Skip to main content

FORECAST

The FORECAST commands lets you forecast a time series, given a time and a column to predict. The command will append many columns to the output - the most important is forecast, which is the column we are predicting on.

The underlying model to our forecasting uses Prophet, created by Facebook.

Syntax

FORECAST(<column_name>, [, time=<time_column>, horizon=<horizon>])
  • column_name the input column name to forecast on. This must be a numerical input.
  • time is the column name of the time column. This column must have the format YYYY-MM-DD HH:MM:SS or YYYY-MM-DD.
    • The default is ds. Replace this if your time column is another time.

Returns

The command will append many column to the output, described below:

  • forecast is the median forecasted value.
  • forecast_upper is the 90th quantile forecast value.
  • forecast_lower is the 10th quantile forecast value.
  • trend is the median trend of the forecast.
  • trend_upper is the 90th quantile trend of the forecast.
  • trend_lower is the 10th quantile trend of the forecast.
  • weekly is the median weekly trend of the forecast.
  • weekly_upper is the 90th quantile weekly trend of the forecast.
  • weekly_lower is the 10th quantile weekly trend of the forecast.
  • yearly is the median yearly trend of the forecast.
  • yearly_upper is the 90th quantile yearly trend of the forecast.
  • yearly_lower is the 10th quantile yearly trend of the forecast.

Examples

Forecast the number of bookings in the next year.

SELECT * FROM airbnb FORECAST(total_bookings, time=date, horizon=365)