Forecast

Description

The Forecast step reads a time series from an input table, fits a forecasting model, and writes future predicted values to an output table. Use it when you have historical values by date and want the workflow to produce a repeatable forecast.

Configuration

Forecast Parameters

• Input Table (required) — the table that contains the historical time series.
• Output Table (required) — the table where forecast rows are written.
• Date Column (required) — the mapped target column that contains the date or timestamp for each observation. The default is ds.
• Value Column (required) — the mapped target column that contains the numeric value to forecast.
• Series Columns (optional) — mapped target columns that identify separate series within the same table, such as customer, product, region, or account.
• Frequency — Auto, Hourly, Daily, Weekly, Monthly, or Quarterly. Use an explicit frequency when the input dates are sparse or irregular.
• Horizon — number of future periods to forecast. The default is 12.
• Model — Automatic by default. You can also select AutoETS, AutoARIMA, Seasonal Naive, Naive, or AutoTheta.

Table Data Selection

Use the table data selection tab to shape the input table before forecasting. The mapped target columns must include the Date Column, Value Column, and any Series Columns.

Use Inspect Source to populate the mapping from the selected input table. You can populate both sides of the mapping, only the source side, or only the target side before adjusting the columns.

Data Filters

Use the filters tab to limit rows, apply conditions, aggregate input data, or slice the result before the forecast runs.

Behavior

The step creates or updates the configured output table with forecast rows for the selected horizon. When you configure Series Columns, PlaidCloud forecasts each series separately and includes the series values in the output.

Resources

Forecast processing runs in a separate job. Enable Advanced Resource Management only when the forecast needs more CPU or memory than the default request. Increasing resource requests can delay Forecast starts while capacity is scheduled.