def _csr_normalize_columns(m):
factor = m.sum(0)
nnzeros = np.where(factor > 0)
factor[nnzeros] = 1.0 / factor[nnzeros]
factor = np.array(factor)[0]
spt.csr_scale_columns(m.shape[0], m.shape[1], m.indptr, m.indices, m.data, factor)
return m
def _csr_normalize_columns(m):
factor = m.sum(0)
nnzeros = np.where(factor > 0)
factor[nnzeros] = 1.0 / factor[nnzeros]
factor = np.array(factor)[0]
spt.csr_scale_columns(m.shape[0], m.shape[1], m.indptr, m.indices, m.data,
factor)
return m
def normalize_sets_sparse (train, test):
norm = train.copy()
norm.data **= 2 # Square every value
norm = norm.sum(axis=0) # Sum every column
n_nonzeros = np.where(norm > 0)
norm[n_nonzeros] = 1.0 / np.sqrt(norm[n_nonzeros])
norm = np.array(norm).T[0]
sparsetools.csr_scale_columns(train.shape[0], train.shape[1], train.indptr, train.indices, train.data, norm)
sparsetools.csr_scale_columns(test.shape[0], test.shape[1], test.indptr, test.indices, test.data, norm)
return train, test
def normalize_sets_sparse(train, test):
norm = train.copy()
norm.data **= 2 # Square every value
norm = norm.sum(axis=0) # Sum every column
n_nonzeros = np.where(norm > 0)
norm[n_nonzeros] = 1.0 / np.sqrt(norm[n_nonzeros])
norm = np.array(norm).T[0]
sparsetools.csr_scale_columns(train.shape[0], train.shape[1], train.indptr,
train.indices, train.data, norm)
sparsetools.csr_scale_columns(test.shape[0], test.shape[1], test.indptr,
test.indices, test.data, norm)
return train, test
def test_scale_rows_and_cols(self):
D = matrix([[1,0,0,2,3],
[0,4,0,5,0],
[0,0,6,7,0]])
#TODO expose through function
S = csr_matrix(D)
v = array([1,2,3])
csr_scale_rows(3,5,S.indptr,S.indices,S.data,v)
assert_equal(S.todense(), diag(v)*D )
S = csr_matrix(D)
v = array([1,2,3,4,5])
csr_scale_columns(3,5,S.indptr,S.indices,S.data,v)
assert_equal(S.todense(), D*diag(v) )
# blocks
E = kron(D,[[1,2],[3,4]])
S = bsr_matrix(E,blocksize=(2,2))
v = array([1,2,3,4,5,6])
bsr_scale_rows(3,5,2,2,S.indptr,S.indices,S.data,v)
assert_equal(S.todense(), diag(v)*E )
S = bsr_matrix(E,blocksize=(2,2))
v = array([1,2,3,4,5,6,7,8,9,10])
bsr_scale_columns(3,5,2,2,S.indptr,S.indices,S.data,v)
assert_equal(S.todense(), E*diag(v) )
E = kron(D,[[1,2,3],[4,5,6]])
S = bsr_matrix(E,blocksize=(2,3))
v = array([1,2,3,4,5,6])
bsr_scale_rows(3,5,2,3,S.indptr,S.indices,S.data,v)
assert_equal(S.todense(), diag(v)*E )
S = bsr_matrix(E,blocksize=(2,3))
v = array([1,2,3,4,5,6,7,8,9,10,11,12,13,14,15])
bsr_scale_columns(3,5,2,3,S.indptr,S.indices,S.data,v)
assert_equal(S.todense(), E*diag(v) )
def csr_scale_columns(A, x):
sparsetools.csr_scale_columns(A.shape[0], A.shape[1], A.indptr, A.indices,
A.data, x)
def _csr_scale_columns(m, scale):
spt.csr_scale_columns(m.shape[0], m.shape[1], m.indptr, m.indices, m.data, scale)
def _csr_scale_columns(m, scale):
spt.csr_scale_columns(m.shape[0], m.shape[1], m.indptr, m.indices, m.data,
scale)
def csr_scale_columns(A, x):
sparsetools.csr_scale_columns(A.shape[0], A.shape[1],
A.indptr, A.indices, A.data,
x)