fitDistance.py: print and save results, update codes of quadratic program. This is the last version with edit costs having no other constraints than being no smaller than 0.

preimage/fitDistance.py

@@ -19,7 +19,6 @@ import cvxpy as cp
 
 import sys
 sys.path.insert(0, "../")
-from pygraph.utils.graphfiles import loadDataset
 from ged import GED, get_nb_edit_operations
 from utils import kernel_distance_matrix
 
@@ -43,9 +42,11 @@ def fit_GED_to_kernel_distance(Gn, gkernel, itr_max):
     
     residual_list = []
     edit_cost_list = []
+    time_list = []
     
     for itr in range(itr_max):
         print('\niteration', itr)
+        time0 = time.time()
         # compute GEDs and numbers of edit operations.
         edit_cost_constant = [i for i in edit_costs]
         edit_cost_list.append(edit_cost_constant)
@@ -71,11 +72,20 @@ def fit_GED_to_kernel_distance(Gn, gkernel, itr_max):
         
         for idx, item in enumerate(idx_nonzeros):
             edit_costs[item] = edit_costs_new[idx]
+        
+        time_list.append(time.time() - time0)
             
         print('edit_costs:', edit_costs)
         print('residual_list:', residual_list)
+        
+        
+    edit_cost_list.append(edit_costs)
+    ged_all, ged_mat, n_edit_operations = compute_geds(Gn, edit_costs, 
+            idx_nonzeros, parallel=True)
+    residual = np.sqrt(np.sum(np.square(np.array(ged_all) - dis_k_vec)))
+    residual_list.append(residual)
     
-    return edit_costs, residual_list, edit_cost_list, dis_k_mat, ged_mat
+    return edit_costs, residual_list, edit_cost_list, dis_k_mat, ged_mat, time_list
 
 
 def compute_geds(Gn, edit_cost_constant, idx_nonzeros, parallel=False):
@@ -166,33 +176,33 @@ def compute_better_costs(nb_cost_mat, dis_k_vec):
 #    edit_costs_new, residual = optimize.nnls(nb_cost_mat, dis_k_vec)
     
     # method 3: solve as a quadratic program with constraints: x_i >= 0, sum(x) = 1.
-    P = np.dot(nb_cost_mat.T, nb_cost_mat)
-    q_T = -2 * np.dot(dis_k_vec.T, nb_cost_mat)
-    G = -1 * np.identity(nb_cost_mat.shape[1])
-    h = np.array([0 for i in range(nb_cost_mat.shape[1])])
-    A = np.array([1 for i in range(nb_cost_mat.shape[1])])
-    b = 1
+#    P = np.dot(nb_cost_mat.T, nb_cost_mat)
+#    q_T = -2 * np.dot(dis_k_vec.T, nb_cost_mat)
+#    G = -1 * np.identity(nb_cost_mat.shape[1])
+#    h = np.array([0 for i in range(nb_cost_mat.shape[1])])
+#    A = np.array([1 for i in range(nb_cost_mat.shape[1])])
+#    b = 1
+#    x = cp.Variable(nb_cost_mat.shape[1])
+#    prob = cp.Problem(cp.Minimize(cp.quad_form(x, P) + q_T@x),
+#                      [G@x <= h])
+#    prob.solve()
+#    edit_costs_new = x.value
+#    residual = prob.value - np.dot(dis_k_vec.T, dis_k_vec)
+    
+#    G = -1 * np.identity(nb_cost_mat.shape[1])
+#    h = np.array([0 for i in range(nb_cost_mat.shape[1])])
     x = cp.Variable(nb_cost_mat.shape[1])
-    prob = cp.Problem(cp.Minimize(cp.quad_form(x, P) + q_T@x),
-                      [G@x <= h])
+    cost = cp.sum_squares(nb_cost_mat * x - dis_k_vec)
+    constraints = [x >= [0 for i in range(nb_cost_mat.shape[1])]]
+    prob = cp.Problem(cp.Minimize(cost), constraints)
     prob.solve()
     edit_costs_new = x.value
-    residual = prob.value - np.dot(dis_k_vec.T, dis_k_vec)
+    residual = np.sqrt(prob.value)
+    
+    # method 4: 
     
     return edit_costs_new, residual
 
 
 if __name__ == '__main__':
-    from utils import remove_edges
-    ds = {'name': 'MUTAG', 'dataset': '../datasets/MUTAG/MUTAG_A.txt',
-          'extra_params': {}}  # node/edge symb
-    Gn, y_all = loadDataset(ds['dataset'], extra_params=ds['extra_params'])
-#    Gn = Gn[0:10]
-    remove_edges(Gn)
-    gkernel = 'marginalizedkernel'
-    itr_max = 10
-    time0 = time.time()
-    edit_costs, residual_list, edit_cost_list, dis_k_mat, ged_mat = \
-        fit_GED_to_kernel_distance(Gn, gkernel, itr_max)
-    total_time = time.time() - time0
-    print('total time:', total_time)
+    print('check test_fitDistance.py')

preimage/test_fitDistance.py

@@ -0,0 +1,114 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""
+Created on Thu Oct 24 11:50:56 2019
+
+@author: ljia
+"""
+from matplotlib import pyplot as plt
+import numpy as np
+
+from pygraph.utils.graphfiles import loadDataset
+from utils import remove_edges
+from fitDistance import fit_GED_to_kernel_distance
+from utils import normalize_distance_matrix
+
+def test_anycosts():
+    ds = {'name': 'MUTAG', 'dataset': '../datasets/MUTAG/MUTAG_A.txt',
+          'extra_params': {}}  # node/edge symb
+    Gn, y_all = loadDataset(ds['dataset'], extra_params=ds['extra_params'])
+#    Gn = Gn[0:10]
+    remove_edges(Gn)
+    gkernel = 'marginalizedkernel'
+    itr_max = 10
+    edit_costs, residual_list, edit_cost_list, dis_k_mat, ged_mat, time_list = \
+        fit_GED_to_kernel_distance(Gn, gkernel, itr_max)
+    total_time = np.sum(time_list)
+    print('\nedit_costs:', edit_costs)
+    print('\nresidual_list:', residual_list)
+    print('\nedit_cost_list:', edit_cost_list)
+    print('\ndistance matrix in kernel space:', dis_k_mat)
+    print('\nged matrix:', ged_mat)
+    print('total time:', total_time)
+    np.savez('results/fit_distance.any_costs.gm', edit_costs=edit_costs, 
+             residual_list=residual_list, edit_cost_list=edit_cost_list,
+             dis_k_mat=dis_k_mat, ged_mat=ged_mat, time_list=time_list, 
+             total_time=total_time)
+    
+    # normalized distance matrices.
+#    gmfile = np.load('results/fit_distance.any_costs.gm.npz')
+#    edit_costs = gmfile['edit_costs']
+#    residual_list = gmfile['residual_list']
+#    edit_cost_list = gmfile['edit_cost_list']
+#    dis_k_mat = gmfile['dis_k_mat']
+#    ged_mat = gmfile['ged_mat']
+#    total_time = gmfile['total_time']
+    
+    norm_dis_k_mat = normalize_distance_matrix(dis_k_mat)
+    plt.imshow(norm_dis_k_mat)
+    plt.colorbar()
+    plt.savefig('results/norm_dis_k_mat.any_costs' + '.eps', format='eps', dpi=300)
+#    plt.savefig('results/norm_dis_k_mat.any_costs' + '.jpg', format='jpg')
+#    plt.show()
+    plt.clf()
+    norm_ged_mat = normalize_distance_matrix(ged_mat)
+    plt.imshow(norm_ged_mat)
+    plt.colorbar()
+    plt.savefig('results/norm_ged_mat.any_costs' + '.eps', format='eps', dpi=300)
+#    plt.savefig('results/norm_ged_mat.any_costs' + '.jpg', format='jpg')
+#    plt.show()
+    plt.clf()
+    
+
+def test_cs_leq_ci_plus_cr():
+    """c_vs <= c_vi + c_vr, c_es <= c_ei + c_er
+    """
+    ds = {'name': 'MUTAG', 'dataset': '../datasets/MUTAG/MUTAG_A.txt',
+          'extra_params': {}}  # node/edge symb
+    Gn, y_all = loadDataset(ds['dataset'], extra_params=ds['extra_params'])
+#    Gn = Gn[0:10]
+    remove_edges(Gn)
+    gkernel = 'marginalizedkernel'
+    itr_max = 10
+    edit_costs, residual_list, edit_cost_list, dis_k_mat, ged_mat, time_list = \
+        fit_GED_to_kernel_distance(Gn, gkernel, itr_max)
+    total_time = np.sum(time_list)
+    print('\nedit_costs:', edit_costs)
+    print('\nresidual_list:', residual_list)
+    print('\nedit_cost_list:', edit_cost_list)
+    print('\ndistance matrix in kernel space:', dis_k_mat)
+    print('\nged matrix:', ged_mat)
+    print('total time:', total_time)
+    np.savez('results/fit_distance.cs_leq_ci_plus_cr.gm', edit_costs=edit_costs, 
+             residual_list=residual_list, edit_cost_list=edit_cost_list,
+             dis_k_mat=dis_k_mat, ged_mat=ged_mat, time_list=time_list, 
+             total_time=total_time)
+    
+    # normalized distance matrices.
+#    gmfile = np.load('results/fit_distance.cs_leq_ci_plus_cr.gm.npz')
+#    edit_costs = gmfile['edit_costs']
+#    residual_list = gmfile['residual_list']
+#    edit_cost_list = gmfile['edit_cost_list']
+#    dis_k_mat = gmfile['dis_k_mat']
+#    ged_mat = gmfile['ged_mat']
+#    total_time = gmfile['total_time']
+    
+    norm_dis_k_mat = normalize_distance_matrix(dis_k_mat)
+    plt.imshow(norm_dis_k_mat)
+    plt.colorbar()
+    plt.savefig('results/norm_dis_k_mat.cs_leq_ci_plus_cr' + '.eps', format='eps', dpi=300)
+#    plt.savefig('results/norm_dis_k_mat.cs_leq_ci_plus_cr' + '.jpg', format='jpg')
+#    plt.show()
+    plt.clf()
+    norm_ged_mat = normalize_distance_matrix(ged_mat)
+    plt.imshow(norm_ged_mat)
+    plt.colorbar()
+    plt.savefig('results/norm_ged_mat.cs_leq_ci_plus_cr' + '.eps', format='eps', dpi=300)
+#    plt.savefig('results/norm_ged_mat.cs_leq_ci_plus_cr' + '.jpg', format='jpg')
+#    plt.show()
+    plt.clf()
+    
+    
+if __name__ == '__main__':
+    test_anycosts()
+    test_cs_leq_ci_plus_cr()

preimage/utils.py

@@ -106,4 +106,10 @@ def get_same_item_indices(ls):
             idx_dict[item].append(idx)
         else:
             idx_dict[item] = [idx]
-    return idx_dict
+    return idx_dict
+
+
+def normalize_distance_matrix(D):
+    max_value = np.amax(D)
+    min_value = np.amin(D)
+    return (D - min_value) / (max_value - min_value)