Added activation functions (#968)

learning.py

@@ -4,7 +4,8 @@
     removeall, unique, product, mode, argmax, argmax_random_tie, isclose, gaussian,
     dotproduct, vector_add, scalar_vector_product, weighted_sample_with_replacement,
     weighted_sampler, num_or_str, normalize, clip, sigmoid, print_table,
-    open_data, sigmoid_derivative, probability, norm, matrix_multiplication, relu, relu_derivative
+    open_data, sigmoid_derivative, probability, norm, matrix_multiplication, relu, relu_derivative,
+    tanh, tanh_derivative, leaky_relu, leaky_relu_derivative, elu, elu_derivative
 )
 
 import copy
@@ -746,8 +747,15 @@ def BackPropagationLearner(dataset, net, learning_rate, epochs, activation=sigmo
             # The activation function used is relu or sigmoid function
             if node.activation == sigmoid:
                 delta[-1] = [sigmoid_derivative(o_nodes[i].value) * err[i] for i in range(o_units)]
-            else:
+            elif node.activation == relu:
                 delta[-1] = [relu_derivative(o_nodes[i].value) * err[i] for i in range(o_units)]
+            elif node.activation == tanh:
+                delta[-1] = [tanh_derivative(o_nodes[i].value) * err[i] for i in range(o_units)]
+            elif node.activation == elu:
+                delta[-1] = [elu_derivative(o_nodes[i].value) * err[i] for i in range(o_units)]
+            else:
+                delta[-1] = [leaky_relu_derivative(o_nodes[i].value) * err[i] for i in range(o_units)]
+
 
             # Backward pass
             h_layers = n_layers - 2
@@ -762,9 +770,18 @@ def BackPropagationLearner(dataset, net, learning_rate, epochs, activation=sigmo
                 if activation == sigmoid:
                     delta[i] = [sigmoid_derivative(layer[j].value) * dotproduct(w[j], delta[i+1])
                             for j in range(h_units)]
-                else:
+                elif activation == relu:
                     delta[i] = [relu_derivative(layer[j].value) * dotproduct(w[j], delta[i+1])
                             for j in range(h_units)]
+                elif activation == tanh:
+                    delta[i] = [tanh_derivative(layer[j].value) * dotproduct(w[j], delta[i+1])
+                            for j in range(h_units)]
+                elif activation == elu:
+                    delta[i] = [elu_derivative(layer[j].value) * dotproduct(w[j], delta[i+1])
+                            for j in range(h_units)]
+                else:
+                    delta[i] = [leaky_relu_derivative(layer[j].value) * dotproduct(w[j], delta[i+1])
+                            for j in range(h_units)]
 
             #  Update weights
             for i in range(1, n_layers):

utils.py

@@ -9,6 +9,7 @@
 import random
 import math
 import functools
+import numpy as np
 from itertools import chain, combinations
 
 
@@ -273,9 +274,47 @@ def sigmoid(x):
     """Return activation value of x with sigmoid function"""
     return 1 / (1 + math.exp(-x))
 
+
+
+def relu_derivative(value):
+	if value > 0:
+		return 1
+	else:
+		return 0
+
+def elu(x, alpha=0.01):
+	if x > 0:
+		return x
+	else:
+		return alpha * (math.exp(x) - 1)
+
+def elu_derivative(value, alpha = 0.01):
+	if value > 0:
+		return 1
+	else:
+		return alpha * math.exp(value)
+
+def tanh(x):
+	return np.tanh(x)
+
+def tanh_derivative(value):
+	return (1 - (value ** 2))
+
+def leaky_relu(x, alpha = 0.01):
+	if x > 0:
+		return x
+	else:
+		return alpha * x
+
+def leaky_relu_derivative(value, alpha=0.01):
+	if value > 0:
+		return 1
+	else:
+		return alpha
+
 def relu(x):
 	return max(0, x)
-
+	
 def relu_derivative(value):
 	if value > 0:
 		return 1