7 examples of 'sklearn cosine similarity' in Python

10
def cosine_similarity(x, y):
11
    numerator = sum(a * b for a, b in zip(x, y))
12
    denominator = square_rooted(x) * square_rooted(y)
13
    try:
14
        return numerator / float(denominator)
15
    except ZeroDivisionError:
16
        return 0.0

45
def test_cosine(self):
46
    """Do a quick basic test for index/search functionality"""
47
    data = [
48
        'hello world',
49
        'oh hello there',
50
        'Play it',
51
        'Play it again Sam',
52
    ]
53
54
    features = [dict([(x, 1) for x in f.split()]) for f in data]
55
    features = DictVectorizer().fit_transform(features)
56
57
    cluster_index = ci.ClusterIndex(features, data)
58
59
    ret = cluster_index.search(features, k=1, k_clusters=1,
60
                               return_distance=False)
61
    self.assertEqual([[d] for d in data], ret)

81
def test_cosine_identical(self):
82
    cosine = CosineTextSimilarity(self.ilist)
83
    cosine_sim = cosine(self.ilist[0], self.ilist[0])
84
    self.assertAlmostEqual(cosine_sim, 1, places=5)

158
def cosine_distance(s1, s2, k):
159
    """Compute the cosine difference of the strings as kmer vectors
160
    """
161
    vec1, vec2 = to_kmer_vector(s1, s2, k)
162
163
    intersection = set(vec1.keys()) & set(vec2.keys())
164
    numerator = sum([vec1[x] * vec2[x] for x in intersection])
165
166
    sum1 = sum([vec1[x] ** 2 for x in vec1.keys()])
167
    sum2 = sum([vec2[x] ** 2 for x in vec2.keys()])
168
    denominator = math.sqrt(sum1) * math.sqrt(sum2)
169
    if not denominator:
170
        return 0.0
171
    else:
172
        return float(numerator) / denominator

452
def cosine(vec1, vec2):
453
    vec1=debug_print(vec1, 'vec1')
454
    vec2=debug_print(vec2, 'vec2')
455
    norm_uni_l=T.sqrt((vec1**2).sum())
456
    norm_uni_r=T.sqrt((vec2**2).sum())
457
    
458
    dot=T.dot(vec1,vec2.T)
459
    
460
    simi=debug_print(dot/(norm_uni_l*norm_uni_r), 'uni-cosine')
461
    return simi.reshape((1,1))

1184
@keras_export(
1185
    'keras.losses.cosine_similarity',
1186
    v1=[
1187
        'keras.metrics.cosine_proximity',
1188
        'keras.metrics.cosine',
1189
        'keras.losses.cosine_proximity',
1190
        'keras.losses.cosine',
1191
        'keras.losses.cosine_similarity',
1192
    ])
1193
def cosine_similarity(y_true, y_pred, axis=-1):
1194
  """Computes the cosine similarity between labels and predictions.
1195
1196
  Note that it is a negative quantity between -1 and 0, where 0 indicates
1197
  orthogonality and values closer to -1 indicate greater similarity. This makes
1198
  it usable as a loss function in a setting where you try to maximize the
1199
  proximity between predictions and targets.
1200
1201
  `loss = -sum(y_true * y_pred)`
1202
1203
  Args:
1204
    y_true: Tensor of true targets.
1205
    y_pred: Tensor of predicted targets.
1206
    axis: Axis along which to determine similarity.
1207
1208
  Returns:
1209
    Cosine similarity tensor.
1210
  """
1211
  y_true = nn.l2_normalize(y_true, axis=axis)
1212
  y_pred = nn.l2_normalize(y_pred, axis=axis)
1213
  return -math_ops.reduce_sum(y_true * y_pred, axis=axis)

307
def CosineSimilarity(v1, v2):
308
  """ Implements the Cosine similarity metric.
309
   This is the recommended metric in the LaSSI paper
310
311
  **Arguments**:
312
313
    - two vectors (sequences of bit ids)
314
315
  **Returns**: a float.
316
317
  **Notes**
318
319
    - the vectors must be sorted
320
321
  >>> print('%.3f'%CosineSimilarity( (1,2,3,4,10), (2,4,6) ))
322
  0.516
323
  >>> print('%.3f'%CosineSimilarity( (1,2,2,3,4), (2,2,4,5,6) ))
324
  0.714
325
  >>> print('%.3f'%CosineSimilarity( (1,2,2,3,4), (1,2,2,3,4) ))
326
  1.000
327
  >>> print('%.3f'%CosineSimilarity( (1,2,2,3,4), (5,6,7) ))
328
  0.000
329
  >>> print('%.3f'%CosineSimilarity( (1,2,2,3,4), () ))
330
  0.000
331
332
  """
333
  d1 = Dot(v1, v1)
334
  d2 = Dot(v2, v2)
335
  denom = math.sqrt(d1 * d2)
336
  if not denom:
337
    res = 0.0
338
  else:
339
    numer = Dot(v1, v2)
340
    res = numer / denom
341
  return res