I intend to setup a query on BigQuery that when it runs, it will run a machine learning on the elements that has not run it on previously.

For example, if I use object_tables to track files on cloud storage, how do I make sure that each time I run ML.PROCESS_DOCUMENT I’m not running it against the same documents I have run it previously?

Gemini suggested something as the following, but I’m not sure that ML.PROCESS_DOCUMENT function can be used that way:

    SELECT
    document_id,
    ML.PROCESS_DOCUMENT(
        MODEL `path_to_model_name`,
        (
            SELECT *
            FROM `path_to_storage`
            WHERE document_id = t.document_id
        )
    ) AS processed_document
FROM (
    SELECT document_id
    FROM `path_to_storage`
    WHERE
        NOT EXISTS (
            SELECT 1
            FROM `parsed_table`
            WHERE document_id = t.document_id
        )
        AND NOT EXISTS (
            SELECT 1
            FROM `parsed_table$__STREAMING_BUFFER`
            WHERE document_id = t.document_id
        )
) AS t;

I do get the following error:

Syntax error: Expected “)” but got identifier path_to_model_name at [4:15]