Anaplan AutoML and Amazon Ensemble

This article explains individual algorithms provided by PlanIQ. However, it does not address the automated methods provided by PlanIQ.

As a customer you have several options:

  1. Use one of the automated methods as described below.
  2. Use combination of individual algorithms and automated methods and perform algorithm or method selection within an Anaplan model.

Let’s describe those automated options:

Anaplan AutoML – trains a variety of PlanIQ algorithms based on the data collection parameters. For example, Anaplan AutoML will include or exclude algorithms such as DeepAR+ based on the minimum data requirements. It performs tuning of the underlying algorithms using the knowledge and understanding of the data in the data collection. Anaplan AutoML trains about 10 different algorithms and then chooses a single best algorithm for the data collection.

In order to choose the algorithm, PlanIQ is evaluating the optimization metric for every individual item (time series) in the data collection and picks the algorithms that provided the best performance (minimal error rate) for the majority of the items in the data collection. This algorithm will be used in order to perform forecasts.

You should choose the optimization metric that is most suitable for your business.

Amazon Ensemble – is an ensemble method provided by Amazon Forecast. Within the ensemble method, Amazon runs all the applicable algorithms and provides a forecast result based on the combination of the results provided by those algorithms. Therefore, not a single algorithm is chosen in this case.

Amazon Ensemble can also leverage the optimization metric setting described above in order to optimize its performance.

Please note that both of those methods have longer model training and forecast times due to the higher complexity of the computation and processing.