def _do_sweep(self, covs, Q, prev_contrast):
"""
Perform a single sweep.
:param covs: The covariance matrices.
:param Q: he rotation matrix.
:param prev_contrast: The previous contrast.
:return: The maximum improvement in contrast, rotated matrices, the current
contrast.
"""
# perform a single sweep
# initialize maximal improvement in a single sweep
max_increase = -1
# shuffle rotation order
numx_rand.shuffle(self.rot_axis)
# sweep through all axes combinations
for (i, j) in self.rot_axis:
# get the angle that minimizes the contrast
# and the contrast value
angle, contrast = self._givens_angle(i, j, covs)
if contrast == 0:
# we hit numerical precision in case when b_sfa == 0
# we can only break things from here on, better quit!
max_increase = -1
break
# relative improvement in the contrast function
relative_diff = old_div((prev_contrast - contrast),
abs(prev_contrast))
if relative_diff < 0:
# if rate of change is negative we hit numerical precision
# or we already sit on the optimum for this pair of axis.
# don't rotate anymore and go to the next pair
continue
# update the rotation matrix
rotate(Q, angle, [i, j])
# rotate the covariance matrices
covs.rotate(angle, [i, j])
# store maximum and previous rate of change
max_increase = max(max_increase, relative_diff)
prev_contrast = contrast
return max_increase, covs, Q, contrast
def _do_sweep(self, covs, Q, prev_contrast):
"""
Perform a single sweep.
:param covs: The covariance matrices.
:param Q: he rotation matrix.
:param prev_contrast: The previous contrast.
:return: The maximum improvement in contrast, rotated matrices, the current
contrast.
"""
# perform a single sweep
# initialize maximal improvement in a single sweep
max_increase = -1
# shuffle rotation order
numx_rand.shuffle(self.rot_axis)
# sweep through all axes combinations
for (i, j) in self.rot_axis:
# get the angle that minimizes the contrast
# and the contrast value
angle, contrast = self._givens_angle(i, j, covs)
if contrast == 0:
# we hit numerical precision in case when b_sfa == 0
# we can only break things from here on, better quit!
max_increase = -1
break
# relative improvement in the contrast function
relative_diff = old_div((prev_contrast-contrast),abs(prev_contrast))
if relative_diff < 0:
# if rate of change is negative we hit numerical precision
# or we already sit on the optimum for this pair of axis.
# don't rotate anymore and go to the next pair
continue
# update the rotation matrix
rotate(Q, angle, [i, j])
# rotate the covariance matrices
covs.rotate(angle, [i, j])
# store maximum and previous rate of change
max_increase = max(max_increase, relative_diff)
prev_contrast = contrast
return max_increase, covs, Q, contrast
def _label_one(self, pattern, threshold):
pattern = mdp.utils.bool_to_sign(pattern)
has_converged = False
while not has_converged:
has_converged = True
iter_order = range(len(self._weight_matrix))
if self._shuffled_update:
numx_rand.shuffle(iter_order)
for row in iter_order:
w_row = self._weight_matrix[row]
thresh_row = threshold[row]
new_pattern_row = numx.sign(numx.dot(w_row, pattern) - thresh_row)
if new_pattern_row == 0:
# Following McKay, Neural Networks, we do nothing
# when the new pattern is zero
pass
elif pattern[row] != new_pattern_row:
has_converged = False
pattern[row] = new_pattern_row
return mdp.utils.sign_to_bool(pattern)
def _label_one(self, pattern, threshold):
pattern = mdp.utils.bool_to_sign(pattern)
has_converged = False
while not has_converged:
has_converged = True
iter_order = range(len(self._weight_matrix))
if self._shuffled_update:
numx_rand.shuffle(iter_order)
for row in iter_order:
w_row = self._weight_matrix[row]
thresh_row = threshold[row]
new_pattern_row = numx.sign(
numx.dot(w_row, pattern) - thresh_row)
if new_pattern_row == 0:
# Following McKay, Neural Networks, we do nothing
# when the new pattern is zero
pass
elif pattern[row] != new_pattern_row:
has_converged = False
pattern[row] = new_pattern_row
return mdp.utils.sign_to_bool(pattern)
