0
$\begingroup$

There are ways to estimate the importance of features in the model.

However, if I have a new feature and want to know whether this can help my existing classification model, what can analysis I do? Assuming retraining models with new features + old features is too computationally intensive.

$\endgroup$
0
$\begingroup$

I suppose you could create a new model with no large computational cost that takes as input the old features, extended with the new one.

Then use the output of this new model as input to the trained existing model.

Then train this combination to see if the new feature improves the result.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ How could you use "the output of this new model as input to the trained existing model"? The input of the trained existing model is those old features only, right? $\endgroup$ – user2149631 Jan 29 '17 at 19:15
  • $\begingroup$ If the new feature does not contribute any information needed for the classification that is not already in the old features, then the output of the new model should converge to those old features. But who knows ... It's just a thought. $\endgroup$ – Ytsen de Boer Jan 29 '17 at 22:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.