bigquery-forecast

About

A bigquery-forecast tool forecasts time series data in BigQuery. It’s compatible with the following sources:

• bigquery

bigquery-forecast constructs and executes a SELECT * FROM AI.FORECAST(...) query based on the provided parameters:

• history_data (string, required): This specifies the source of the historical time series data. It can be either a fully qualified BigQuery table ID (e.g., my-project.my_dataset.my_table) or a SQL query that returns the data.
• timestamp_col (string, required): The name of the column in your history_data that contains the timestamps.
• data_col (string, required): The name of the column in your history_data that contains the numeric values to be forecasted.
• id_cols (array of strings, optional): If you are forecasting multiple time series at once (e.g., sales for different products), this parameter takes an array of column names that uniquely identify each series. It defaults to an empty array if not provided.
• horizon (integer, optional): The number of future time steps you want to predict. It defaults to 10 if not specified.

The behavior of this tool is influenced by the writeMode setting on its bigquery source:

• allowed (default) and blocked: These modes do not impose any special restrictions on the bigquery-forecast tool.
• protected: This mode enables session-based execution. The tool will operate within the same BigQuery session as other tools using the same source. This allows the history_data parameter to be a query that references temporary resources (e.g., TEMP tables) created within that session.

The tool’s behavior is also influenced by the allowedDatasets restriction on the bigquery source:

• Without allowedDatasets restriction: The tool can use any table or query for the history_data parameter.
• With allowedDatasets restriction: The tool verifies that the history_data parameter only accesses tables within the allowed datasets.
  • If history_data is a table ID, the tool checks if the table’s dataset is in the allowed list.
  • If history_data is a query, the tool performs a dry run to analyze the query and rejects it if it accesses any table outside the allowed list.

Example

tools:
 forecast_tool:
    kind: bigquery-forecast
    source: my-bigquery-source
    description: Use this tool to forecast time series data in BigQuery.

Sample Prompt

You can use the following sample prompts to call this tool:

• Can you forecast the history time series data in bigquery table bqml_tutorial.google_analytic? Use project_id myproject.
• What are the future total_visits in bigquery table bqml_tutorial.google_analytic?

Reference