Spark 2.0 toPandas method

Question

I have a spark data frame like below:

topics.show(2)
+-----+--------------------+--------------------+--------------------+
|topic|         termIndices|         termWeights|        topics_words|
+-----+--------------------+--------------------+--------------------+
|    0|[0, 39, 68, 43, 5...|[0.06362107696025...|[, management, sa...|
|    1|[3, 1, 8, 6, 4, 1...|[0.03164821806301...|[objectives, lear...|
+-----+--------------------+--------------------+--------------------+
only showing top 2 rows

However when I try to convert to pandas data frame using below method which works in 1.6, I get an error.

topics.toPandas()

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-165-4c1231b68769> in <module>()
----> 1 topics.toPandas()

/Users/i854319/spark2/python/pyspark/sql/dataframe.pyc in toPandas(self)
   1440         """
   1441         import pandas as pd
-> 1442         return pd.DataFrame.from_records(self.collect(), columns=self.columns)
   1443 
   1444     ##########################################################################################

/Users/i854319/spark2/python/pyspark/sql/dataframe.pyc in collect(self)
    307         [Row(age=2, name=u'Alice'), Row(age=5, name=u'Bob')]
    308         """
--> 309         with SCCallSiteSync(self._sc) as css:
    310             port = self._jdf.collectToPython()
    311         return list(_load_from_socket(port, BatchedSerializer(PickleSerializer())))

/Users/i854319/spark2/python/pyspark/traceback_utils.pyc in __enter__(self)
     70     def __enter__(self):
     71         if SCCallSiteSync._spark_stack_depth == 0:
---> 72             self._context._jsc.setCallSite(self._call_site)
     73         SCCallSiteSync._spark_stack_depth += 1
     74 

AttributeError: 'NoneType' object has no attribute 'setCallSite'

So not sure if there is a bug in this method in Spark 2.0.2 or something is going wrong?

Answer 1

Copying my answer from a related question:

There's an open issue around this:

https://issues.apache.org/jira/browse/SPARK-27335?jql=text%20~%20%22setcallsite%22

The poster suggests forcing to sync your DF's backend with your Spark context:

df.sql_ctx.sparkSession._jsparkSession = spark._jsparkSession
df._sc = spark._sc

This worked for us, hopefully can work in other cases as well.