File location for training model files

A location must be specified to save training model files produced by the Batch Forecast Training task. This location is specified in the FORECAST_MODEL_PATH registry parameter. The files may be saved to a local directory on the job scheduler server or to a network or cloud location, such as a directory in an Amazon S3 bucket.

When the task runs, it creates model files for each driver that it trains. The file names for these files include the identifier for the relevant driver. These files can be useful to review when troubleshooting forecast results.

We recommend that customers using machine learning forecasting enable asynchronous forecast creation. This configures the Forecasting screen to generate user-created forecasts on the job scheduler, which has direct access to the model files.

When asynchronous forecast creation is not enabled, the application server requires access to the model files to create forecasts. If the job scheduler and application server are on different physical servers, then the model files must be copied from the job scheduler to the same file path on the application server. Note that this does not affect the Batch Forecasting task, which always creates forecasts on the job scheduler.

For details on asynchronous forecast creation, see Forecast Async Task.