0 votes

I have a list of ~30 features and among them some can really reduce the number of prediction errors if only I could give them more weight.

Is this possible ?

Thanks !

1 Answer

0 votes
Hi Jérémie,

We do not support this in the model interface. Indeed, we think that modifying the coefficients by hand is not the best way to go since they are optimised based on past data. I am not sure modifying them will actually increase your classifier performance.  

If you really want to pursue this idea, it is actually possible via python. You would need to export the model to an ipython notebook, run it again and play with the coef_ attribute of your classifier.
1,296 questions
1,323 answers
11,862 users

©Dataiku 2012-2018 - Privacy Policy