Skip to content

Running predictions without a destination table #3116

Open
@romainr

Description

@romainr

Is your feature request related to a problem?

As described in the excellent documentation, executing inferences requires a destination table.

Not requiring a destination table would help execute queries more interactively and be simpler to the end users.

I could not find the explanation in the design documentation why it was not possible to provide it as an option.

FROM table_references
  [WHERE where_condition]
  [LIMIT row_count]
TO PREDICT result_table_reference
[WITH
  attr_expr [, attr_expr ...]]
USING model_table_reference;

Describe the solution you'd like

TO PREDICT result_table_reference --> [TO PREDICT result_table_reference] optional

FROM table_references
  [WHERE where_condition]
  [LIMIT row_count]
[WITH
  attr_expr [, attr_expr ...]]
USING model_table_reference;

and returns:

predicted_label column1 column2...

where predicted_label comes from the model used.

If a user wants to save the predictions, it would add the TO PREDICT class like today.

Describe alternatives
e.g. BigQuery ML does not require a destination table and a regular SELECT * works: https://cloud.google.com/bigquery-ml/docs/reference/standard-sql/bigqueryml-syntax-predict

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions