pySpark mapping multiple columns

I need to be able to compare two dataframes using multiple columns.

pySpark attempt

# get PrimaryLookupAttributeValue values from reference table in a dictionary to compare them to df1. 

primaryAttributeValue_List = [ p.PrimaryLookupAttributeValue for p in AttributeLookup.select('PrimaryLookupAttributeValue').distinct().collect() ]
primaryAttributeValue_List #dict of value, vary by filter 

Out: ['Archive',
 'Pending Security Deposit',
 'Partially Abandoned',
 'Revision Contract Review',
 'Open',
 'Draft Accounting In Review',
 'Draft Returned']


# compare df1 to PrimaryLookupAttributeValue
output = dataset_standardFalse2.withColumn('ConformedLeaseStatusName', f.when(dataset_standardFalse2['LeaseStatus'].isin(primaryAttributeValue_List), "FOUND").otherwise("TBD"))

display(output)

Solution 1:

From my understanding, you can create a map based on columns from reference_df (I assumed this is not a very big dataframe):

map_key = concat_ws('\0', PrimaryLookupAttributeName, PrimaryLookupAttributeValue)
map_value = OutputItemNameByValue

and then use this mapping to get the corresponding values in df1:

from itertools import chain
from pyspark.sql.functions import collect_set, array, concat_ws, lit, col, create_map

d = reference_df.agg(collect_set(array(concat_ws('\0','PrimaryLookupAttributeName','PrimaryLookupAttributeValue'), 'OutputItemNameByValue')).alias('m')).first().m
#[['LeaseStatus\x00Abandoned', 'Active'],
# ['LeaseRecoveryType\x00Gross-modified', 'Modified Gross'],
# ['LeaseStatus\x00Archive', 'Expired'],
# ['LeaseStatus\x00Terminated', 'Terminated'],
# ['LeaseRecoveryType\x00Gross w/base year', 'Modified Gross'],
# ['LeaseStatus\x00Draft', 'Pending'],
# ['LeaseRecoveryType\x00Gross', 'Gross']]

mappings = create_map([lit(i) for i in chain.from_iterable(d)])

primaryLookupAttributeName_List = ['LeaseType', 'LeaseRecoveryType', 'LeaseStatus']

df1.select("*", *[ mappings[concat_ws('\0', lit(c), col(c))].alias("Matched[{}]OutputItemNameByValue".format(c)) for c in primaryLookupAttributeName_List ]).show()
+----------------+...+---------------------------------------+-----------------------------------------------+-----------------------------------------+
|SourceSystemName|...|Matched[LeaseType]OutputItemNameByValue|Matched[LeaseRecoveryType]OutputItemNameByValue|Matched[LeaseStatus]OutputItemNameByValue|
+----------------+...+---------------------------------------+-----------------------------------------------+-----------------------------------------+
|          ABC123|...|                                   null|                                          Gross|                               Terminated|
|          ABC123|...|                                   null|                                 Modified Gross|                                  Expired|
|          ABC123|...|                                   null|                                 Modified Gross|                                  Pending|
+----------------+...+---------------------------------------+-----------------------------------------------+-----------------------------------------+

UPDATE: to set Column names from the information retrieved through reference_df dataframe:

# a list of domains to retrieve
primaryLookupAttributeName_List = ['LeaseType', 'LeaseRecoveryType', 'LeaseStatus']

# mapping from domain names to column names: using `reference_df`.`TargetAttributeForName`
NEWprimaryLookupAttributeName_List = dict(reference_df.filter(reference_df['DomainName'].isin(primaryLookupAttributeName_List)).agg(collect_set(array('DomainName', 'TargetAttributeForName')).alias('m')).first().m)

test = dataset_standardFalse2.select("*",*[ mappings[concat_ws('\0', lit(c), col(c))].alias(c_name) for c,c_name in NEWprimaryLookupAttributeName_List.items()]) 

Note-1: it is better to loop through primaryLookupAttributeName_List so the order of the columns are preserved and in case any entries in primaryLookupAttributeName_List is missing from the dictionary, we can set a default column-name, i.e. Unknown-<col>. In the old method, columns with the missing entries are simply discarded.

test = dataset_standardFalse2.select("*",*[ mappings[concat_ws('\0', lit(c), col(c))].alias(NEWprimaryLookupAttributeName_List.get(c,"Unknown-{}".format(c))) for c in primaryLookupAttributeName_List])

Note-2: per comments, to overwrite the existing column names(untested):

(1) use select:

test = dataset_standardFalse2.select([c for c in dataset_standardFalse2.columns if c not in NEWprimaryLookupAttributeName_List.values()] + [ mappings[concat_ws('\0', lit(c), col(c))].alias(NEWprimaryLookupAttributeName_List.get(c,"Unknown-{}".format(c))) for c in primaryLookupAttributeName_List]).show()

(2) use reduce (not recommended if the List is very long):

from functools import reduce

df_new = reduce(lambda d, c: d.withColumn(c, mappings[concat_ws('\0', lit(c), col(c))].alias(NEWprimaryLookupAttributeName_List.get(c,"Unknown-{}".format(c)))), primaryLookupAttributeName_List, dataset_standardFalse2)

reference: PySpark create mapping from a dict