From 2c0744b55d69e3755d1d99ea164b5eaedb382b6d Mon Sep 17 00:00:00 2001
From: epapoutsellis <epapoutsellis@gmail.com>
Date: Thu, 23 May 2019 17:39:28 +0100
Subject: delete cvx sol

---
 .../demos/PDHG_examples/PDHG_TGV_Denoising.py      | 74 +---------------------
 1 file changed, 2 insertions(+), 72 deletions(-)

(limited to 'Wrappers')

diff --git a/Wrappers/Python/demos/PDHG_examples/PDHG_TGV_Denoising.py b/Wrappers/Python/demos/PDHG_examples/PDHG_TGV_Denoising.py
index 761c025..4d6da00 100755
--- a/Wrappers/Python/demos/PDHG_examples/PDHG_TGV_Denoising.py
+++ b/Wrappers/Python/demos/PDHG_examples/PDHG_TGV_Denoising.py
@@ -30,8 +30,8 @@ Problem:     min_{x} \alpha * ||\nabla x - w||_{2,1} +
             where fidelity can be as follows depending on the noise characteristics
             of the data:
                 * Norm2Squared     \frac{1}{2} * || x - g ||_{2}^{2}
-                * KullbackLeibler 
-                * L1Norm
+                * KullbackLeibler  \int u - g * log(u) + Id_{u>0}
+                * L1Norm           ||u - g||_{1}
 
              \alpha: Regularization parameter
              \beta:  Regularization parameter
@@ -219,73 +219,3 @@ plt.title('Middle Line Profiles')
 plt.show()
 
 
-#%% Check with CVX solution
-
-from ccpi.optimisation.operators import SparseFiniteDiff
-
-try:
-    from cvxpy import *
-    cvx_not_installable = True
-except ImportError:
-    cvx_not_installable = False
-
-if cvx_not_installable:    
-    
-    u = Variable(ig.shape)
-    w1 = Variable((N, N))
-    w2 = Variable((N, N))
-    
-    # create TGV regulariser
-    DY = SparseFiniteDiff(ig, direction=0, bnd_cond='Neumann')
-    DX = SparseFiniteDiff(ig, direction=1, bnd_cond='Neumann')
-    
-    regulariser = alpha * sum(norm(vstack([DX.matrix() * vec(u) - vec(w1), \
-                                           DY.matrix() * vec(u) - vec(w2)]), 2, axis = 0)) + \
-                  beta * sum(norm(vstack([ DX.matrix().transpose() * vec(w1), DY.matrix().transpose() * vec(w2), \
-                                      0.5 * ( DX.matrix().transpose() * vec(w2) + DY.matrix().transpose() * vec(w1) ), \
-                                      0.5 * ( DX.matrix().transpose() * vec(w2) + DY.matrix().transpose() * vec(w1) ) ]), 2, axis = 0  ) )  
-    
-    constraints = []
-    fidelity = pnorm(u - noisy_data.as_array(),1)    
-    solver = MOSEK
-
-        # choose solver
-    if 'MOSEK' in installed_solvers():
-        solver = MOSEK
-    else:
-        solver = SCS  
-        
-    obj =  Minimize( regulariser +  fidelity)
-    prob = Problem(obj)
-    result = prob.solve(verbose = True, solver = solver)
-    
-    diff_cvx = numpy.abs( pdhg.get_output()[0].as_array() - u.value )
-        
-    plt.figure(figsize=(15,15))
-    plt.subplot(3,1,1)
-    plt.imshow(pdhg.get_output()[0].as_array())
-    plt.title('PDHG solution')
-    plt.colorbar()
-    plt.subplot(3,1,2)
-    plt.imshow(u.value)
-    plt.title('CVX solution')
-    plt.colorbar()
-    plt.subplot(3,1,3)
-    plt.imshow(diff_cvx)
-    plt.title('Difference')
-    plt.colorbar()
-    plt.show()    
-    
-    plt.plot(np.linspace(0,N,N), pdhg.get_output()[0].as_array()[int(N/2),:], label = 'PDHG')
-    plt.plot(np.linspace(0,N,N), u.value[int(N/2),:], label = 'CVX')
-    plt.legend()
-    plt.title('Middle Line Profiles')
-    plt.show()
-            
-    print('Primal Objective (CVX) {} '.format(obj.value))
-    print('Primal Objective (PDHG) {} '.format(pdhg.objective[-1][0]))
-
-
-
-
-
-- 
cgit v1.2.3