def testAlignments(test_data, classifiers):
x0 = Symbol('x0')
featuresets = []
correctlyAlignedIndicesList = []
for (i, iIndex, wordproblem, equationTemplate, solution) in test_data:
try:
classifier = classifiers[str(equationTemplate)]
except KeyError:
continue
alignmentFeatures, correctlyAlignedIndices = extractAlignmentFeatures(wordproblem, equationTemplate, solution, i)
featuresets.append(alignmentFeatures)
alignedString = ''
for c in correctlyAlignedIndices:
alignedString += str(c)
correctlyAlignedIndicesList.append(alignedString)
prediction = classifier.predict(featuresets)
predictedAlignedIndices = []
correct = 0
for each in prediction:
item = []
for char in each:
item.append(int(char))
predictedAlignedIndices.append(item)
#print 'sdsdsdsdsdsdsd'
#print predictedAlignedIndices
predictedAlignedValues = []
for x in range(0, len(test_data)):
try:
i, iIndex,wordproblem, equationTemplate, solution = test_data[x]
numberVector = extractNumberVectorFromQuestion(wordproblem)
correctAlignValues = []
for each in predictedAlignedIndices[x]:
correctAlignValues.append(numberVector[each])
predictedAlignedValues.append(correctAlignValues)
#print 'correctalignedvalues'
#print correctAlignValues
result = solveEquations(equationTemplate, getNumberslots(equationTemplate), correctAlignValues)
if solution[0] == result[x0]:
correct += 1
except:
pass
#print 'all values'
#print predictedAlignedValues
for i in range(0,len(prediction)):
print correctlyAlignedIndicesList[i], prediction[i]
if correctlyAlignedIndicesList[i] == prediction[i]:
correct += 1
print 'Correct: ' + str(correct)
def trainAlignmentClassifierScikit(labeled_word_problems, algorithm):
featuresets = []
correctlyAlignedIndicesList = []
for (i, iIndex, wordproblem, equationTemplate, solution) in labeled_word_problems:
print iIndex
alignmentFeatures, correctlyAlignedIndices = extractAlignmentFeatures(wordproblem, equationTemplate, solution, i, 'train')
if len(correctlyAlignedIndices) != 0 and len(alignmentFeatures) != 0:
featuresets.append(alignmentFeatures)
alignedString = ''
for i in correctlyAlignedIndices:
alignedString += str(i)
correctlyAlignedIndicesList.append(alignedString)
#sfeaturesets = numpy.array(featuresets)
#print type(featuresets)
#correctlyAlignedIndicesList = numpy.array(correctlyAlignedIndicesList)
#print featuresets
#print correctlyAlignedIndicesList
classifier = None
if len(featuresets) != 0:
if algorithm == 'SVM':
classifier = SVC()
elif algorithm == 'NaiveBayes':
classifier = GaussianNB()
elif algorithm == 'DecisionTree':
classifier = DecisionTreeClassifier()
elif algorithm == 'MaxEnt':
classifier = LogisticRegression()
classifier.fit(featuresets, correctlyAlignedIndicesList)
return classifier
def trainAlignmentClassifier(labeled_word_problems, algorithm):
featuresets = []
for (i, iIndex, wordproblem, equationTemplate, solution) in labeled_word_problems:
alignmentFeatures, correctlyAlignedIndices = extractAlignmentFeatures(wordproblem, equationTemplate, solution, i)
if len(correctlyAlignedIndices) != 0:
featuresets.append((alignmentFeatures, correctlyAlignedIndices))
#featuresets = [(extractAlignmentFeatures(wordproblem, equationTemplate, solution), i)
#for (i, iIndex, wordproblem, equationTemplate, solution) in labeled_word_problems]
print featuresets
train_set = featuresets
if algorithm == 'DecisionTree':
classifier = nltk.DecisionTreeClassifier.train(train_set)
elif algorithm == 'NaiveBayes':
classifier = nltk.NaiveBayesClassifier.train(train_set)
elif algorithm == 'MaxEntMegam':
classifier = nltk.classify.MaxentClassifier.train(train_set, 'MEGAM', trace=0, max_iter=1)
elif algorithm == 'MaxEnt':
classifier = nltk.MaxentClassifier.train(train_set)
return classifier
