def read_first_matrix(specs, delete_failures=False):
"""
Attempt to read each path provided by specs, until one exists. If none of
the specs point to an existing path, raise a FileNotFoundError.
specs should be a list where each element is a dictionary specifying a
potential path from which to read a matrix. Currently, the spec dictionary
supports the following keys:
- path: path to the file
- transpose: whether to transpose the file after reading it. If omitted,
then False.
- file_format: format of the matrix. If omitted, then infer.
"""
paths = list()
for spec in specs:
path = pathlib.Path(spec['path'])
paths.append(str(path))
if not path.is_file():
continue
transpose = spec.get('transpose', False)
file_format = spec.get('file_format', 'infer')
try:
matrix = read_matrix(path, file_format=file_format)
except Exception as error:
logging.warning(f'Error reading matrix at {path}:\n{error}')
if delete_failures:
path.unlink()
logging.warning(f'Deleting file at {path}')
continue
if transpose:
matrix = matrix.transpose()
return matrix
raise FileNotFoundError(
f'No matrix files found at the specified paths:\n' + '\n'.join(paths))
def normalize(matrix, vector, axis, damping_exponent):
"""
Normalize a 2D numpy.ndarray.
Parameters
==========
matrix : numpy.ndarray or scipy.sparse
vector : numpy.ndarray
Vector used for row or column normalization of matrix.
axis : str
'rows' or 'columns' for which axis to normalize
damping_exponent : float
exponent to use in scaling a node's row or column
"""
assert matrix.ndim == 2
assert vector.ndim == 1
if damping_exponent == 0:
return matrix
with numpy.errstate(divide='ignore'):
vector **= -damping_exponent
vector[numpy.isinf(vector)] = 0
vector = scipy.sparse.diags(vector)
if axis == 'rows':
# equivalent to `vector @ matrix` but returns scipy.sparse.csc not scipy.sparse.csr # noqa: E501
matrix = (matrix.transpose() @ vector).transpose()
else:
matrix = matrix @ vector
return matrix