Skip to content

Added ability to add quad term when solving admm; changes to affine t… #10

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
kevinbfry wants to merge 1 commit into
base: master
Choose a base branch
from
Open

Added ability to add quad term when solving admm; changes to affine t… #10

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
10 changes: 8 additions & 2 deletions regreg/affine/__init__.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -854,8 +854,10 @@ class aslinear(linear_transform):
def __init__(self, transform):
self._transform = astransform(transform)
self.affine_offset = None
self.input_shape = self._transform.output_shape
self.output_shape = self._transform.input_shape
# self.input_shape = self._transform.output_shape
# self.output_shape = self._transform.input_shape
self.input_shape = self._transform.input_shape
self.output_shape = self._transform.output_shape

def linear_map(self, x):
return self._transform.linear_map(x)
Expand All @@ -877,6 +879,8 @@ def __init__(self, transform):
self.affine_offset = None
self.input_shape = self.transform.output_shape
self.output_shape = self.transform.input_shape
# self.input_shape = self.transform.input_shape
# self.output_shape = self.transform.output_shape

def linear_map(self, x):
return self.transform.adjoint_map(x)
Expand Down Expand Up @@ -948,6 +952,8 @@ def __init__(self, *transforms):
self.transforms = [astransform(t) for t in transforms]
self.input_shape = self.transforms[-1].input_shape
self.output_shape = self.transforms[0].output_shape

# if self.transforms[-1]

# compute the affine_offset
affine_offset = self.affine_map(np.zeros(self.input_shape))
Expand Down
20 changes: 15 additions & 5 deletions regreg/problems/admm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,6 +33,9 @@ def __init__(self,
quadratic=None,
fit_args={}):

# if quadratic is None:
# quadratic = identity_quadratic(0,0,0,0)

(self.loss,
self.atom,
self.transform,
Expand All @@ -41,22 +44,22 @@ def __init__(self,
atom,
astransform(transform),
augmented_param,
quadratic)
identity_quadratic(0,0,0,0) if quadratic is None else quadratic)

self.loss_coefs = self.loss.coefs # x in ADMM notes
self.dual_coefs = np.zeros(self.transform.output_shape) # y in ADMM notes
self.atom_coefs = np.zeros(self.transform.output_shape) # z in ADMM notes
self.augmented_param = augmented_param # rho in ADMM notes

self.linear_transform = aslinear(self.transform) # D
self.linear_transform = aslinear(self.transform) # D; using transpose b/c for some reason returns transform.T
qloss = quadratic_loss.squared_transform(self.linear_transform) # x^TD^TDx / 2
qloss.coef *= self.augmented_param # scale by rho
self.augmented_loss = smooth_sum([self.loss,
qloss])

self.fit_args = fit_args # for the smooth_sum FISTA run

assert (self.loss.shape == self.transform.output_shape)
# assert (self.loss.shape == self.transform.output_shape)

def update_loss_coefs(self):
"""
Expand All @@ -68,7 +71,8 @@ def update_loss_coefs(self):
self.transform.affine_offset)
if alpha is None:
alpha = 0
linear_term = self.linear_transform.dot(y - rho * (z - alpha))
# linear_term = self.linear_transform.dot(y - rho * (z - alpha))
linear_term = self.linear_transform.T.dot(y - rho * (z - alpha))
self.loss_coefs[:] = self.augmented_loss.solve(quadratic=identity_quadratic(0, 0, linear_term, 0),
**self.fit_args)

Expand All @@ -89,9 +93,15 @@ def update_dual_coefs(self):
rho = self.augmented_param
self.dual_coefs[:] += rho * (self.transform.affine_map(self.loss_coefs) - self.atom_coefs)

def solve(self, niter=20):
def solve(self, quadratic=None, niter=20):
if quadratic is None:
quadratic = identity_quadratic(0, 0, 0, 0)
oldq, self.quadratic = self.quadratic, self.quadratic + quadratic

for _ in range(niter):
self.update_loss_coefs()
self.update_atom_coefs()
self.update_dual_coefs()

self.quadratic = oldq