KNNRecommenderPredict Example - Teradata Vantage

Input

SQL Call

The call recommends five movies for ten users.

SELECT * FROM KnnRecommenderPredict (
  ON (SELECT * FROM ml_ratings WHERE userid <= 10) AS Ratings
    PARTITION BY userid
  ON ml_bias AS Bias DIMENSION
  ON ml_weights AS Weights DIMENSION
  USING
  TopK (5)
) AS dt ORDER BY 1 DESC;

Output

 userid itemid prediction         
 ------ ------ ------------------ 
      5    454   4.43944806005997
      5   2890  4.353130720557537
      5     34   4.43944806005997
      5    420 4.5586737193706535
      5    357 4.5150229309770635
      4   2023  4.472692557514834
      4   1258  4.472692557514834
      4    300  4.622911229674568
      4   1704  4.472692557514834
      4   2571  4.722330311219902
      3   1258  4.396245483590272
      3   1704  4.396245483590272
      3   2023  4.396245483590272
      3   2571  4.645740472524265
      3   1136  4.146753684279514
      3   1711  4.146753684279514
      3   1267  4.146753684279514
      2    515  4.540419986677616
      2    480  4.540419986677616
      2    296  4.463307355858737
      2    265  4.540419986677616
      2    648  4.540419986677616
      1   1278  4.228140795465176
      1    553  3.973010311922945
      1     21   4.02887733802919
      1    420  4.466978626573491
      1   1639  4.479795437756031

Some predicted ratings are higher than 5, even though the maximum rating is 5. The weighted KNN recommendation algorithm does not limit its results to the range of the input data. The outcome of interest are the items with the highest recommendation score; if the resulting ratings must be limited to a specific range, normalize the output data.

Download a zip file of all examples and a SQL script file that creates their input tables from the attachment in the left sidebar.