def createInstance(self):
bayes = orange.BayesLearner()
if hasattr(self, "estimatorConstructor"):
bayes.estimatorConstructor = self.estimatorConstructor
if hasattr(self, "m"):
if hasattr(bayes.estimatorConstructor, "m"):
raise AttributeError(
"invalid combination of attributes: 'estimatorConstructor' does not expect 'm'"
)
else:
self.estimatorConstructor.m = self.m
elif hasattr(self, "m"):
bayes.estimatorConstructor = orange.ProbabilityEstimatorConstructor_m(
m=self.m)
if hasattr(self, "conditionalEstimatorConstructor"):
bayes.conditionalEstimatorConstructor = self.conditionalEstimatorConstructor
elif bayes.estimatorConstructor:
bayes.conditionalEstimatorConstructor = orange.ConditionalProbabilityEstimatorConstructor_ByRows(
)
bayes.conditionalEstimatorConstructor.estimatorConstructor = bayes.estimatorConstructor
if hasattr(self, "conditionalEstimatorConstructorContinuous"):
bayes.conditionalEstimatorConstructorContinuous = self.conditionalEstimatorConstructorContinuous
return bayes
def test_m(self):
pec = orange.ProbabilityEstimatorConstructor_m(m=3)
pe = pec([22, 5, 2], [1, 1, 1])
self.assertEqual(list(pe()), [23 / 32, 6 / 32, 3 / 32])
pe = pec([22, 5, 2], [6, 2, 2])
self.assertEqual(list(pe()), [(22 + .6 * 3) / 32, (5 + .2 * 3) / 32,
(2 + .2 * 3) / 32])
self.assertRaises(ValueError, pec, [22, 5, 2])
def __init__(self, parent=None, signalManager = None, name='NaiveBayes'):
OWWidget.__init__(self, parent, signalManager, name, wantMainArea = 0, resizingEnabled = 0)
self.inputs = [("Data", ExampleTable, self.setData), ("Preprocess", PreprocessedLearner, self.setPreprocessor)]
self.outputs = [("Learner", orange.Learner),("Naive Bayesian Classifier", orange.BayesClassifier)]
self.m_estimator = orange.ProbabilityEstimatorConstructor_m()
self.estMethods=[("Relative Frequency", orange.ProbabilityEstimatorConstructor_relative()),
("Laplace", orange.ProbabilityEstimatorConstructor_Laplace()),
#("m-Estimate", self.m_estimator)
]
self.condEstMethods=[("<same as above>", None),
("Relative Frequency", orange.ConditionalProbabilityEstimatorConstructor_ByRows(estimatorConstructor=orange.ProbabilityEstimatorConstructor_relative())),
("Laplace", orange.ConditionalProbabilityEstimatorConstructor_ByRows(estimatorConstructor=orange.ProbabilityEstimatorConstructor_Laplace())),
("m-Estimate", orange.ConditionalProbabilityEstimatorConstructor_ByRows(estimatorConstructor=self.m_estimator))]
self.m_estimator.m = 2.0
self.name = 'Naive Bayes'
self.probEstimation = 0
self.condProbEstimation = 0
self.adjustThreshold = 0
self.windowProportion = 0.5
self.loessPoints = 100
self.preprocessor = None
self.data = None
self.loadSettings()
OWGUI.lineEdit(self.controlArea, self, 'name', box='Learner/Classifier Name', \
tooltip='Name to be used by other widgets to identify your learner/classifier.')
OWGUI.separator(self.controlArea)
glay = QGridLayout()
box = OWGUI.widgetBox(self.controlArea, 'Probability estimation', orientation = glay)
#glay.addWidget(OWGUI.separator(box, height=5), 0, 0)
glay.addWidget(OWGUI.widgetLabel(box, "Prior"), 1, 0)
glay.addWidget(OWGUI.comboBox(box, self, 'probEstimation', items=[e[0] for e in self.estMethods], tooltip='Method for estimating the prior probability.'),
1, 2)
glay.addWidget(OWGUI.widgetLabel(box, "Conditional (for discrete)"), 2, 0)
glay.addWidget(OWGUI.comboBox(box, self, 'condProbEstimation', items=[e[0] for e in self.condEstMethods], tooltip='Conditional probability estimation method used for discrete attributes.', callback=self.refreshControls),
2, 2)
glay.addWidget(OWGUI.widgetLabel(box, " " + "Parameter for m-estimate" + " "), 3, 0)
mValid = QDoubleValidator(self.controlArea)
mValid.setRange(0,10000,1)
self.mwidget = OWGUI.lineEdit(box, self, 'm_estimator.m', valueType = float, validator = mValid)
glay.addWidget(self.mwidget, 3, 2)
glay.addWidget(OWGUI.separator(box), 4, 0)
glay.addWidget(OWGUI.widgetLabel(box, 'Size of LOESS window'), 5, 0)
kernelSizeValid = QDoubleValidator(self.controlArea)
kernelSizeValid.setRange(0,1,3)
glay.addWidget(OWGUI.lineEdit(box, self, 'windowProportion',
tooltip='Proportion of examples used for local learning in loess.\nUse 0 to learn from few local instances (3) and 1 to learn from all in the data set (this kind of learning is not local anymore).',
valueType = float, validator = kernelSizeValid),
5, 2)
glay.addWidget(OWGUI.widgetLabel(box, 'LOESS sample points'), 6, 0)
pointsValid = QIntValidator(20, 1000, self.controlArea)
glay.addWidget(OWGUI.lineEdit(box, self, 'loessPoints',
tooltip='Number of points in computation of LOESS (20-1000).',
valueType = int, validator = pointsValid),
6, 2)
OWGUI.separator(self.controlArea)
OWGUI.checkBox(self.controlArea, self, "adjustThreshold", "Adjust threshold (for binary classes)", box = "Threshold")
OWGUI.separator(self.controlArea)
# box = OWGUI.widgetBox(self.controlArea, "Apply", orientation=1)
applyButton = OWGUI.button(self.controlArea, self, "&Apply", callback=self.applyLearner, default=True)
OWGUI.rubber(self.controlArea)
self.refreshControls()
self.applyLearner()
# Description: Shows how to use probability estimators with measure of attribute quality
# Category: attribute quality
# Classes: MeasureAttribute, MeasureAttribute_info, ProbabilityEstimatorConstructor_m, ConditionalProbabilityEstimatorConstructor_ByRows
# Uses: lenses
# Referenced: MeasureAttribute.htm
import orange
data = orange.ExampleTable("lenses")
ms = (0, 2, 5, 10, 20)
measures = []
for m in ms:
meas = orange.MeasureAttribute_info()
meas.estimatorConstructor = orange.ProbabilityEstimatorConstructor_m(m=m)
meas.conditionalEstimatorConstructor = orange.ConditionalProbabilityEstimatorConstructor_ByRows(
)
meas.conditionalEstimatorConstructor.estimatorConstructor = meas.estimatorConstructor
measures.append(meas)
print "%15s\t%5i\t%5i\t%5i\t%5i\t%5i\t" % (("attr", ) + ms)
for attr in data.domain.attributes:
print "%15s\t%5.3f\t%5.3f\t%5.3f\t%5.3f\t%5.3f" % (
(attr.name, ) + tuple([meas(attr, data) for meas in measures]))