aufgabe5

uebungen/aufgabe3/uebung3.py

@@ -4,7 +4,7 @@ import matplotlib.pyplot as plt
 
 
 def load_data():
-    df_orig_train = pd.read_csv('mnist.csv')
+    df_orig_train = pd.read_csv('uebungen/aufgabe3/mnist.csv')
     df_digits = df_orig_train.drop('label', axis=1)
 
     return df_digits.to_numpy()
@@ -39,13 +39,10 @@ class RBM:
         self.hidden_bias = np.zeros(self.hidden_size) * 0.1
 
     def activate(self, v0):
-        return self.sample(sigmoid(np.matmul(v0.T, self.weights) + self.hidden_bias))
+        return sigmoid(np.matmul(v0.T, self.weights) + self.hidden_bias)
 
     def reactivate(self, h0):
-        return self.sample(sigmoid(np.matmul(self.weights, h0.T) + self.visible_bias))
-
-    def sample(self, a):
-        return np.where(a > np.random.uniform(0, 1, a.shape), 1, 0)
+        return sigmoid(np.matmul(self.weights, h0.T) + self.visible_bias)
 
     def contrastive_divergence(self, v0, h0, v1, h1):
         # Gradient

uebungen/aufgabe5/uebung5.py

@@ -0,0 +1,44 @@
+import numpy as np
+import matplotlib.pyplot as plt
+
+start = np.array([0, 0])
+bias = np.array([-3.37, 0.125])
+weights = np.array([
+    [-4,1.5],
+    [-1.5,0]
+])
+
+def activate(input):
+    # calculate activation as matrix
+    # o1 = w11 * o1 + w12 * o2 + b1
+    # o2 = w21 * o1 + w22 * o2 + b2
+    return np.matmul(weights, input) + bias
+        
+def predict(n):
+    current_output = start.copy()
+    points = np.zeros((2, n))
+
+    for i in range(n):
+        # calculate output with tanh(x)
+        current_output = np.tanh(activate(current_output))
+
+        #save datapoint
+        points[:, i] = current_output.copy()
+    
+    return points
+
+timespan = 50
+timespan_range = range(timespan)
+
+predictions = predict(timespan)
+
+# fetch o1, o2 from datapoint prediction
+o1 = predictions[0, :]
+o2 = predictions[1, :]
+
+plt.title("Recurrent Neural Network")
+plt.plot(timespan_range, o1)
+plt.plot(timespan_range, o2)
+plt.legend(["o1", "o2"], loc="upper left")
+plt.show()
+