3.13.2 Table Predictions with Topic Modeling

This section describes how to generate predictions for a table of data with topic modeling.

Before You Begin

• Connect to your HeatWave Database System.

• Complete the steps to train the model with topic modeling. See Section 3.13.1, “Training a Model with Topic Modeling”.

Generating Predictions with Topic Modeling

To generate predictions on a table with topic modeling:

1. Use the ML_MODEL_LOAD routine to load the model with topic modeling:

   mysql> CALL sys.ML_MODEL_LOAD(@topic_modeling, NULL);

2. Use the ML_PREDICT_TABLE routine to generate predictions with topic modeling:

   mysql> CALL sys.ML_PREDICT_TABLE('schema_name.input_table_name', @topic_modeling, 
   'schema_name.output_table_name', NULL);

   Where:

   • schema_name is the database name that contains the table. Update this with the appropriate database.

   • `input_table_name` is the input table that contains the training dataset. Update this with the appropriate input table.

   • @topic_modeling is the session variable that contains the model handle. Update this as needed.

   • `output_table_name` is the output table that will have the predictions. No existing table can have the same name.

   • The target column argument is set to NULL because topic modeling is an unsupervised task and does not need labeled data to train the model.

Once the output table of predictions is generated:

• The output table will have the same columns as the input table with the added column of ml_results

• The ml_results JSON object literal contains the array of word groups that represent the trained text.

• Every row in ml_results has predictions and every prediction has the new key tags which has the generated word groups.

To modify the number of word groups in the ml_results column, you can set the topk option. This option must be an integer greater or equal to one.

The following example generates predictions on a table with topic modeling and uses the topk option to limit the number of word groups to five:

mysql> CALL sys.ML_PREDICT_TABLE('schema_name.input_table_name', @topic_modeling, 
'schema_name.output_table_name',JSON_OBJECT('topk', 5));