3.16.6 ML_PREDICT_TABLE

ML_PREDICT_TABLE generates predictions for an entire table of unlabeled data and saves the results to an output table. HeatWave AutoML performs the predictions in parallel.

ML_PREDICT_TABLE is a compute intensive process. Limiting operations to batches of 10 to 100 rows by splitting large tables into smaller tables is recommended.

A loaded model is required to run ML_PREDICT_TABLE. See Section 3.14.3, “Loading Models”.

The returned table includes a primary key:

• If the input table has a primary key, the output table will have the same primary key.

• If the input table does not have a primary key, the output table will have a new primary key column that auto increments.

  • As of MySQL 8.4.1, the name of the new primary key column is _4aad19ca6e_pk_id. The input table must not have a column with the name _4aad19ca6e_pk_id that is not a primary key.

  • Before MySQL 8.4.1, the name of the new primary key column is _id. The input table must not have a column with the name _id that is not a primary key.

MySQL 8.4.0 adds support to include prediction intervals to forecasting models.

The returned table includes the ml_results column which contains the prediction results and the data. The combination of results and data must be less than 65,532 characters.

ML_PREDICT_TABLE supports data drift detection for classification and regression models with the following:

• The options parameter includes the additional_details boolean value.

• The ml_results column includes the drift JSON object literal.

See: Section 3.14.11, “Data Drift Detection”.

ML_PREDICT_TABLE Syntax

MySQL 8.4.0 adds the prediction_interval option to the forecasting task.

mysql> CALL sys.ML_PREDICT_TABLE(table_name, model_handle, output_table_name), [options]);
 
options: {
     JSON_OBJECT("key","value"[,"key","value"] ...)
          "key","value": {
          ['threshold', 'N']
          ['topk', 'N']
          ['recommend', {'ratings'|'items'|'users'|'users_to_items'|'items_to_users'|'items_to_items'|'users_to_users'}|NULL]
          ['remove_seen', {'true'|'false'}]
          ['batch_size', 'N']
          ['additional_details', {'true'|'false'}]
          ['prediction_interval', 'N']
          }
}

A call to ML_PREDICT_TABLE can include columns that were not present during ML_TRAIN. A table can include extra columns, and still use the HeatWave AutoML model. This allows side by side comparisons of target column labels, ground truth, and predictions in the same table. ML_PREDICT_TABLE ignores any extra columns, and appends them to the results.

ML_PREDICT_TABLE parameters:

• table_name: Specifies the fully qualified name of the input table (schema_name.table_name). The input table should contain the same feature columns as the training dataset but no target column.

• model_handle: Specifies the model handle or a session variable containing the model handle

• output_table_name: Specifies the table where predictions are stored. The table is created if it does not exist. A fully qualified table name must be specified (schema_name.table_name). If the table already exists, an error is returned.

• options: A set of options in JSON format.

  As of MySQL 8.4.0, this parameter supports the forecasting task.

  This parameter supports the recommendation and anomaly_detection tasks. For all other tasks, set this parameter to NULL.

  • threshold: The optional threshold for use with the anomaly_detection task to convert anomaly scores to 1: an anomaly or 0: normal. 0 < threshold < 1. The default value is (1 - contamination)-th percentile of all the anomaly scores.

  • topk: The optional top K rows for use with the anomaly_detection and recommendation tasks. A positive integer between 1 and the table length.

    For the anomaly_detection task, the results include the top K rows with the highest anomaly scores. If topk is not set, ML_PREDICT_TABLE uses threshold.

    For an anomaly_detection task, do not set both threshold and topk. Use threshold or topk, or set options to NULL.

    For the recommendation task, the number of recommendations to provide. The default is 3.

    A recommendation task with implicit feedback can use both threshold and topk.

  • recommend: Use with the recommendation task to specify what to recommend. Permitted values are:

    • ratings: Use this option to predict ratings. This is the default value.

      The target column is prediction, and the values are float.

      The input table must contain at least two columns with the same names as the user column and item column from the training model.

    • items: Use this option to recommend items for users.

      The target column is item_recommendation, and the values are:

      JSON_OBJECT("column_item_id_name", JSON_ARRAY("item_1", ... , "item_k"), 
                  "column_rating_name" , JSON_ARRAY(rating_1, ..., rating_k))

      The input table must at least contain a column with the same name as the user column from the training model.

    • users: Use this option to recommend users for items.

      The target column is user_recommendation, and the values are:

      JSON_OBJECT("column_user_id_name", JSON_ARRAY("user_1", ... , "user_k"), 
                  "column_rating_name" , JSON_ARRAY(rating_1, ..., rating_k))

      The input table must at least contain a column with the same name as the item column from the training model.

    • users_to_items: This is the same as items.

    • items_to_users: This is the same as users.

    • items_to_items: Use this option to recommend similar items for items.

      The target column is item_recommendation, and the values are:

      JSON_OBJECT("column_item_id_name", JSON_ARRAY("item_1", ... , "item_k"))

      The input table must at least contain a column with the same name as the item column from the training model.

    • users_to_users: Use this option to recommend similar users for users.

      The target column is user_recommendation, and the values are:

      JSON_OBJECT("column_user_id_name", JSON_ARRAY("user_1", ... , "user_k"))

      The input table must at least contain a column with the same name as the user column from the training model.

  • remove_seen: If the input table overlaps with the training table, and remove_seen is true, then the model will not repeat existing interactions. The default is true. Set remove_seen to false to repeat existing interactions from the training table.

  • batch_size: The size of each batch. 1 ≤ batch_size ≤ 1,000. The default is 1,000, and this provides the best results.

  • additional_details: Set to true for ml_results to include the JSON object literal, drift.

  • prediction_interval: Use this to generate forecasted values with lower and upper bounds based on a specific prediction interval (level of confidence). For the prediction_interval value:

    • The default value is 0.95.

    • The data type for this value must be FLOAT.

    • The value must be greater than 0 and less than 1.0.

Syntax Examples

• A typical usage example that specifies the fully qualified name of the table to generate predictions for, the session variable containing the model handle, and the fully qualified output table name:

  mysql> CALL sys.ML_PREDICT_TABLE('ml_data.iris_test', @iris_model, 
            'ml_data.iris_predictions', NULL);

  To view ML_PREDICT_TABLE results, query the output table. The table shows the predictions and the feature column values used to make each prediction. The table includes the primary key, _4aad19ca6e_pk_id, and the ml_results column, which uses JSON format:

  mysql> SELECT * from ml_data.iris_predictions LIMIT 5;
  +-------------------+--------------+-------------+--------------+-------------+-----------------+-----------------+---------------------------------------------------------------------------------------------------------------------------------------+
  | _4aad19ca6e_pk_id | sepal length | sepal width | petal length | petal width | class           | Prediction      | ml_results                                                                                                                            |
  +-------------------+--------------+-------------+--------------+-------------+-----------------+-----------------+---------------------------------------------------------------------------------------------------------------------------------------+
  |                 1 |          7.3 |         2.9 |          6.3 |         1.8 | Iris-virginica  | Iris-virginica  | {'predictions': {'class': 'Iris-virginica'}, 'probabilities': {'Iris-setosa': 0.0, 'Iris-versicolor': 0.13, 'Iris-virginica': 0.87}}  |
  |                 2 |          6.1 |         2.9 |          4.7 |         1.4 | Iris-versicolor | Iris-versicolor | {'predictions': {'class': 'Iris-versicolor'}, 'probabilities': {'Iris-setosa': 0.0, 'Iris-versicolor': 1.0, 'Iris-virginica': 0.0}}   |
  |                 3 |          6.3 |         2.8 |          5.1 |         1.5 | Iris-virginica  | Iris-versicolor | {'predictions': {'class': 'Iris-versicolor'}, 'probabilities': {'Iris-setosa': 0.0, 'Iris-versicolor': 0.6, 'Iris-virginica': 0.4}}   |
  |                 4 |          6.3 |         3.3 |          4.7 |         1.6 | Iris-versicolor | Iris-versicolor | {'predictions': {'class': 'Iris-versicolor'}, 'probabilities': {'Iris-setosa': 0.0, 'Iris-versicolor': 0.99, 'Iris-virginica': 0.01}} |
  |                 5 |          6.1 |           3 |          4.9 |         1.8 | Iris-virginica  | Iris-virginica  | {'predictions': {'class': 'Iris-virginica'}, 'probabilities': {'Iris-setosa': 0.0, 'Iris-versicolor': 0.32, 'Iris-virginica': 0.68}}  |
  +-------------------+--------------+-------------+--------------+-------------+-----------------+-----------------+---------------------------------------------------------------------------------------------------------------------------------------+
  5 rows in set (0.00 sec)

See also:

• Section 3.9.3, “Using a Forecasting Model”

• Section 3.9.4.1, “Using Forecasting Models with Prediction Intervals”

• Section 3.10.3, “Using an Anomaly Detection Model”

• Section 3.11.3, “Using a Recommendation Model”

• Section 3.15, “Progress tracking”.

• Section 3.14.11, “Data Drift Detection”