PySpark: TypeError: 'Row' object does not support item assignment

Question

I used the following code to replace the None value in a DataFrame row to an empty string:

def replaceNone(row):
  row_len = len(row)
  for i in range(0, row_len):
    if row[i] is None:
      row[i] = ""    
  return row

in my pyspark code:

data_out = df.rdd.map(lambda row : replaceNone(row)).map(
  lambda row : "\t".join( [x.encode("utf-8") if isinstance(x, basestring) else str(x).encode("utf-8") for x in row])
)

Then I got the following errors:

File "<ipython-input-10-8e5d8b2c3a7f>", line 1, in <lambda>
  File "<ipython-input-2-d1153a537442>", line 6, in replaceNone
TypeError: 'Row' object does not support item assignment

Does anyone have any idea about the error? How do I replace a "None" value in a row to an empty string? Thanks!

Answer 1

Row is a subclass of tuple and tuples in Python are immutable hence don't support item assignment. If you want to replace an item stored in a tuple you have rebuild it from scratch:

## replace "" with placeholder of your choice 
tuple(x if x is not None else "" for x in row)  

If you want to simply concatenate flat schema replacing null with empty string you can use concat_ws:

from pyspark.sql.functions import concat_ws

df.select(concat_ws("\t", *df.columns)).rdd.flatMap(lambda x: x)

To prepare output it makes more sense to use spark-csv and specify nullValue, delimiter and quoteMode.