​
Description

The SELECT statement fetches predictions from the model table. The data is returned on the fly and the result set is not persisted.

But there are ways to save predictions data! You can save your predictions as a view using the CREATE VIEW statement. Please note that a view is a saved query and does not store data like a table. Another way is to create a table using the CREATE TABLE statement or insert your predictions into an existing table using the INSERT INTO statement.

​
Syntax

Here is the syntax for making batch predictions by joining one or more data source tables with one or more model tables:

SELECT t1.column, t2.column, m1.target, m2.target
FROM integration_name.table_name1 AS t1
JOIN integration_name.table_name2 AS t2 ON t1.column = t2.column
JOIN ...
JOIN mindsdb.model_name1 AS m1
JOIN mindsdb.model_name2 AS m2
JOIN ...
[ON t1.input_data = m1.expected_argument]
WHERE m1.parameter = 'value'
AND m2.parameter = 'value';

Where:

  • There are the data tables that provide input to the models: integration_name.table_name1, integration_name.table_name2.
  • These are the AI tables: mindsdb.model_name1, mindsdb.model_name2.

Note that you can provide input to the models from the data tables and also in the WHERE clause.

When querying for predictions, you can specify the partition_size parameter to split data into partitions and run prediction on different workers. Note that the ML task queue needs to be enabled to use this parameter.

To use the partition_size parameter, provide it in the USING clause, specifying the partition size, like this:

...
USING partition_size=100

Follow this doc page to learn more about AI Tables.

​
Example

Let’s make bulk predictions to predict the rental_price value using the home_rentals_model model joined with the data source table.

SELECT t.sqft, t.location, t.neighborhood, t.days_on_market, t.rental_price AS real_price,
       m.rental_price AS predicted_rental_price
FROM example_db.demo_data.home_rentals AS t
JOIN mindsdb.home_rentals_model AS m
LIMIT 5;

On execution, we get:

+-------+----------+-----------------+----------------+--------------+-----------------------------+
| sqft  | location | neighborhood    | days_on_market | real_price   | predicted_rental_price      |
+-------+----------+-----------------+----------------+--------------+-----------------------------+
| 917   | great    | berkeley_hills  | 13             | 3901         | 3886                        |
| 194   | great    | berkeley_hills  | 10             | 2042         | 2007                        |
| 543   | poor     | westbrae        | 18             | 1871         | 1865                        |
| 503   | good     | downtown        | 10             | 3026         | 3020                        |
| 1066  | good     | thowsand_oaks   | 13             | 4774         | 4748                        |
+-------+----------+-----------------+----------------+--------------+-----------------------------+

Follow this doc page to see examples of joining multiple data table with multiple models.