def dot(self, other):
"""
Matrix or scalar multiplication. Delegate to divisi2.operators to
sort it out.
"""
from divisi2 import operators
return operators.dot(self, other)
def test_incremental_svd():
U_sparse, S_sparse, V_sparse = mat_4x3.svd(2)
rec = dot(U_sparse * S_sparse, V_sparse.T)
rec2 = ReconstructedMatrix.make_random(mat_4x3.row_labels,
mat_4x3.col_labels,
2,
learning_rate = 0.01)
for iter in xrange(1000):
for row in xrange(4):
for col in xrange(3):
rec2.hebbian_step(row, col, mat_4x3[row, col])
print np.linalg.norm(rec2.to_dense() - rec)
dense = rec2.to_dense()
assert rec.same_labels_as(rec2)
assert np.linalg.norm(rec2.to_dense() - rec) < 0.1
def __getitem__(self, indices):
if not isinstance(indices, tuple):
indices = (indices,)
if len(indices) < 2:
indices += (SLICE_ALL,) * (2 - len(indices))
leftpart = self.left[indices[0], :]
rightpart = self.right[:, indices[1]]
if (not np.isscalar(indices[0])) and (not np.isscalar(indices[1])):
# slice by slice. Reconstruct a new matrix to avoid huge
# computations.
return ReconstructedMatrix(leftpart, rightpart, self.shifts)
else:
# in any other case, just return the dense result
row_shift = self.row_shift[indices[0]]
col_shift = self.col_shift[indices[1]]
return dot(leftpart, rightpart) + row_shift + col_shift + self.total_shift
def __getitem__(self, indices):
if not isinstance(indices, tuple):
indices = (indices, )
if len(indices) < 2:
indices += (SLICE_ALL, ) * (2 - len(indices))
leftpart = self.left[indices[0], :]
rightpart = self.right[:, indices[1]]
if (not np.isscalar(indices[0])) and (not np.isscalar(indices[1])):
# slice by slice. Reconstruct a new matrix to avoid huge
# computations.
return ReconstructedMatrix(leftpart, rightpart, self.shifts)
else:
# in any other case, just return the dense result
row_shift = self.row_shift[indices[0]]
col_shift = self.col_shift[indices[1]]
return (dot(leftpart, rightpart) + row_shift + col_shift +
self.total_shift)
def right_category(self, vec):
return dot(self.left, aligned_matrix_multiply(self.right, vec))
def left_category(self, vec):
return dot(aligned_matrix_multiply(vec, self.left), self.right)
def matvec_transp(self, vec):
return dot(self.right.T, dot(self.left.T, vec))
def matvec(self, vec):
return dot(self.left, dot(self.right, vec))
def to_dense(self):
return dot(self.left, self.right)
def to_dense(self):
return dot(self.left, self.right)
def right_category(self, vec):
return dot(self.left, aligned_matrix_multiply(self.right, vec))
def left_category(self, vec):
return dot(aligned_matrix_multiply(vec, self.left), self.right)
def matvec_transp(self, vec):
return dot(self.right.T, dot(self.left.T, vec))
def matvec(self, vec):
return dot(self.left, dot(self.right, vec))