def match(self, queryWords, docWords):
words = self.sharedWords(queryWords, docWords)
idf = [self.invDocFreq(w) for w in words]
docVec = self.wordVector(words, docWords)
maxFreq = docVec.largest()
docVec = Vector([docVec[i] / maxFreq * idf[i]
for i in range(len(words))])
queryVec = self.wordVector(words, queryWords)
maxFreq = queryVec.largest()
queryVec = Vector([0 if idf[i] == 0 else \
(0.5 + (0 if maxFreq == 0 else 0.5*queryVec[i]/maxFreq)) \
* idf[i] for i in range(len(words))])
return queryVec.dot(docVec) / \
(queryVec.length() * docVec.length())
def testCosineSimilarity(self):
matcher = matching.CosineSimilarity({
'docFreq': {
'a': 4,
'b': 2,
'c': 1,
},
'docCount': 16
})
# ---- test one -----
docWords = {'a': 13, 'c': 9}
queryWords = {'a': 7, 'b': 4}
idf = {w: matcher.invDocFreq(w) for w in ['a', 'b', 'c']}
docVec = Vector([13/13*idf['a'], 0, 9/13*idf['c']])
queryVec = Vector([(0.5+0.5*7/7)*idf['a'], (0.5+0.5*4/7)*idf['b'],
0.5*idf['c']])
scoreComputed = docVec[0]*queryVec[0] + docVec[1]*queryVec[1] + \
docVec[2]*queryVec[2]
scoreComputed = scoreComputed / docVec.length()
scoreComputed = scoreComputed / queryVec.length()
score = matcher.match(queryWords, docWords)
self.assertApproxEqual(score, scoreComputed)
# ---- test two -----
docWords = {'a': 13, 'b': 9, 'c': 1}
queryWords = {'a': 7, 'b': 4}
docVec = Vector([13/13*idf['a'], 9/13*idf['b'], 1/13*idf['c']])
queryVec = Vector([(0.5+0.5*7/7)*idf['a'], (0.5+0.5*4/7)*idf['b'],
0.5*idf['c']])
scoreComputed = docVec[0]*queryVec[0] + docVec[1]*queryVec[1] + \
docVec[2]*queryVec[2]
scoreComputed = scoreComputed / docVec.length()
scoreComputed = scoreComputed / queryVec.length()
score = matcher.match(queryWords, docWords)
self.assertApproxEqual(score, scoreComputed)
# ---- test three -----
docWords = {'a': 13, 'b': 9}
queryWords = {'a': 7, 'b': 4, 'c': 1}
docVec = Vector([13/13*idf['a'], 9/13*idf['b'], 0])
queryVec = Vector([(0.5+0.5*7/7)*idf['a'], (0.5+0.5*4/7)*idf['b'],
(0.5+0.5*1/7)*idf['c']])
scoreComputed = docVec[0]*queryVec[0] + docVec[1]*queryVec[1] + \
docVec[2]*queryVec[2]
scoreComputed = scoreComputed / docVec.length()
scoreComputed = scoreComputed / queryVec.length()
score = matcher.match(queryWords, docWords)
self.assertApproxEqual(score, scoreComputed)
def test_length(self): vec = Vector([3,4]) length = vec.length() self.assertEqual(length, 5)
