def _ttm_compute(self, V, mode, transp):
Z = self.unfold(mode, transp=True).tocsr()
if transp:
V = V.T
Z = Z.dot(V.T)
shape = copy(self.shape)
shape[mode] = V.shape[0]
if issparse_mat(Z):
newT = unfolded_sptensor((Z.data, (Z.row, Z.col)), [mode], None, shape=shape).fold()
else:
newT = unfolded_dtensor(Z.T, mode, shape).fold()
return newT
def nvecs(X, n, rank, do_flipsign=True, use_eigsh=False, dtype=np.float):
"""
Eigendecomposition of mode-n unfolding of a tensor
Parameters
----------
X : sptensor or dtensor
n : int
rank : int
number of eigenvectors to return
do_flipsign: bool, optional
Flip sign of factor matrices such that largest magnitude
element will be positive
use_eigsh : bool, optional
if True, use the Implicitly Restarted Lanczos Method (scipy.sparse.linalg.eigsh)
to find eigenvectors of matrix unfolding.
eigsh is effective for large eigenvalues of sparse matrices,
but works poorly for dense matrices or when many eigenvalues are needed.
dtype: data-type, optional
"""
Xn = X.unfold(n)
if issparse_mat(Xn):
Xn = csr_matrix(Xn, dtype=dtype)
Y = Xn.dot(Xn.T)
if not use_eigsh:
if issparse_mat(Y):
Y = Y.todense()
N = Y.shape[0]
_, U = eigh(Y, eigvals=(N - rank, N - 1))
#_, U = eigsh(Y, rank, which='LM')
else:
_, U = eigsh(Y, rank, which='LM')
# reverse order of eigenvectors such that eigenvalues are decreasing
U = array(U[:, ::-1])
# flip sign
if do_flipsign:
U = flipsign(U)
return U
def _ttm_compute(self, V, mode, transp):
Z = self.unfold(mode, transp=True).tocsr()
if transp:
V = V.T
Z = Z.dot(V.T)
shape = np.copy(self.shape)
shape[mode] = V.shape[0]
if issparse_mat(Z):
newT = unfolded_sptensor((Z.data, (Z.row, Z.col)), [mode],
None,
shape=shape).fold()
else:
newT = unfolded_dtensor(Z.T, mode, shape).fold()
return newT
def nvecs(X, n, rank, do_flipsign=True):
"""
Eigendecomposition of mode-n unfolding of a tensor
"""
Xn = X.unfold(n)
Y = Xn.dot(Xn.T)
if issparse_mat(Y):
_, U = eigsh(Y, rank, which='LM')
else:
N = Y.shape[0]
_, U = eigh(Y, eigvals=(N - rank, N - 1))
#_, U = eigsh(Y, rank, which='LM')
# reverse order of eigenvectors such that eigenvalues are decreasing
U = array(U[:, ::-1])
# flip sign
if do_flipsign:
U = flipsign(U)
return U
def nvecs(X, n, rank, do_flipsign=True, dtype=np.float):
"""
Eigendecomposition of mode-n unfolding of a tensor
"""
Xn = X.unfold(n)
if issparse_mat(Xn):
Xn = csr_matrix(Xn, dtype=dtype)
Y = Xn.dot(Xn.T)
_, U = eigsh(Y, rank, which='LM')
else:
Y = Xn.dot(Xn.T)
N = Y.shape[0]
_, U = eigh(Y, eigvals=(N - rank, N - 1))
# reverse order of eigenvectors such that eigenvalues are decreasing
U = np.array(U[:, ::-1])
# flip sign
if do_flipsign:
U = flipsign(U)
return U