CODEWITHSUNDEEP
pythonapache-sparkpysparkapache-spark-sql
Harish
Harish

Reputation: 999

find mean and corr of 10,000 columns in pyspark Dataframe

I have DF with 10K columns and 70Million rows. I want to calculate the mean and corr of 10K columns. I did below code but it wont work due to code size 64K issue (https://issues.apache.org/jira/browse/SPARK-16845)

Data:

region dept week sal val1  val2  val3 ... val10000   
 US    CS   1     1    2    1     1   ...  2 
 US    CS   2     1.5  2    3     1   ...  2
 US    CS   3     1    2    2     2.1      2
 US    ELE  1     1.1  2    2     2.1      2
 US    ELE  2     2.1  2    2     2.1      2
 US    ELE  3     1    2    1     2   .... 2
 UE    CS   1     2    2    1     2   .... 2

Code:

aggList =  [func.mean(col) for col in df.columns]  #exclude keys
df2= df.groupBy('region', 'dept').agg(*aggList)

code 2

aggList =  [func.corr('sal', col).alias(col) for col in df.columns]  #exclude keys
df2  = df.groupBy('region', 'dept', 'week').agg(*aggList)

this fails. Is there any alternative way to overcome this bug? and any one tried DF with 10K columns?. Is there any suggestion on performance improvement?

Upvotes: 4

Views: 2366

Answers (1)

Joachim Rosskopf
Joachim Rosskopf

Reputation: 1269

We also ran into the 64KB issue, but in a where clause, which is filed under another bug report. What we used as a workaround, is simply, to do the operations/transformations in several steps.

In your case, this would mean, that you don't do all the aggregatens in one step. Instead loop over the relevant columns in an outer operation:

  • Use select to create a temporary dataframe, which just contains columns you need for the operation.
  • Use the groupBy and agg like you did, except not for a list of aggregations, but just for on (or two, you could combine the mean and corr.
  • After you received references to all temporary dataframes, use withColumn to append the aggregated columns from the temporary dataframes to a result df.

Due to the lazy evaluation of a Spark DAG, this is of course slower as doing it in one operation. But it should evaluate the whole analysis in one run.

Upvotes: 1

Related Questions