The ML.HOLIDAY_INFO function

This document describes the ML.HOLIDAY_INFO function, which you can use to return the list of holidays being modeled by an ARIMA_PLUS or ARIMA_PLUS_XREG time series forecasting model.

Syntax

ML.HOLIDAY_INFO(
  MODEL `PROJECT_ID.DATASET.MODEL_NAME`
)

Arguments

ML.HOLIDAY_INFO takes the following arguments:

  • PROJECT_ID: your project ID.
  • DATASET: the BigQuery dataset that contains the model.
  • MODEL_NAME: the name of the model.

Output

ML.HOLIDAY_INFO returns the following columns:

  • region: a STRING value that identifies the holiday region.
  • holiday_name: a STRING value that identifies the holiday.
  • primary_date: a DATE value that identifies the date the holiday falls on.
  • preholiday_days: an INT64 value that identifies the start of the holiday window around the holiday that was taken into account when modeling.
  • postholiday_days: an INT64 value that identifies the end of the holiday window around the holiday that was taken into account when modeling.

Example

The following example returns the results for a model that uses a custom holiday:

SELECT * FROM
ML.HOLIDAY_INFO(MODEL `mydataset.arima_model`);

The output looks similar to the following:

+-----------------------+--------------+-----------------+------------------+
| region | holiday_name | primary_date | preholiday_days | postholiday_days |
+--------------------------------------------------------+------------------+
| US     | Members day  | 2001-10-21   | 3               | 1                |
+-----------------------+--------------+-----------------+------------------+
| US     | Members day  | 2002-10-22   | 3               | 1                |
+-----------------------+--------------+-----------------+------------------+
| US     | Members day  | 2003-10-21   | 3               | 1                |
+-----------------------+--------------+-----------------+------------------+
| US     | Members day  | 2004-10-23   | 3               | 1                |
+-----------------------+--------------+-----------------+------------------+

Limitation

  • Results returned by ML.HOLIDAY_INFO only indicate the holiday information used during model fitting. They don't necessarily indicate the detection of a holiday effect. Use ML.EXPLAIN_FORECAST instead for actual holiday effect results.

What's next