Time series forecasting with overlay data

classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

Time series forecasting with overlay data

José Heitor
Hi,

I am trying to make forecasts for periods beyond the end of the training
data. I have overlay data that extends beyond the training data period, but
cannot figure out how to structure the ARFF file so that the plugin can
recognise it and use it for predictions. This is how I have tried to
structure the dataset (simplified example):

<https://weka.8497.n7.nabble.com/file/t7061/Screenshot_20200428_182621.png>

I am wanting to train on the data until the 'missing' values for the target
(the first three instances in the above example), and make predictions for
the last two periods (with the question-marks). I keep getting a warning
about missing overlay data beyond the end of the training data...

How must I structure the dataset? And/or configure the plugin?

Thanks,
Jose



--
Sent from: https://weka.8497.n7.nabble.com/
_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to [hidden email]
To unsubscribe send an email to [hidden email]
To subscribe, unsubscribe, etc., visit https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|

Re: Time series forecasting with overlay data

Mark Hall
This can't be accomplished entirely within the Forecast tab in the Explorer because this environment is really geared towards evaluation against known target values. You will need to use both the Explorer and the KnowledgeFlow. Here is the procedure:

1. Load your data into the Forecast panel in the Explorer
2. Configure your target and overlay fields
3. In Advanced configuration/Output uncheck both "Output future predictions beyond end of series" and "Graph future predictions beyond end of series"
4. Click "Start" to build your model, and then save it to a file by right clicking over the entry in the "Result list"
5. Launch the KnowledgeFlow and set up a flow like so: ArffLoader --dataSet--> TimeSeriesForcasting --instance--> InstanceStreamToBatchMaker --dataSet--> TextViewer
6. Configure the ArffLoader with your dataset and the TimeSeriesForecasting step with the path to your saved model
7. Execute the Flow

The TextViewer will contain the full dataset with missing target values filled in with prediction. Replace InstanceStreamToBatchMaker and TextViewer with a connection to an ArffSaver in order to write the data with predictions out to an ARFF file.

Cheers,
Mark.

On 29/04/20, 10:53 AM, "José Heitor" <[hidden email]> wrote:

    Hi,
   
    I am trying to make forecasts for periods beyond the end of the training
    data. I have overlay data that extends beyond the training data period, but
    cannot figure out how to structure the ARFF file so that the plugin can
    recognise it and use it for predictions. This is how I have tried to
    structure the dataset (simplified example):
   
    <https://weka.8497.n7.nabble.com/file/t7061/Screenshot_20200428_182621.png>
   
    I am wanting to train on the data until the 'missing' values for the target
    (the first three instances in the above example), and make predictions for
    the last two periods (with the question-marks). I keep getting a warning
    about missing overlay data beyond the end of the training data...
   
    How must I structure the dataset? And/or configure the plugin?
   
    Thanks,
    Jose
   
   
   
    --
    Sent from: https://weka.8497.n7.nabble.com/
    _______________________________________________
    Wekalist mailing list -- [hidden email]
    Send posts to [hidden email]
    To unsubscribe send an email to [hidden email]
    To subscribe, unsubscribe, etc., visit https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
    List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
   
_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to [hidden email]
To unsubscribe send an email to [hidden email]
To subscribe, unsubscribe, etc., visit https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html