FellegiSunterPredict Example 1: Unsupervised Learning Model - Teradata Vantage

Input

• Input table: fspredict_input, created from the output of StringSimilarity Example 2: Compare src_text2 to tar_text with this SQL code:

  DROP TABLE fspredict_input;
  
  CREATE MULTISET TABLE fspredict_input AS (
    SELECT * FROM StringSimilarity (
      ON strsimilarity_input PARTITION BY ANY
      USING
      ComparisonColumnPairs (
                    'jaro (src_text2 , tar_text ) AS jaro1_sim',
                    'LD (src_text2 , tar_text, 2) AS ld1_sim',
                    'n_gram (src_text2 , tar_text, 2) AS ngram1_sim',
                    'jaro_winkler (src_text2 , tar_text, 2) AS jw1_sim'
      )
      CaseSensitive ('true')
      Accumulate ('id','src_text2','tar_text')
    ) AS dt1
  ) WITH DATA AS dt2 PARTITION BY id;
  
  SELECT * FROM fspredict_input ORDER BY 1;

• Model table: fg_unsupervised_model, output by FellegiSunter Example 1: Unsupervised Learning

SQL Call

SELECT * FROM FellegiSunterPredict (
  ON fspredict_input PARTITION BY ANY
  ON fg_unsupervised_model AS model DIMENSION
  USING
  Accumulate ('id', 'src_text2', 'tar_text', 'jaro1_sim',
              'ld1_sim','ngram1_sim', 'jw1_sim')
) AS dt ORDER BY id;

Output

The final column, match_result, contains the model prediction—M for match, U for no match. The weight column contains the weight of the object pair.