PySpark: Convert Python Array/List to Spark Data Frame

Hi venu,

There are several things you need to know:

• collect function will request all data in the data frame to be sent to your driver application.
• From Spark later versions, you can directly use DataFrame APIs to transform instead of using RDD and loop through.
• Similarly for saving as CSV, you can also directly use DataFrame APIs.

Thus, to utilize parallelism and to improve performance, I would suggest the following changes:

1. Repartition your DataFrame df using repartition function if there is appropriate partition keys. 
2. Directly use df to do all kinds of transformations. You can find more information here: pyspark.sql.DataFrame — PySpark 3.2.0 documentation (apache.org). Remember to read the documentation of your Spark version. 
3. Use df.write to save the data into HDFS. 

Hi Raymond,

But it takes lot of time because of df.collect()

Is there any way to fasten this process? I tried to use --num-executors 5 in spark-submit but no change in performance. Also if possible please provide a solution for this too on how can we leverage --num-executors in this case. Since it's a 'pyspark dataframe' i also used df1 = df.toPandas() but no change in performance.