Most relevant features

User 1273 | 2/11/2015, 2:58:28 PM

I would like to know: when doing a prediction how can I export the features that were relevant for that prediction's result?

Comments

User 91 | 2/13/2015, 4:25:43 AM

Vince

If you are using classifiers or recommenders, you can do model['coefficients'] to interpret the results.

See the section on interpreting results: http://dato.com/learn/gallery/notebooks/intro-regression.html


User 1273 | 2/17/2015, 5:25:32 PM

If I understood that well those are the most relevant features for the whole classifier, but I am searching for the most relevant features for the single prediction, that may be different every time (while the coefficients for the trained classifier remain the same).


User 1273 | 2/17/2015, 7:33:47 PM

A boosted trees classifier for example: how do I see which nodes were navigated?


User 1273 | 2/24/2015, 3:13:28 PM

Up?


User 1375 | 4/2/2015, 11:00:14 PM

model.show() is not drawing anything on ipython notebook after I call gl.canvas.set_target('ipynb'). However, show() does work as expected with SFrame and SArray.


User 1190 | 4/6/2015, 6:07:36 PM

Hi @msainz,

This seems to be a bug for model.show when 'ipynb' is the canvas target. We will get that fixed.

As workaround, please use the 'browser' target for now in ipython notebook.

Thanks, -jay