How to include external features for a time series forecasting training job?

siddhesh_s Registered Posts: 1 ✭✭✭

I am trying to train a multivariate time series model with the "train and evaluate forecasting models" method available in the "forecast" plugin. I have added an external feature(holidays) to the training by checking the "Add external features" check box and selecting the external feature using drop down.. but the model(feed forward network) seems to be not accepting the external feature, cos the model parameters option shows "false" for use_external_features parameter in the post training results. Can someone let me know how to set it as "true"? Thanks.

Screen Shot 2022-09-05 at 5.48.52 AM.png

Screen Shot 2022-09-05 at 5.44.23 AM.png


Best Answer

  • ClemenceB
    ClemenceB Dataiker, Dataiku DSS Core Designer, Dataiku DSS ML Practitioner, Product Ideas Manager Posts: 18 Dataiker
    Answer ✓

    Hi @siddhesh_s

    The reason is that external features are only usable by AutoARIMA, DeepAR, Transformer, and MQ-CNN algorithms (cf the Plugin Documentation). The FeedForward model does not support external features.

    For the record, starting with V11, Time Series Forecasting is available in our Visual ML capabilities (see documentation) to enjoy a complete visual experience from design to deployment.



Setup Info
      Help me…