Using prediction model in the lab, we noticed that each time our input dataset is regenerated/updated the current model settings are updated by Dataiku ?
For instance, some features are activated/included in the model whereas they were deactivated for the last train session. Is there an option/setting which can avoid this behavior ?
It would be very helpful as it's quite time consuming to review systematically all the features settings each time we update the input dataset.
Could you please share a screenshot example of a ml model where the feature activation changed after you updated your underlying dataset? Did the meaning of that feature change? For example from text to bigint?
I don't have an example at this time as we finalized the model.
In our case, the input dataset is generated with a Python recipe. And we noticed that any change on this input dataset may switch a feature from inactive (in the model) to active without changing the data type.
Moreover, most of the features have the buttons 'Accept' or 'Keep my settings' in the design/features handling screen. It can be very cumbersome to process feature by feature when you have a long list of features in the model.
Please, do you know in which context this automatic change happens ? is there a way to avoid it ?