Create deeplearning_utils.py

Author: Owen-CH-Leung

utils/deeplearning_utils.py

@@ -0,0 +1,74 @@
+import numpy as np 
+
+class ActivationFunction:
+    @staticmethod
+    def sigmoid(W, X):
+        '''
+        Parameters
+        ----------
+        W : numpy array of size (D, M)
+        X : numpy array of size (D, N)
+            
+        Returns
+        -------
+        numpy array of size (M, N)
+        '''
+        return 1 / (1 + np.exp(-1 * (W.T.dot(X))))
+    
+    @staticmethod
+    def tanh(W, X):
+        '''
+        Parameters
+        ----------
+        W : numpy array of size (D, M)
+        X : numpy array of size (D, N)
+            
+        Returns
+        -------
+        numpy array of size (M, N)
+        '''
+        x = W.T.dot(X)
+        return (np.exp(x) - np.exp(-x)) / (np.exp(x) + np.exp(-x))
+    
+    @staticmethod
+    def relu(W, X):
+        '''
+        Parameters
+        ----------
+        W : numpy array of size (D, M)
+        X : numpy array of size (D, N)
+            
+        Returns
+        -------
+        numpy array of size (M, N)
+        '''
+        x = W.T.dot(X)
+        return np.maximum(x, 0)
+    
+    def softmax(W, X):
+        '''
+        Parameters
+        ----------
+        W : numpy array of size (M, K)
+        X : numpy array of size (M, N)
+            
+        Returns
+        -------
+        numpy array of size (M, N)
+        '''
+        x = W.T.dot(X)
+        exp_X = np.exp(x)
+        return exp_X / exp_X.sum(axis=1, keepdims=True)
+    
+
+    
+if __name__ == '__main__':
+    N = 10
+    D = 2
+    M = 4
+    K = 3
+    X = np.random.random((D, N)) - 0.5
+    W1 = np.random.random((D, M)) 
+    W2 = np.random.random((M, K))
+    z1 = ActivationFunction.sigmoid(W1, X)
+    pred_y = ActivationFunction.softmax(W2, z1)