def validationDialog(self, parent):
"""Return the classifier specific (cross-)validation widget."""
if self._cvwidget is None:
self._cvwidget = CrossValidationDialog(parent, self)
return self._cvwidget
class Svc(Classifier):
KERNEL = "rbf"
name = "Support Vector Classifier"
def __init__(self, *args, **kw):
super(Svc, self).__init__(*args, **kw)
self._pp = None
self._cvwidget = None
self._pwidget = None
self._clf = sklearn.svm.SVC(C=1.0,
kernel=self.KERNEL,
gamma=0.0,
probability=True)
def parameterWidget(self, parent):
"""Returns the classifier specific parameter widget."""
if self._pwidget is None:
self._pwidget = SvcParameterWidget(parent)
return self._pwidget
def validationDialog(self, parent):
"""Return the classifier specific (cross-)validation widget."""
if self._cvwidget is None:
self._cvwidget = CrossValidationDialog(parent, self)
return self._cvwidget
def description(self):
if self._cvwidget is None:
return ""
else:
return self._cvwidget.text()
def train(self, features, labels):
self.setupPreProcessor(features)
self._clf.fit(self._pp(features), labels)
def predict(self, features):
if self._clf is None:
return super(Svc, self).predict(features)
else:
features = self._pp(features)
proba = self._clf.predict_proba(features)
# turn indices into class labels, asumeing ascending order
predictions = [
self._classes.keys()[i] for i in np.argmax(proba, axis=1)
]
classes = [self.classes[pred].clone() for pred in predictions]
for c, p in zip(classes, proba):
c.score = dict(
((k, v) for k, v in zip(self.classes.keys(), p)))
return classes
def saveToHdf(self,
name,
file_,
feature_selection,
description,
overwrite=False,
labels=None,
sample_info=None):
if self._cvwidget is None:
raise HdfError(("You need to validate the classifier first\n"
"Open the Cross validation dialog"))
writer = SvcWriter(name, file_, description, overwrite)
writer.saveTrainingSet(self._pp.data, feature_selection.values())
writer.saveAnnotations(labels)
writer.saveClassDef(self.classes, self._clf.get_params())
writer.saveNormalization(self._pp)
writer.saveConfusionMatrix(self._cvwidget.confusion_matrix)
writer.saveSampleInfo(sample_info)
writer.flush()
class Svc(Classifier):
KERNEL = "rbf"
name = "Support Vector Classifier"
def __init__(self, *args, **kw):
super(Svc, self).__init__(*args, **kw)
self._pp = None
self._cvwidget = None
self._pwidget = None
self._clf = sklearn.svm.SVC(C=1.0, kernel=self.KERNEL, gamma=0.0,
probability=True)
def parameterWidget(self, parent=None):
"""Returns the classifier specific parameter widget."""
if self._pwidget is None:
self._pwidget = SvcParameterWidget(parent)
return self._pwidget
def validationDialog(self, parent):
"""Return the classifier specific (cross-)validation widget."""
if self._cvwidget is None:
self._cvwidget = CrossValidationDialog(parent, self)
return self._cvwidget
def description(self):
if self._cvwidget is None:
return ""
else:
return self._cvwidget.text()
def checkParameters(self, parent):
"""Pop up a dialog and ask the user wether to proceed with current
unoptimized parameters, run cross validation or cancel.
"""
title = "Grid search & Cross Validation"
text = "You need to run Grid Search & Cross Validation before you predict"
QApplication.restoreOverrideCursor()
qmb = QMessageBox(QMessageBox.Question, title, text, parent=parent)
ignoreBtn = qmb.addButton(qmb.Ignore)
cancelBtn = qmb.addButton(qmb.Cancel)
runBtn = qmb.addButton(
"Run Grid Search", qmb.AcceptRole)
qmb.exec_()
QApplication.setOverrideCursor(QCursor(Qt.WaitCursor))
if qmb.clickedButton() == runBtn:
# parent sets some flag
parent.validateClassifier()
return True
elif qmb.clickedButton() == ignoreBtn:
return True
elif qmb.clickedButton() == cancelBtn:
return False
raise RuntimeError()
def train(self, features, labels):
self.setupPreProcessor(features)
self._clf.fit(self._pp(features), labels)
def predict(self, features, dummy=None):
if self._clf is None:
return super(Svc, self).predict(features)
else:
features = self._pp(features)
proba = self._clf.predict_proba(features)
# turn indices into class labels, asumeing ascending order
predictions = [self._classes.keys()[i]
for i in np.argmax(proba, axis=1)]
classes = [self.classes[pred].clone() for pred in predictions]
for c, p in zip(classes, proba):
c.score = dict(
((k, v) for k, v in zip(self.classes.keys(), p)))
return classes
def saveToHdf(self, name, file_, feature_selection, description,
overwrite=False, labels=None, sample_info=None):
if self._cvwidget is None:
raise HdfError(("You need to validate the classifier first\n"
"Open the Cross validation dialog"))
writer = SvcWriter(name, file_, description, overwrite)
writer.saveTrainingSet(self._pp.data, feature_selection.values())
writer.saveAnnotations(labels)
writer.saveClassDef(self.classes, self._clf.get_params())
writer.saveNormalization(self._pp)
writer.saveConfusionMatrix(self._cvwidget.confusion_matrix)
writer.saveSampleInfo(sample_info)
writer.flush()
def validationDialog(self, parent):
"""Return the classifier specific (cross-)validation widget."""
if self._cvwidget is None:
self._cvwidget = CrossValidationDialog(parent, self)
return self._cvwidget
class Svc(Classifier):
KERNEL = "rbf"
name = "Support Vector Classifier"
def __init__(self, *args, **kw):
super(Svc, self).__init__(*args, **kw)
self._pp = None
self._cvwidget = None
self._pwidget = None
self._clf = sklearn.svm.SVC(C=1.0, kernel=self.KERNEL, gamma=0.0,
probability=True)
def parameterWidget(self, parent):
"""Returns the classifier specific parameter widget."""
if self._pwidget is None:
self._pwidget = SvcParameterWidget(parent)
return self._pwidget
def validationDialog(self, parent):
"""Return the classifier specific (cross-)validation widget."""
if self._cvwidget is None:
self._cvwidget = CrossValidationDialog(parent, self)
return self._cvwidget
def description(self):
if self._cvwidget is None:
return ""
else:
return self._cvwidget.text()
def train(self, features, labels):
self.setupPreProcessor(features)
self._clf.fit(self._pp(features), labels)
def predict(self, features):
if self._clf is None:
return super(Svc, self).predict(features)
else:
features = self._pp(features)
proba = self._clf.predict_proba(features)
# turn indices into class labels, asumeing ascending order
predictions = [self._classes.keys()[i]
for i in np.argmax(proba, axis=1)]
classes = [self.classes[pred].clone() for pred in predictions]
for c, p in zip(classes, proba):
c.score = dict(
((k, v) for k, v in zip(self.classes.keys(), p)))
return classes
def saveToHdf(self, name, file_, feature_selection, description,
overwrite=False, labels=None, sample_info=None):
if self._cvwidget is None:
raise HdfError(("You need to validate the classifier first\n"
"Open the Cross validation dialog"))
writer = SvcWriter(name, file_, description, overwrite)
writer.saveTrainingSet(self._pp.data, feature_selection.values())
writer.saveAnnotations(labels)
writer.saveClassDef(self.classes, self._clf.get_params())
writer.saveNormalization(self._pp)
writer.saveConfusionMatrix(self._cvwidget.confusion_matrix)
writer.saveSampleInfo(sample_info)
writer.flush()
