übung 3

Übung 3/SLN.py

@@ -37,23 +37,22 @@ class SLN:
         for it in range(maxIter):
             Y = self.neuron(X)  # nur ein Aufruf
             err = error_rate(Y, T)
-            if (it % 10) == 0:
+            # if (it % 10) == 0:
-                print('#{} err:{}\n{}\n{}'.format(it, err, self._W, self._b))
+            print('#{} err:{}\n{}\n{}'.format(it, err, self._W, self._b))
-                nnwplot.plotTwoFeatures(X, T, self.neuron)
+            nnwplot.plotTwoFeatures(X, T, self.neuron)
-                plt.pause(0.05)  # warte auf GUI event loop (je nach python Version notwendig oder nicht)
+            plt.pause(0.05)  # warte auf GUI event loop (je nach python Version notwendig oder nicht)
             if err <= maxErrorRate:
                 break
-            # update _W
             for n in range(N):
-                for j in range(self.dIn):
+                delta = T[n] - Y[n]  # Fehlersignal pro Beispiel (mit Vorzeichen!)
-                    weighted_error = err * X[j, n]
+
-                    to_learn = weighted_error * eta
+                # update _W
-                    #print("{:.2f} = {} * {} * {}".format(to_learn, err, X[j, n], eta))
+                for j in range(self.dIn):
-                    self._W[j, 0] += to_learn
+                    self._W[j, 0] += eta * delta * X[j, n]
-            # update _b
+
-            self._b[0][0] += err * eta
+                # update _b
+                self._b[0][0] += eta * delta
 
-            ...
         print('#{} err:{}\n{}\n{}'.format(it, err, self._W, self._b))
         nnwplot.plotTwoFeatures(X, T, self.neuron)
         plt.pause(0.05)  # warte auf GUI event loop (je nach python Version notwendig oder nicht)

Übung 3/nnwplot.py

@@ -4,26 +4,25 @@ from matplotlib import pyplot as plt
 from matplotlib import colors
 
 
-def plotTwoFeatures(X,T,pred_func):
+def plotTwoFeatures(X, T, pred_func):
-    if X.ndim!=2:
+    if X.ndim != 2:
         raise ValueError('X be a matrix (2 dimensional array).')
-    tp=False
+    tp = False
-    if X.shape[0]!=2 and X.shape[1]==2: 
+    if X.shape[0] != 2 and X.shape[1] == 2:
-        X=X.T
+        X = X.T
-        tp=True
+        tp = True
-    if X.shape[0]!=2:
+    if X.shape[0] != 2:
         raise ValueError('X must contain exactly 2 features.')
-        
+
     # determine canvas borders
-    mins = np.amin(X,1); 
+    mins = np.amin(X, 1)
-    mins = mins - 0.1*np.abs(mins);
+    mins = mins - np.where(mins == 0, 0.1, 0.1 * np.abs(mins))
-    maxs = np.amax(X,1); 
+    maxs = np.amax(X, 1)
-    maxs = maxs + 0.1*maxs;
+    maxs = maxs + 0.1 * maxs
 
     ## generate dense grid
-    xs,ys = np.meshgrid(np.linspace(mins[0],maxs[0],300), 
+    xs, ys = np.meshgrid(np.linspace(mins[0], maxs[0], 300),
-            np.linspace(mins[1], maxs[1], 300));
+                         np.linspace(mins[1], maxs[1], 300))
-
 
     # evaluate model on the dense grid
     try:
@@ -32,18 +31,18 @@ def plotTwoFeatures(X,T,pred_func):
         Z = pred_func(np.c_[xs.flatten(), ys.flatten()]);
 
     if tp:
-        Z=Z.T
+        Z = Z.T
-        T=T.T
+        T = T.T
-    
+
-    if Z.ndim>1 and Z.shape[0]>1: # onehot? -> convert
+    if Z.ndim > 1 and Z.shape[0] > 1:  # onehot? -> convert
-        Z=Z.argmax(0)
+        Z = Z.argmax(0)
     Z = Z.reshape(xs.shape)
-    
+
-    if T.ndim>1 and T.shape[0]>1: # onehot? -> convert
+    if T.ndim > 1 and T.shape[0] > 1:  # onehot? -> convert
-        T=T.argmax(0)
+        T = T.argmax(0)
-        
+
     # Plot the contour and training examples
     plt.contourf(xs, ys, Z, cmap=plt.cm.Spectral)
-    plt.scatter(X[0,:], X[1,:], c=T, s=50,
+    plt.scatter(X[0, :], X[1, :], c=T, s=50,
-            cmap=colors.ListedColormap(['orange', 'blue', 'green']))
+                cmap=colors.ListedColormap(['blue', 'green', 'red']))
     plt.show()