Split training & test according to a temporal attribute

[Nydia Neuss]

Hello

I have a time series and I want to split it in test (most recent values) and training (the other ones). I saw that the Split Data task allows you to assign the first portion to the training and the last to the test using “Data shuffle option” = “no shuffle” and setting the starting patter for training and test and it works!

I wonder if it is possible to assign the test set based on the value of a column (for example all records with a date greater than a certain threshold).

Thank you!

[Nicolò]

Hi!

You can do it using the Datamanager task:

• perform the filter that select your desired test set using the query manager
• right-click on data “test set” in the “modelling set” section (top-right)
• click on “assign the displayed row to test set”
• clear the query manager
• save