Test + comparaison initiale vs optimisée vs optimal

wdnmf_dual/methods/Optim_pb.py

@@ -86,7 +86,7 @@ def optim_w(w_init, h, data, sigma, dictionary, reg, ent, rho1, rho2, cost,
                  ent = ent, rho1 = rho1, rho2=rho2, cost = cost, method = method, numItermax= numItermax,
                  thr = thr, verbose = verbose)
     
-    print(init_pb)
+    print("Coût initial : " + str(float(init_pb)))
     
     n, r = w_init.shape
     _, m = h.shape

wdnmf_dual/scripts/running_functions.py

@@ -9,28 +9,9 @@ m = 4
 r = 2
 rho1 = 0.1
 
-aim = tn.ones(n, m) + tn.rand(n,m)
-aim = simplex_norm(aim)
-
-#dictionary = tn.tensor([[0.05, 0.05],[0.9, 0.05], [0.05, 0.9]], dtype=tn.float64)
-dictionary = tn.abs(tn.rand(n,2, dtype=tn.float64))
-dictionary = simplex_norm(dictionary)
-
-#w = tn.tensor([[0.9, 0.05],[0.05, 0.05], [0.05, 0.9]], dtype=tn.float64)
-w = tn.ones(n, r) + tn.rand(n,r)
-w = simplex_norm(w)
-
-h = tn.ones(r, m) + tn.rand(r,m)
-h = simplex_norm(h)
-
-data = dictionary@h
-
 sigma = [0,1]
-
 reg = 0.1
-
 ent = 0.1
-
 rho2 = 0.1
 
 cost = tn.zeros((n, n), dtype=tn.float64)
@@ -40,21 +21,43 @@ for i in range(n):
 
 matK = compute_matK(cost, ent)
 
-
-out = pb_dual(aim, w, rho1, h, data, sigma, dictionary, reg, ent, matK)
-
-out2 = grad1_pb_dual(aim, w, rho1, h, data, ent, matK)
-
-print(w)
-
-print(dictionary)
-
-
-out3 = optim_w(w, h, data, sigma, dictionary, reg, ent, rho1, rho2, cost)
-
-print(out3)
-
-out4 = pb(w = out3, h = h, sigma = sigma, data = data, dictionary = dictionary, reg = reg, 
-     ent = ent, rho1 = rho1, rho2=rho2, cost = cost)
-
-print(out4)
\ No newline at end of file
+for i in range(100):
+    aim = tn.ones(n, m) + tn.rand(n,m)
+    aim = simplex_norm(aim)
+    
+    #dictionary = tn.tensor([[0.05, 0.05],[0.9, 0.05], [0.05, 0.9]], dtype=tn.float64)
+    dictionary = tn.abs(tn.rand(n,2, dtype=tn.float64))
+    dictionary = simplex_norm(dictionary)
+    
+    #w = tn.tensor([[0.9, 0.05],[0.05, 0.05], [0.05, 0.9]], dtype=tn.float64)
+    w = tn.ones(n, r) + tn.rand(n,r)
+    w = simplex_norm(w)
+    
+    h = tn.ones(r, m) + tn.rand(r,m)
+    h = simplex_norm(h)
+    
+    data = dictionary@h
+    
+
+    
+    
+    out = pb_dual(aim, w, rho1, h, data, sigma, dictionary, reg, ent, matK)
+    
+    out2 = grad1_pb_dual(aim, w, rho1, h, data, ent, matK)
+    
+    #print(w)
+    
+    #print(dictionary)
+    
+    
+    out3 = optim_w(w, h, data, sigma, dictionary, reg, ent, rho1, rho2, cost)
+    
+    #print(out3)
+    
+    out4 = pb(w = out3, h = h, sigma = sigma, data = data, dictionary = dictionary, reg = reg, 
+         ent = ent, rho1 = rho1, rho2=rho2, cost = cost)
+    
+    print("Coût après optim : " + str(float(out4)))
+    print("Coût optimal : " + str(float(pb(w = dictionary, h = h, sigma = sigma, data = data, dictionary = dictionary, reg = reg, 
+         ent = ent, rho1 = rho1, rho2=rho2, cost = cost))))
+    print("~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~")
\ No newline at end of file