Add algorithm_v3 which uses pandas udf instead of spark udf

Output is slightly modified now from an array of objects to that of
an array of json. This change will need to be accounted for in the
ETL pipeline