Building a time-aware model factory

[Chinthaka]

Hello,

I am in need of building time series forecasting models for our individual centers. when I run the model factory based on some examples, it only builds regression models. how would I include Feature derivation window and prediction window through the API.

[mel]

Hi Chinthaka, here are some docs on specifying the feature derivation window and the forecast window through the API.

The code might look like this:

time_partition = dr.DatetimePartitioningSpecification(
    use_time_series      = True,
    datetime_partition_column = 'Date',
    multiseries_id_columns  = ['series_id'],
    feature_derivation_window_start= -8, # this is how far back each individual forecast requires
      feature_derivation_window_end=0,
      forecast_window_start=1,
      forecast_window_end=25, # this is how far into the future you want each forecast
   )

project.analyze_and_model(
      target = 'Revenue',    
      mode  = dr.AUTOPILOT_MODE.FULL_AUTO, 
      partitioning_method = time_partition,
      worker_count=-1)