def score(self, Kyx, cy):
if self.scenario == 'multiclass':
cy = tuple_labels_to_list_labels(cy)
return self._score_one_vs_one(Kyx, cy)
elif self.scenario == 'versus_null':
cy = tuple_labels_to_list_labels(cy)
return self._score_one_vs_rest(Kyx, cy)
elif self.scenario == 'versus_null_sklearn':
cy = tuple_labels_to_list_labels(cy)
return self._score_one_vs_rest_sklearn(Kyx, cy)
def score(self, Kyx, cy):
if self.scenario == "multiclass":
cy = tuple_labels_to_list_labels(cy)
return self._score_one_vs_one(Kyx, cy)
elif self.scenario == "versus_null":
cy = tuple_labels_to_list_labels(cy)
return self._score_one_vs_rest(Kyx, cy)
elif self.scenario == "versus_null_sklearn":
cy = tuple_labels_to_list_labels(cy)
return self._score_one_vs_rest_sklearn(Kyx, cy)
def fit(self, Kxx, cx):
if self.scenario == "multiclass":
cx = tuple_labels_to_list_labels(cx)
self._fit_one_vs_one(Kxx, cx)
elif self.scenario == "versus_null":
cx = tuple_labels_to_list_labels(cx)
self._fit_one_vs_rest(Kxx, cx)
elif self.scenario == "versus_null_sklearn":
cx = tuple_labels_to_list_labels(cx)
self._fit_one_vs_rest_sklearn(Kxx, cx)
return self
def fit(self, Kxx, cx):
if self.scenario == 'multiclass':
cx = tuple_labels_to_list_labels(cx)
self._fit_one_vs_one(Kxx, cx)
elif self.scenario == 'versus_null':
cx = tuple_labels_to_list_labels(cx)
self._fit_one_vs_rest(Kxx, cx)
elif self.scenario == 'versus_null_sklearn':
cx = tuple_labels_to_list_labels(cx)
self._fit_one_vs_rest_sklearn(Kxx, cx)
return self
def fit(self, Kxx, cx):
""" Fits SVM to kernel matrix and cross-validate the regularization
weight across a logarithmic scale.
"""
cx = tuple_labels_to_list_labels(cx)
my_svm = svm.SVC(kernel="precomputed")
c_values = np.power(3.0, np.arange(-2, 8))
tuned_parameters = [{"C": c_values}]
splits = StratifiedShuffleSplit(cx, 3, test_size=0.3)
self.clf = GridSearchCV(my_svm, tuned_parameters, score_func=zero_one_score, cv=splits, n_jobs=4)
self.clf.fit(Kxx, cx)
return self
def fit(self, Kxx, cx):
""" Fits SVM to kernel matrix and cross-validate the regularization
weight across a logarithmic scale.
"""
cx = tuple_labels_to_list_labels(cx)
my_svm = svm.SVC(kernel='precomputed')
c_values = np.power(3.0, np.arange(-2, 8))
tuned_parameters = [{'C': c_values}]
splits = StratifiedShuffleSplit(cx, 3, test_size=0.3)
self.clf = GridSearchCV(my_svm,
tuned_parameters,
score_func=zero_one_score,
cv=splits,
n_jobs=4)
self.clf.fit(Kxx, cx)
return self
def score(self, Kyx, cy):
""" Return the accuracy score / zero-one score. """
cy = tuple_labels_to_list_labels(cy)
return self.clf.score(Kyx, cy) * 100
def predict(self, Kyx, cy):
cy = tuple_labels_to_list_labels(cy)
if self.scenario == "multiclass":
return self._predict_one_vs_one(Kyx, cy)
return None
def score(self, Kyx, cy):
""" Return the accuracy score / zero-one score. """
cy = tuple_labels_to_list_labels(cy)
return self.clf.score(Kyx, cy) * 100
def predict(self, Kyx, cy):
cy = tuple_labels_to_list_labels(cy)
if self.scenario == 'multiclass':
return self._predict_one_vs_one(Kyx, cy)
return None
