def cosine_sparse(csr1,
csr2,
topn=10,
min_value=0.4,
dense=True,
similarity=True,
expected_density=1):
"""Computes the cosine distance between the rows of `csr`,
smaller than the cut-off distance `epsilon`.
"""
if not dense:
csr1 = csr_matrix(csr1).astype(bool, copy=False).astype(int,
copy=False)
csr1 = csr1.astype('float64', copy=False)
csr1 = normalize(csr1, norm='l2', axis=1)
csr2 = csr_matrix(csr2).astype(bool, copy=False).astype(int,
copy=False)
csr2 = csr2.astype('float64', copy=False)
csr2 = normalize(csr2, norm='l2', axis=1)
intrsct = dot_product(csr1, csr2.T, topn, min_value, expected_density)
intrsct.data[intrsct.data >= 1] = 1
if not similarity:
intrsct.data = 1 - intrsct.data
else:
csr1 = csr_matrix(csr1).astype('float64', copy=False)
csr1 = normalize(csr1, norm='l2', axis=1)
csr2 = csr_matrix(csr2).astype('float64', copy=False)
csr2 = normalize(csr2, norm='l2', axis=1)
intrsct = dot_product(csr1, csr2.T, topn, min_value, expected_density)
if not similarity:
intrsct.data = 1 - intrsct.data
return intrsct
def pairwise_cosine_sparse_sim(csr1,csr2, topn=4000, min_value = 0.4, word_embedings = True):
"""Computes the cosine distance between the rows of `csr`,
smaller than the cut-off distance `epsilon`.
"""
if not word_embedings:
csr1 = csr_matrix(csr1).astype(bool,copy=False).astype(int,copy=False)
csr1 = csr1.astype('float64',copy=False)
csr1 = normalize(csr1, norm='l2', axis=1)
csr2 = csr_matrix(csr2).astype(bool,copy=False).astype(int,copy=False)
csr2 = csr2.astype('float64',copy=False)
csr2 = normalize(csr2, norm='l2', axis=1)
intrsct = dot_product(csr1,csr2.T, topn, min_value)
intrsct.data[intrsct.data>=1] = 1
else:
csr1 = csr_matrix(csr1).astype('float64',copy=False)
csr1 = normalize(csr1, norm='l2', axis=1)
csr2 = csr_matrix(csr2).astype('float64',copy=False)
csr2 = normalize(csr2, norm='l2', axis=1)
intrsct = dot_product(csr1,csr2.T, topn, min_value)
return intrsct
def pairwise_cosine_sparse(csr, topn=2000, min_value=0.1):
"""Computes the cosine distance between the rows of `csr`,
smaller than the cut-off distance `epsilon`.
"""
csr = csr_matrix(csr).astype(bool, copy=False).astype(int, copy=False)
csr = csr.astype('float64', copy=False)
csr = normalize(csr, norm='l2', axis=1)
intrsct = dot_product(csr, csr.T, topn, min_value)
intrsct.data[intrsct.data >= 1] = 1
intrsct.data = 1 - intrsct.data
return intrsct
def pairwise_cosine_sparse_sim(csr, topn=4000, min_value = 0.4,expected_density = 1, sparse = True,normalize = True):
"""Computes the cosine distance between the rows of `csr`,
smaller than the cut-off distance `epsilon`.
"""
if sparse:
csr = csr_matrix(csr).astype(bool,copy=False).astype(int,copy=False)
csr = csr.astype('float64',copy=False)
if normalize:
csr = normalize(csr, norm='l2', axis=1)
else:
pass
intrsct = dot_product(csr,csr.T, topn, min_value, expected_density)
#intrsct.data[intrsct.data>=1] = 1
else:
csr = csr.astype('float64',copy=False)
if normalize:
csr = normalize(csr, norm='l2', axis=1)
else:
pass
intrsct = dot_product(csr,csr.T, topn, min_value, expected_density)
intrsct.data[intrsct.data<=0] = 0
return intrsct
def pairwise_jaccard_sparse(csr, topn=2000, min_value=2):
"""Computes the Jaccard distance between the rows of `csr`,
smaller than the cut-off distance `epsilon`.
"""
csr = csr_matrix(csr).astype(bool, copy=False).astype(int, copy=False)
csr = csr.astype('float64', copy=False)
csr_rownnz = csr.getnnz(axis=1)
intrsct = dot_product(csr, csr.T, topn, min_value - 0.1)
nnz_i = np.repeat(csr_rownnz, intrsct.getnnz(axis=1))
unions = nnz_i + csr_rownnz[intrsct.indices] - intrsct.data
intrsct.data = 1.0 - intrsct.data / unions
return intrsct