update

HW2/Untitled-1.py

@@ -0,0 +1,5 @@
+
+import numpy as np
+a = np.array([[0.00324953]])
+print(a)
+print(-a)

HW2/WolfePowellSearch.py

@@ -54,19 +54,21 @@
 # should return t=0.0938
 
 
+from re import T
 import numpy as np
 
 
 def matrnr():
     # set your matriculation number here
-    matrnr = 0
+    matrnr = 23062789
     return matrnr
 
 
 def WolfePowellSearch(f, x: np.array, d: np.array, sigma=1.0e-3, rho=1.0e-2, verbose=0):
     fx = f.objective(x)
     gradx = f.gradient(x)
     descent = gradx.T @ d
+    print("wp",descent)
 
     if descent >= 0:
         raise TypeError('descent direction check failed!')
@@ -82,6 +84,72 @@ def WolfePowellSearch(f, x: np.array, d: np.array, sigma=1.0e-3, rho=1.0e-2, ver
 
     t = 1
 
-    MISSING CODE
-
+    # MISSING CODE
+
+    fxd = f.objective(x + t*d)
+    gradxd = f.gradient(x + t*d)
+    descentxd = gradxd.T @ d
+
+    W1 = fxd <= (fx + t*sigma*descent)
+    W2 = descentxd >= (rho*descent)
+    if W1 == False:
+        t = t/2
+        fxd = f.objective(x + t*d)
+        W1 = fxd <= (fx + t*sigma*descent)
+        gradxd = f.gradient(x + t*d)
+        descentxd = gradxd.T @ d
+        W2 = descentxd >= (rho*descent)
+        while W1 == False:
+            t = t/2
+            fxd = f.objective(x + t*d)
+            W1 = fxd <= (fx + t*sigma*descent)
+            gradxd = f.gradient(x + t*d)
+            descentxd = gradxd.T @ d
+            W2 = descentxd >= (rho*descent)
+        t_min = t
+        t_pl = 2*t
+
+    elif W2 == True:
+        t_star = t
+        return t_star
+
+    else:
+        t = 2*t
+        fxd = f.objective(x + t*d)
+        W1 = fxd <= (fx + t*sigma*descent)
+        gradxd = f.gradient(x + t*d)
+        descentxd = gradxd.T @ d
+        W2 = descentxd >= (rho*descent)
+        while W1 == True:
+            t = 2*t
+            fxd = f.objective(x + t*d)
+            W1 = fxd <= (fx + t*sigma*descent)
+            gradxd = f.gradient(x + t*d)
+            descentxd = gradxd.T @ d
+            W2 = descentxd >= (rho*descent)
+        t_min = t/2
+        t_pl = t
+
+    t = t_min
+    fxd = f.objective(x + t*d)
+    W1 = fxd <= (fx + t*sigma*descent)
+    gradxd = f.gradient(x + t*d)
+    descentxd = gradxd.T @ d
+    W2 = descentxd >= (rho*descent)
+    while W2 == False:
+        print("4")
+        t = (t_min + t_pl)/2
+        fxd = f.objective(x + t*d)
+        W1 = fxd <= (fx + t*sigma*descent)
+        gradxd = f.gradient(x + t*d)
+        descentxd = gradxd.T @ d
+        W2 = descentxd >= (rho*descent)
+        if W1 == True:
+            t_min = t
+        else:
+            t_pl = t
+        print(t_min)
+    t_star = t_min
+
+    t = np.copy(t_star)
     return t

HW2/globalNewtonDescent.py

@@ -49,7 +49,7 @@
 
 def matrnr():
     # set your matriculation number here
-    matrnr = 0
+    matrnr = 23062789
     return matrnr
 
 
@@ -62,14 +62,30 @@ def globalNewtonDescent(f, x0: np.array, eps=1.0e-3, verbose=0):
         print('Start globalNewtonDescent...')
 
     countIter = 0
-    x = MISSING
+    xk = x0
 
-
-    while MISSING STATEMENT:
-        MISSING CODE
+    gradx = f.gradient(xk)
+    hessx = f.hessian(xk)
+
+    while np.linalg.norm(gradx) > eps:
+        print("tur")
+        # MISSING CODE
+        Bk = hessx
+        dk = PCG.PrecCGSolver(Bk,xk)
+        descent = gradx.T @ dk
+        print(dk)
+        if descent >= 0:
+            print("girdi")
+            dk = -gradx
+        print(descent)
+        tk = WP.WolfePowellSearch(f,xk,dk)
+        xk = xk + tk*dk
+        hessx = f.hessian(xk)
+        Bk = hessx
 
         countIter = countIter + 1
 
+    x = np.copy(xk)
 
     if verbose:
         gradx = f.gradient(x)