def __trainSkipGram(self):
"""
Main method for training the SkipGram version of Word2Vec algorithm.
"""
iteration = Iteration(self.__corpus, self.__parameter)
currentSentence = self.__corpus.getSentence(
iteration.getSentenceIndex())
outputs = Vector()
outputs.initAllSame(self.__parameter.getLayerSize(), 0.0)
outputUpdate = Vector()
outputUpdate.initAllSame(self.__parameter.getLayerSize(), 0)
self.__corpus.shuffleSentences(1)
while iteration.getIterationCount(
) < self.__parameter.getNumberOfIterations():
iteration.alphaUpdate()
wordIndex = self.__vocabulary.getPosition(
currentSentence.getWord(iteration.getSentencePosition()))
currentWord = self.__vocabulary.getWord(wordIndex)
outputs.clear()
outputUpdate.clear()
b = randrange(self.__parameter.getWindow())
for a in range(b, self.__parameter.getWindow() * 2 + 1 - b):
c = iteration.getSentencePosition(
) - self.__parameter.getWindow() + a
if a != self.__parameter.getWindow(
) and currentSentence.safeIndex(c):
lastWordIndex = self.__vocabulary.getPosition(
currentSentence.getWord(c))
l1 = lastWordIndex
outputUpdate.clear()
if self.__parameter.isHierarchicalSoftMax():
for d in range(currentWord.getCodeLength()):
l2 = currentWord.getPoint(d)
f = self.__wordVectors.getRowVector(l1).dotProduct(
self.__wordVectorUpdate.getRowVector(l2))
if f <= -NeuralNetwork.MAX_EXP or f >= NeuralNetwork.MAX_EXP:
continue
else:
f = self.__expTable[int(
(f + NeuralNetwork.MAX_EXP) *
(NeuralNetwork.EXP_TABLE_SIZE //
NeuralNetwork.MAX_EXP // 2))]
g = (1 - currentWord.getCode(d) -
f) * iteration.getAlpha()
outputUpdate.addVector(
self.__wordVectorUpdate.getRowVector(
l2).product(g))
self.__wordVectorUpdate.addRowVector(
l2,
self.__wordVectors.getRowVector(l1).product(g))
else:
for d in range(
self.__parameter.getNegativeSamplingSize() +
1):
if d == 0:
target = wordIndex
label = 1
else:
target = self.__vocabulary.getTableValue(
randrange(
self.__vocabulary.getTableSize()))
if target == 0:
target = randrange(
self.__vocabulary.size() - 1) + 1
if target == wordIndex:
continue
label = 0
l2 = target
f = self.__wordVectors.getRowVector(l1).dotProduct(
self.__wordVectorUpdate.getRowVector(l2))
g = self.__calculateG(f, iteration.getAlpha(),
label)
outputUpdate.addVector(
self.__wordVectorUpdate.getRowVector(
l2).product(g))
self.__wordVectorUpdate.addRowVector(
l2,
self.__wordVectors.getRowVector(l1).product(g))
self.__wordVectors.addRowVector(l1, outputUpdate)
currentSentence = iteration.sentenceUpdate(currentSentence)
class VectorTest(unittest.TestCase):
data1 = [2, 3, 4, 5, 6]
def setUp(self):
data2 = [8, 7, 6, 5, 4]
self.smallVector1 = Vector(self.data1)
self.smallVector2 = Vector(data2)
largeData1 = []
for i in range(1, 1001):
largeData1.append(i)
self.largeVector1 = Vector(largeData1)
largeData2 = []
for i in range(1, 1001):
largeData2.append(1000 - i + 1)
self.largeVector2 = Vector(largeData2)
def test_Biased(self):
biased = self.smallVector1.biased()
self.assertEqual(1, biased.getValue(0))
self.assertEqual(self.smallVector1.size() + 1, biased.size())
def test_ElementAdd(self):
self.smallVector1.add(7)
self.assertEqual(7, self.smallVector1.getValue(5))
self.assertEqual(6, self.smallVector1.size())
self.smallVector1.remove(5)
def test_Insert(self):
self.smallVector1.insert(3, 6)
self.assertEqual(6, self.smallVector1.getValue(3))
self.assertEqual(6, self.smallVector1.size())
self.smallVector1.remove(3)
def test_Remove(self):
self.smallVector1.remove(2)
self.assertEqual(5, self.smallVector1.getValue(2))
self.assertEqual(4, self.smallVector1.size())
self.smallVector1.insert(2, 4)
def test_SumOfElementsSmall(self):
self.assertEqual(20, self.smallVector1.sumOfElements())
self.assertEqual(30, self.smallVector2.sumOfElements())
def test_SumOfElementsLarge(self):
self.assertEqual(20, self.smallVector1.sumOfElements())
self.assertEqual(30, self.smallVector2.sumOfElements())
self.assertEqual(500500, self.largeVector1.sumOfElements())
self.assertEqual(500500, self.largeVector2.sumOfElements())
def test_MaxIndex(self):
self.assertEqual(4, self.smallVector1.maxIndex())
self.assertEqual(0, self.smallVector2.maxIndex())
def test_Sigmoid(self):
smallVector3 = Vector(self.data1)
smallVector3.sigmoid()
self.assertAlmostEqual(0.8807971, smallVector3.getValue(0), 6)
self.assertAlmostEqual(0.9975274, smallVector3.getValue(4), 6)
def test_SkipVectorSmall(self):
smallVector3 = self.smallVector1.skipVector(2, 0)
self.assertEqual(2, smallVector3.getValue(0))
self.assertEqual(6, smallVector3.getValue(2))
smallVector3 = self.smallVector1.skipVector(3, 1)
self.assertEqual(3, smallVector3.getValue(0))
self.assertEqual(6, smallVector3.getValue(1))
def test_SkipVectorLarge(self):
largeVector3 = self.largeVector1.skipVector(2, 0)
self.assertEqual(250000, largeVector3.sumOfElements())
largeVector3 = self.largeVector1.skipVector(5, 3)
self.assertEqual(100300, largeVector3.sumOfElements())
def test_VectorAddSmall(self):
self.smallVector1.addVector(self.smallVector2)
self.assertEqual(50, self.smallVector1.sumOfElements())
self.smallVector1.subtract(self.smallVector2)
def test_VectorAddLarge(self):
self.largeVector1.addVector(self.largeVector2)
self.assertEqual(1001000, self.largeVector1.sumOfElements())
self.largeVector1.subtract(self.largeVector2)
def test_SubtractSmall(self):
self.smallVector1.subtract(self.smallVector2)
self.assertEqual(-10, self.smallVector1.sumOfElements())
self.smallVector1.addVector(self.smallVector2)
def test_SubtractLarge(self):
self.largeVector1.subtract(self.largeVector2)
self.assertEqual(0, self.largeVector1.sumOfElements())
self.largeVector1.addVector(self.largeVector2)
def test_DifferenceSmall(self):
smallVector3 = self.smallVector1.difference(self.smallVector2)
self.assertEqual(-10, smallVector3.sumOfElements())
def test_DifferenceLarge(self):
largeVector3 = self.largeVector1.difference(self.largeVector2)
self.assertEqual(0, largeVector3.sumOfElements())
def test_DotProductWithVectorSmall(self):
dotProduct = self.smallVector1.dotProduct(self.smallVector2)
self.assertEqual(110, dotProduct)
def test_DotProductWithVectorLarge(self):
dotProduct = self.largeVector1.dotProduct(self.largeVector2)
self.assertEqual(167167000, dotProduct)
def test_DotProductWithItselfSmall(self):
dotProduct = self.smallVector1.dotProductWithSelf()
self.assertEqual(90, dotProduct)
def test_DotProductWithItselfLarge(self):
dotProduct = self.largeVector1.dotProductWithSelf()
self.assertEqual(333833500, dotProduct)
def test_ElementProductSmall(self):
smallVector3 = self.smallVector1.elementProduct(self.smallVector2)
self.assertEqual(110, smallVector3.sumOfElements())
def test_ElementProductLarge(self):
largeVector3 = self.largeVector1.elementProduct(self.largeVector2)
self.assertEqual(167167000, largeVector3.sumOfElements())
def test_Divide(self):
self.smallVector1.divide(10.0)
self.assertEqual(2, self.smallVector1.sumOfElements())
self.smallVector1.multiply(10.0)
def test_Multiply(self):
self.smallVector1.multiply(10.0)
self.assertEqual(200, self.smallVector1.sumOfElements())
self.smallVector1.divide(10.0)
def test_Product(self):
smallVector3 = self.smallVector1.product(7.0)
self.assertEqual(140, smallVector3.sumOfElements())
def test_L1NormalizeSmall(self):
self.smallVector1.l1Normalize()
self.assertEqual(1.0, self.smallVector1.sumOfElements())
self.smallVector1.multiply(20)
def test_L1NormalizeLarge(self):
self.largeVector1.l1Normalize()
self.assertEqual(1.0, self.largeVector1.sumOfElements())
self.largeVector1.multiply(500500)
def test_L2NormSmall(self):
norm = self.smallVector1.l2Norm()
self.assertEqual(norm, math.sqrt(90))
def test_L2NormLarge(self):
norm = self.largeVector1.l2Norm()
self.assertEqual(norm, math.sqrt(333833500))
def test_cosineSimilaritySmall(self):
similarity = self.smallVector1.cosineSimilarity(self.smallVector2)
self.assertAlmostEqual(0.8411910, similarity, 6)
def test_cosineSimilarityLarge(self):
similarity = self.largeVector1.cosineSimilarity(self.largeVector2)
self.assertAlmostEqual(0.5007497, similarity, 6)
