
from pyspark.ml import Pipeline, PipelineModel
from pyspark.sql import Row

# Create a test data frame
l = [('Alice', 1), ('Bob', 2)]
rdd = sc.parallelize(l)
Person = Row('name', 'age')
person = rdd.map(lambda r: Person(*r))
df2 = spark.createDataFrame(person)
df2.collect()

# Build a very simple pipeline using two transformers
string_indexer = StringIndexer(inputCol='name', outputCol='name_string_index')

feature_assembler = VectorAssembler(inputCols=[string_indexer.getOutputCol()], outputCol="features")

feature_pipeline = [string_indexer, feature_assembler]

featurePipeline = Pipeline(stages=feature_pipeline)

fittedPipeline = featurePipeline.fit(df2)


# serialize the model:
fittedPipeline.serializeToBundle("jar:file:/tmp/pyspark.example.zip", fittedPipeline.transform(df2))

However I get the following error:

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-2-98a49e4cd023> in <module>()
----> 1 fittedPipeline.serializeToBundle("jar:file:/tmp/pyspark.example.zip", fittedPipeline.transform(df2))

/opt/anaconda2/envs/py345/lib/python3.4/site-packages/mleap/pyspark/spark_support.py in serializeToBundle(self, path, dataset)
     22 
     23 def serializeToBundle(self, path, dataset=None):
---> 24     serializer = SimpleSparkSerializer()
     25     serializer.serializeToBundle(self, path, dataset=dataset)
     26 

/opt/anaconda2/envs/py345/lib/python3.4/site-packages/mleap/pyspark/spark_support.py in __init__(self)
     37     def __init__(self):
     38         super(SimpleSparkSerializer, self).__init__()
---> 39         self._java_obj = _jvm().ml.combust.mleap.spark.SimpleSparkSerializer()
     40 
     41     def serializeToBundle(self, transformer, path, dataset):

TypeError: 'JavaPackage' object is not callable

Please assist?

Answer 1

The answer of @Tshilidzi Madau is correct - what you need to do is to add mleap-spark jar into your spark classpath.

One option in pyspark is to set the spark.jars.packages config while creating the SparkSession:

from pyspark.sql import SparkSession

spark = SparkSession.builder \
    .config('spark.jars.packages', 'ml.combust.mleap:mleap-spark_2.12:0.19.0') \
    .config("spark.jars.excludes", "net.sourceforge.f2j:arpack_combined_all") \ # this exclude is needed as this lib seems not to be available in public maven repos
    .getOrCreate()

I tested it with Spark 3.0.3 and mleap 0.19.0

Answer 2

I managed to fix this problem by downloading and pointing to the missing jar files on the spark submit script. For my case, I had installed MLeap 0.8.1 and was using Spark2 built on Scalar11, so I downloaded the following jar files from MvnRepository:

• metrics-core-2.2.0
• mleap-base_2.11-0.8.1
• mleap-core_2.11-0.8.1
• mleap-runtime_2.11-0.8.1
• mleap-spark_2.11-0.8.1
• mleap-spark-base_2.11-0.8.1
• mleap-tensor_2.11-0.8.1

Then I also pointed to this jar files using the --jar flag on my spark submit file as follows (I also pointed to the maven repository using the --repository flag):

export PYSPARK_SUBMIT_ARGS='--master yarn --deploy-mode client --driver-memory 40g --num-executors 15 --executor-memory 30g --executor-cores 5 --packages ml.combust.mleap:mleap-runtime_2.11:0.8.1 --repositories http://YOUR MAVEN REPO/ --jars arpack_combined_all-0.1.jar,mleap-base_2.11-0.8.1.jar,mleap-core_2.11-0.8.1.jar,mleap-runtime_2.11-0.8.1.jar,mleap-spark_2.11-0.8.1.jar,mleap-spark-base_2.11-0.8.1.jar,mleap-tensor_2.11-0.8.1.jar pyspark-shell'
jupyter notebook --no-browser --ip=$(hostname -f)

-Source