1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
|
import numpy as np
from Orange.classification.sgd import SGDClassificationLearner
from Orange.data import Variable
from Orange.modelling import SklFitter
from Orange.preprocess.score import LearnerScorer
from Orange.regression import SGDRegressionLearner
__all__ = ['SGDLearner']
class _FeatureScorerMixin(LearnerScorer):
feature_type = Variable
class_type = Variable
def score(self, data):
model = self.get_learner(data)(data)
return (np.atleast_2d(np.abs(model.skl_model.coef_)).mean(0),
model.domain.attributes)
class SGDLearner(SklFitter, _FeatureScorerMixin):
name = 'sgd'
__fits__ = {'classification': SGDClassificationLearner,
'regression': SGDRegressionLearner}
def _change_kwargs(self, kwargs, problem_type):
pref = "classification" if problem_type is self.CLASSIFICATION else "regression"
return kwargs | {
attr: kwargs[pattr]
for attr, pattr in ((attr, f"{pref}_{attr}")
for attr in ('loss', 'epsilon'))
if pattr in kwargs}
|
