Impact of Attributes in the data collection

EvgyK

As some of you may know, attributes component of the data collection is mainly used by advanced ML algorithms, such as Amazon Ensemble, Anaplan AutoML, DeepAR+ and CNN-QR. Attributes allows those methods and algorithms to look at similarities across time series (items) and learn from them in order to improve forecast accuracy.

It is important that items that appear in historical data, related data and attributes are aligned, however, sometimes due to presence of empty values sometimes there is lack of alignment. For that, we recommend that you follow the recommendations of definitions of export actions in Anapedia for every component of the data collection.

From the perspective of data processing, if attributes component is present in the data collection, we are assuming that it has the master list of items. So, any items that are not present in the attributes file will not be handled by PlanIQ. We expect this change to provide more clarity to our users and also assist us in future enhancements and developments.

One such enhancement has been recently introduced to Amazon Ensemble algorithm. It now allows you to forecast items without any history provided that they are a part of the attributes. A forecast generated by Amazon Ensemble for those items can look like this:

Amazon Ensemble is using the data from the attributes file to find similar items to the new item in order to produce a forecast.

We will keep you posted on further enhancements in future,

Happy forecasting!