In 1.6.1 random_forrest's get_feature_importance() does not report categorical features

User 2032 | 9/28/2015, 1:18:55 PM

sf = gl.SFrame({
        'cat1': ['1', '1', '2', '2', '2'] * 1000,
        'cat2': ['1', '3', '3', '1', '1'] * 1000,
#         'b': [random() for _ in xrange(100)] * 50,
        'target': [0,0,0,1,1] * 1000
    })
m = gl.random_forest_classifier.create(sf, 'target')
m.get_feature_importance() # <- yields empty sframe

Comments

User 1178 | 9/28/2015, 10:15:59 PM

Hi Johnny,

This is a bug from our side!. We will fix it in coming release!

Thanks! Ping