ipex-llm
ipex-llm copied to clipboard
If covert a spark df to shards with empty partition, will throw exception
Description
Fix https://github.com/intel-analytics/BigDL/issues/5877. Add logic to check if there is empty partition of spark dataframe, if there is and if spark_df_to_pd_sparkxshards
is called, will throw exception.
Why not create an empty spark dataframe partition?
Why not create an empty spark dataframe partition?
You mean create empty pandas df for empty spark dataframe partition? I think it may cause potential problems in the further processing(eg. filter
) and the error message may be confusing? Also I think it may bring extra overheads for many empty partitions?