This is prerelease documentation for a service in preview release. It is subject to change.

The Multi-Quantile Recurrent Neural Network (MQRNN) Recipe

Use the Amazon Forecast Multi-Quantile Recurrent Neural Network (MQRNN) recipe for demand forecasting and with other one-dimensional time series datasets.

How MQRNN Works

The Amazon Forecast MQRNN recipe supports the following features:

  • Forecast horizon

    Amazon Forecast MQRNN can forecast many time points into the future.

  • Time frequency

    Amazon Forecast MQRNN uses the time frequency of the dataset to make smart forecasts based on datetime features of hour of the day, day of the week, and so one, which are inherent in the input dataset.

  • Correlated time-varying information

    Amazon Forecast MQRNN can collect the correlations in many datasets and other information to boost forecasting accuracy. For example, it can include item prices and major holidays to make better-informed forecasts of customer demands in the retail domain.

  • Correlated time-static information

    Similar to the way that it correlates time-varying information, Amazon Forecast MQRNN can include time-invariant information to correlate with the input time series to improve forecasting accuracy. For example, it can include shirt color to forecast demands on shirts.

For general information about the MQRNN recipe, see A Multi-Quantile Recurrent Neural Network on the Cornell University Library website.

MQRNN Hyperparameters

The following table lists the hyperparameters that you can use to tune a MQRNN model.

Parameter Name Description
use_related_ts A flag that indicates whether Amazon Forecast should use correlated information (set to True) or not (set to False). We recommend that novice Amazon Forecast users not use this hyperparameter.
Required

No.

Valid values

True | False

Default value

True

use_datetime-features A flag that indicates whether Amazon Forecast should consider datetime features in forecasting (set to True) or not (set to False). We recommend that novice Amazon Forecast users not use this hyperparameter.
Required

No.

Valid values

True | False

Default value

True