Merge pull request #137 from mrava87/doc-genproxgrad

doc: improved consistency in code and doc in GeneralizedProximalGradient

pyproximal/optimization/primal.py

@@ -308,7 +308,7 @@ def GeneralizedProximalGradient(proxfs, proxgs, x0, tau=None,
     .. math::
 
         \mathbf{x} = \argmin_\mathbf{x} \sum_{i=1}^n f_i(\mathbf{x}) 
-        + \sum_{j=1}^m \tau_j g_j(\mathbf{x}),~~n,m \in \mathbb{N}^+
+        + \sum_{j=1}^m \epsilon_j g_j(\mathbf{x}),~~n,m \in \mathbb{N}^+
 
     where the :math:`f_i(\mathbf{x})` are smooth convex functions with a uniquely
     defined gradient and the :math:`g_j(\mathbf{x})` are any convex function that
@@ -329,7 +329,7 @@ def GeneralizedProximalGradient(proxfs, proxgs, x0, tau=None,
         backtracking is used to adaptively estimate the best tau at each
         iteration.
     epsg : :obj:`float` or :obj:`np.ndarray`, optional
-        Scaling factor of g function
+        Scaling factor(s) of ``g`` function(s)
     niter : :obj:`int`, optional
         Number of iterations of iterative scheme
     acceleration:  :obj:`str`, optional
@@ -352,11 +352,12 @@ def GeneralizedProximalGradient(proxfs, proxgs, x0, tau=None,
 
     .. math::
         \text{for } j=1,\cdots,n, \\
-        ~~~~\mathbf z_j^{k+1} = \mathbf z_j^{k} + \eta_k (prox_{\frac{\tau^k}{\omega_j} g_j}(2 \mathbf{x}^{k} - z_j^{k}) 
-        - \tau^k \sum_{i=1}^n \nabla f_i(\mathbf{x}^{k})) - \mathbf{x}^{k} \\
+        ~~~~\mathbf z_j^{k+1} = \mathbf z_j^{k} + \epsilon_j
+        \left[prox_{\frac{\tau^k}{\omega_j} g_j}\left(2 \mathbf{x}^{k} - \mathbf{z}_j^{k}
+        - \tau^k \sum_{i=1}^n \nabla f_i(\mathbf{x}^{k})\right) - \mathbf{x}^{k} \right] \\
         \mathbf{x}^{k+1} = \sum_{j=1}^n \omega_j f_j \\
         
-    where :math:`\sum_{j=1}^n \omega_j=1`. 
+    where :math:`\sum_{j=1}^n \omega_j=1`. In the current implementation :math:`\omega_j=1/n`.
     """
     # check if epgs is a vector
     if np.asarray(epsg).size == 1.:
@@ -393,23 +394,23 @@ def GeneralizedProximalGradient(proxfs, proxgs, x0, tau=None,
     for iiter in range(niter):
         xold = x.copy()
 
-        # proximal step
+        # gradient
         grad = np.zeros_like(x)
         for i, proxf in enumerate(proxfs):
             grad += proxf.grad(x)
 
-        sol = np.zeros_like(x)
+        # proximal step
+        x = np.zeros_like(x)
         for i, proxg in enumerate(proxgs):
-            tmp = 2 * y - zs[i] - tau * grad
-            tmp[:] = proxg.prox(tmp, tau *len(proxgs) )
-            zs[i] += epsg * (tmp - y)
-            sol += zs[i] / len(proxgs)
-        x[:] = sol.copy()
+            ztmp = 2 * y - zs[i] - tau * grad
+            ztmp = proxg.prox(ztmp, tau * len(proxgs))
+            zs[i] += epsg * (ztmp - y)
+            x += zs[i] / len(proxgs)
 
         # update y
         if acceleration == 'vandenberghe':
             omega = iiter / (iiter + 3)
-        elif acceleration== 'fista':
+        elif acceleration == 'fista':
             told = t
             t = (1. + np.sqrt(1. + 4. * t ** 2)) / 2.
             omega = ((told - 1.) / t)
@@ -781,7 +782,7 @@ def ADMML2(proxg, Op, b, A, x0, tau, niter=10, callback=None, show=False, **kwar
 
         if show:
             if iiter < 10 or niter - iiter < 10 or iiter % (niter // 10) == 0:
-                pf, pg = np.linalg.norm(Op @ x - b), proxg(Ax)
+                pf, pg = 0.5 * np.linalg.norm(Op @ x - b) ** 2, proxg(Ax)
                 msg = '%6g  %12.5e  %10.3e  %10.3e  %10.3e' % \
                       (iiter + 1, x[0], pf, pg, pf + pg)
                 print(msg)

pyproximal/proximal/VStack.py

@@ -51,10 +51,16 @@ def __init__(self, ops, nn=None, restr=None):
             self.xin = cum_nn[:-1]
             self.xin = np.insert(self.xin, 0, 0)
             self.xend = cum_nn
+            # store required size of input
+            self.nx = cum_nn[-1]
         else:
             self.restr = restr
+            # store required size of input
+            self.nx = np.sum([restr.iava.size for restr in self.restr])
 
     def __call__(self, x):
+        if x.size != self.nx:
+            raise ValueError(f'x must have size {self.nx}, instead the provided x has size {x.size}')
         f = 0.
         if hasattr(self, 'nn'):
             for iop, op in enumerate(self.ops):
@@ -66,6 +72,8 @@ def __call__(self, x):
 
     @_check_tau
     def prox(self, x, tau):
+        if x.size != self.nx:
+            raise ValueError(f'x must have size {self.nx}, instead the provided x has size {x.size}')
         if hasattr(self, 'nn'):
             f = np.hstack([op.prox(x[self.xin[iop]:self.xend[iop]], tau)
                            for iop, op in enumerate(self.ops)])
@@ -76,6 +84,8 @@ def prox(self, x, tau):
         return f
 
     def grad(self, x):
+        if x.size != self.nx:
+            raise ValueError(f'x must have size {self.nx}, instead the provided x has size {x.size}')
         if hasattr(self, 'nn'):
             f = np.hstack([op.grad(x[self.xin[iop]:self.xend[iop]])
                            for iop, op in enumerate(self.ops)])

pytests/test_proximal.py

@@ -2,7 +2,7 @@
 
 import numpy as np
 from numpy.testing import assert_array_equal, assert_array_almost_equal
-from pylops import MatrixMult, Identity
+from pylops import Identity, MatrixMult, Restriction
 
 import pyproximal
 from pyproximal.utils import moreau
@@ -84,6 +84,20 @@ def test_Orthogonal(par):
     assert moreau(orth, x, tau)
 
 
+@pytest.mark.parametrize("par", [(par1), (par2)])
+def test_VStack_error(par):
+    """VStack operator error when input has wrong dimensions
+    """
+    np.random.seed(10)
+    nxs = [par['nx'] // 4] * 4
+    nxs[-1] = par['nx'] - np.sum(nxs[:-1])
+    l2 = L2()
+    vstack = VStack([l2] * 4, nxs)
+
+    with pytest.raises(ValueError):
+        vstack.prox(np.ones(nxs[0]), 2)
+
+
 @pytest.mark.parametrize("par", [(par1), (par2)])
 def test_VStack(par):
     """L2 functional with VStack operator of multiple L1s
@@ -109,6 +123,33 @@ def test_VStack(par):
     assert moreau(vstack, x, tau)
 
 
+@pytest.mark.parametrize("par", [(par1), ])
+def test_VStack_restriction(par):
+    """L2 functional with VStack operator of multiple L1s using restriction
+    """
+    np.random.seed(10)
+    nxs = [par['nx'] // 2] * 2
+    nxs[-1] = par['nx'] - np.sum(nxs[:-1])
+    l2 = L2()
+    vstack = VStack([l2] * 2,
+                    restr=[Restriction(par['nx'], np.arange(par['nx'] // 2)),
+                           Restriction(par['nx'], par['nx'] // 2 + np.arange(par['nx'] // 2))])
+
+    # functional
+    x = np.random.normal(0., 1., par['nx']).astype(par['dtype'])
+    assert_array_almost_equal(l2(x), vstack(x), decimal=4)
+
+    # gradient
+    assert_array_almost_equal(l2.grad(x), vstack.grad(x), decimal=4)
+
+    # prox / dualprox
+    tau = 2.
+    assert_array_equal(l2.prox(x, tau), vstack.prox(x, tau))
+
+    # moreau
+    assert moreau(vstack, x, tau)
+
+
 def test_Nonlinear():
     """Nonlinear proximal operator. Since this is a template class simply check
     that errors are raised when not used properly