Adding 2-hidden layer neural network with back propagation using sigmoid activation function

neural_network/2_hidden_layers_neural_network.py

@@ -0,0 +1,167 @@
+import numpy
+
+
+class NeuralNetwork:
+    def __init__(self, input_array: numpy.array, output_array: numpy.array):
+        """
+        input_array : input values for training the neural network.
+        output_array : expected output values of the given inputs.
+        """
+        self.input = input_array
+        # Initial weights are assigned randomly where first argument is the number
+        # of nodes in previous layer and second argument is the
+        # number of nodes in the next layer.
+
+        # random initial weights for the input layer
+        # self.input.shape[1] is used to represent number of nodes in input layer
+        # first hidden layer consists of 4 nodes
+        self.weights1 = numpy.random.rand(self.input.shape[1], 4)
+
+        # random initial weights for the first hidden layer
+        # first hidden layer has 4 nodes
+        # second hidden layer has 3 nodes
+        self.weights2 = numpy.random.rand(4, 3)
+
+        # random initial weights for the second hidden layer
+        # second hidden layer has 3 nodes
+        # output layer has 1 node
+        self.weights3 = numpy.random.rand(3, 1)
+
+        self.y = output_array
+        self.output = numpy.zeros(output_array.shape)
+
+    def feedforward(self):
+        """
+        feedforward propagation using sigmoid activation function between layers
+        return the last layer of the neural network
+        """
+        # layer1 is the layer connecting the input nodes with
+        # the first hidden layer nodes
+        self.layer1 = sigmoid(numpy.dot(self.input, self.weights1))
+
+        # layer2 is the layer connecting the first hidden set of nodes
+        # with the second hidden set of nodes
+        self.layer2 = sigmoid(numpy.dot(self.layer1, self.weights2))
+
+        # layer3 is the layer connecting second hidden layer with the output node
+        self.layer3 = sigmoid(numpy.dot(self.layer2, self.weights3))
+
+        return self.layer3
+
+    def back_propagation(self):
+        """
+        backpropagating between the layers using sigmoid derivative and
+        loss between layers
+        updates the weights between the layers
+        """
+
+        updated_weights3 = numpy.dot(
+            self.layer2.T, 2 * (self.y - self.output) * sigmoid_derivative(self.output)
+        )
+        updated_weights2 = numpy.dot(
+            self.layer1.T,
+            numpy.dot(
+                2 * (self.y - self.output) * sigmoid_derivative(self.output),
+                self.weights3.T,
+            )
+            * sigmoid_derivative(self.layer2),
+        )
+        updated_weights1 = numpy.dot(
+            self.input.T,
+            numpy.dot(
+                numpy.dot(
+                    2 * (self.y - self.output) * sigmoid_derivative(self.output),
+                    self.weights3.T,
+                )
+                * sigmoid_derivative(self.layer2),
+                self.weights2.T,
+            )
+            * sigmoid_derivative(self.layer1),
+        )
+
+        self.weights1 += updated_weights1
+        self.weights2 += updated_weights2
+        self.weights3 += updated_weights3
+
+    def train(self, output: numpy.array, iterations: int):
+        """
+        output : required for calculating loss
+        performs the feeding and back propagation process for the
+        given number of iterations
+        every iteration will update the weights of neural network
+        """
+        for iteration in range(1, iterations + 1):
+            self.output = self.feedforward()
+            self.back_propagation()
+            print(
+                "Iteration %s " % iteration,
+                "Loss: " + str(numpy.mean(numpy.square(output - self.feedforward()))),
+            )
+
+    def predict(self, input: list) -> int:
+        """
+        predict's output for the given input values
+        """
+        self.array = numpy.array((input), dtype=float)
+        self.layer1 = sigmoid(numpy.dot(self.array, self.weights1))
+        self.layer2 = sigmoid(numpy.dot(self.layer1, self.weights2))
+        self.layer3 = sigmoid(numpy.dot(self.layer2, self.weights3))
+        if self.layer3 >= 0.5:
+            return 1
+        else:
+            return 0
+
+
+def sigmoid(value: float) -> float:
+    """
+    applies sigmoid activation function
+    return normalized values
+
+    >>> sigmoid(2)
+    0.8807970779778823
+
+    >>> sigmoid(0)
+    0.5
+    """
+    return 1 / (1 + numpy.exp(-value))
+
+
+def sigmoid_derivative(value: float) -> float:
+    """
+    returns derivative of the sigmoid value
+
+    >>> sigmoid_derivative(0.7)
+    0.22171287329310904
+    """
+    return sigmoid(value) * (1 - sigmoid(value))
+
+
+def example():
+    # input values
+    input = numpy.array(
+        (
+            [0, 0, 0],
+            [0, 0, 1],
+            [0, 1, 0],
+            [0, 1, 1],
+            [1, 0, 0],
+            [1, 0, 1],
+            [1, 1, 0],
+            [1, 1, 1],
+        ),
+        dtype=float,
+    )
+
+    # true output for the given input values
+    output = numpy.array(([0], [1], [1], [0], [1], [0], [0], [1]), dtype=float)
+
+    # calling neural network class
+    Neural_Network = NeuralNetwork(input_array=input, output_array=output)
+
+    # calling training function
+    Neural_Network.train(output=output, iterations=1000)
+    print(Neural_Network.predict([0, 1, 1]))
+
+
+if __name__ == "__main__":
+    example()