cancel
Showing results for 
Search instead for 
Did you mean: 

Time Series changing forecast distance

Time Series changing forecast distance

Hello,

I have recently started using the time series models in data robot for monthly data of mine. I was able to train this data and get a model. However, when forecasting for future values, it is only doing so for 3 months into the future. How can I change this forecast distance/forecast window to encompass a larger range of months. Note I use python to communicate with the data robot api so an answer in python would be great. Thanks.

Labels (1)
0 Kudos
2 Replies
senkin13
DataRobot Employee
DataRobot Employee

Hi netnet,

I think you could change the input values when you start configuration for  Forecast Window,then go to Advanced Options -> Configure Backtesting to check forecasting future period.

0 Kudos
dalilaB
DataRobot Alumni

Here is the code 

If you have a calendar that you want to use you can upload it to DataRobot this way

calendar = dr.CalendarFile.create('Acme_Calendar.csv',
                              calendar_name = 'US_Calendar')


 

def start_modeling(projectName,data_df):
    '''provide a project name, and the dataframe for training
        1. Create project
        2. Set specifications for time-series modeling
        3. Start project'''
    project = dr.Project.create(project_name=projectName, sourcedata=data_df)
    mySpec = dr.DatetimePartitioningSpecification(
                datetime_partition_column=ts_settings['date_col'],
                use_time_series=True,
                multiseries_id_columns=[ts_settings['series_id']],
                default_to_known_in_advance=False,
                feature_derivation_window_start=ts_settings['fdw_start'],
                feature_derivation_window_end= ts_settings['fdw_end'],
                forecast_window_start=ts_settings['fd_start'],
                forecast_window_end=ts_settings['fd_end'],
                calendar_id=calendar.id

    )
    project.set_target(
            target = ts_settings['target'],
            partitioning_method = mySpec,
            mode=dr.AUTOPILOT_MODE.FULL_AUTO  ##### Here is where we can change to dr.AUTOPILOT_MODE.FULL_AUTO
    )
    project.wait_for_autopilot()  ##Wait for autopilot to finish
    return(project)

#Create and Start Project
project = start_modeling(ts_settings['project_name'],ts_settings['use_time_series'])

Also, you can find a full python solution for time-series in here 

 


 

0 Kudos