hugobuddel / orange3

Orange/classification/sgd.py

import numpy

The import numpy could not be resolved.

from sklearn import linear_model

The import sklearn could not be resolved.

from sklearn.preprocessing import Imputer

The import sklearn.preprocessing could not be resolved.

Unused Imputer imported from sklearn.preprocessing

import Orange.data
import Orange.classification
#from Orange.data.continuizer import DomainContinuizer

class SGDLearner(Orange.classification.SklLearner):

    def __init__(self, all_classes, means=None, stds=None):

The __init__ method of the super-class SklLearnerClassification is not called.

        self.all_classes = all_classes
        # The SGD Learner works significantly better with normalized data.
        # However, we cannot get the normalization from the data because we
        # do partial fitting. So it has to be provided.
        self.means = means
        self.stds = stds
        self.clf = linear_model.SGDClassifier(loss='log')
        self.clf.means = self.means
        self.clf.stds = self.stds

        #self.reset()

    def partial_fit(self, X, Y, W, normalize=True):

The argument W seems to be unused.

        X = X.copy()
        if normalize:
            if self.means is not None:
                X -= self.means
            if self.stds is not None:
                X /= self.stds

        # Sometimes this exception occurs, but it is unclear when:
        """
        Exception in thread Thread-3:
        Traceback (most recent call last):
          File "/home/evis/anaconda3/lib/python3.4/threading.py", line 920, in _bootstra
        p_inner
            self.run()
          File "/home/evis/anaconda3/lib/python3.4/threading.py", line 868, in run
            self._target(*self._args, **self._kwargs)
          File "/home/evis/orange3/Orange/widgets/classify/owsgd.py", line 239, in train
        ing_thread
            self.train()
          File "/home/evis/orange3/Orange/widgets/classify/owsgd.py", line 261, in train
            classifier = self.learner.partial_fit(new_instances.X, new_instances.Y, None)
          File "/home/evis/orange3/Orange/classification/sgd.py", line 31, in partial_fit
            self.clf = self.clf.partial_fit(X, Y.reshape(-1), self.all_classes)
          File "/home/evis/anaconda3/lib/python3.4/site-packages/sklearn/linear_model/stochastic_gradient.py", line 526, in partial_fit
            coef_init=None, intercept_init=None)
          File "/home/evis/anaconda3/lib/python3.4/site-packages/sklearn/linear_model/stochastic_gradient.py", line 358, in _partial_fit
            X, y = check_X_y(X, y, 'csr', dtype=np.float64, order="C")
          File "/home/evis/anaconda3/lib/python3.4/site-packages/sklearn/utils/validation.py", line 450, in check_X_y
            _assert_all_finite(y)
          File "/home/evis/anaconda3/lib/python3.4/site-packages/sklearn/utils/validation.py", line 52, in _assert_all_finite
            " or a value too large for %r." % X.dtype)
        ValueError: Input contains NaN, infinity or a value too large for dtype('float64').
        """

This string statement has no effect and could be removed.

        #print("SGD training to %s" % Y.reshape(-1))

        self.clf = self.clf.partial_fit(X, Y.reshape(-1), self.all_classes)
        

Trailing whitespace

        return SGDClassifier(self.clf)
    

Trailing whitespace

    # TODO: Remove fit completely?

TODO and FIXME comments should generally be avoided.

    fit = partial_fit

    # TODO: Is this reset function necessary?

TODO and FIXME comments should generally be avoided.

    def reset(self):
        # 'log' or 'modified_huber' required to predict probabilities.
        self.clf = linear_model.SGDClassifier(loss='log')
        self.clf.means = self.means
        self.clf.stds = self.stds
    

Trailing whitespace

    def decision_function(self, X):
        if self.means is not None and self.stds is not None:
            Xa = (numpy.array(X)-numpy.array(self.means))/numpy.array(self.stds)
        else:
            Xa = X
        

Trailing whitespace

        value = self.clf.decision_function(Xa)
        return value

    def predict(self, X):
        if self.means is not None and self.stds is not None:
            Xa = (numpy.array(X)-numpy.array(self.means))/numpy.array(self.stds)
        else:
            1/0

This statement seems to have no effect and could be removed.

            Xa = X
        

Trailing whitespace

        value = self.clf.predict(Xa)
        return value
        

Trailing whitespace

class SGDClassifier(Orange.classification.SklModel):

    def __init__(self, clf):

The __init__ method of the super-class SklModelClassification is not called.

        self.clf = clf

    def predict(self, X):
        if self.clf.means is not None and self.clf.stds is not None:
            Xa = (numpy.array(X)-numpy.array(self.clf.means))/numpy.array(self.clf.stds)
        else:
            Xa = X
        

Trailing whitespace

        value = self.clf.predict(Xa)
        return value