Happiness: Learning Neural networks 3

-*- coding: utf-8 -*-

"""

@author: Balaji

"""

print('\n 18 March Program for a simple neural network, doing a forward pass with one layer')

print('\n 21 March Added provision for data generation for use in training.')

print('\n 21 March Formating input data as a matrix for faster multiplication with weights')

print('\n 22 March Initialise bias to 0, weights to very small ~0.01 random numbers, put computations inside a class')

import numpy as np

import matplotlib.pyplot as plt

class Layer_dense():

def __init__(self, num_pixels, num_neurons):

np.random.seed(1); #For repeatability

self.weights = np.random.randn(num_neurons ,num_pixels);

self.biases = np.random.randn(num_neurons ,1);

#self.output_beforeActivation = np.zeros((num_neurons,1));

def forward(self, inputs):

self.output_beforeActivation = np.dot(self.weights,inputs)+self.biases;

def applyReluActivation (self):

self.output_afterActivation = np.maximum(0,self.output_beforeActivation);

num_neurons = 4;

num_pixels = 5;

num_images = 10;

# Input data set properties

slope = np.tan(np.deg2rad(30));

intercept = 50;

print('\n Creating input data:')

np.random.seed(1); #For repeatability

inputs = np.zeros((num_pixels,num_images));

x = np.arange(0,num_pixels);

x = x.reshape((num_pixels,1));

for dataSet_iter in range(num_images):

    inputs[:,dataSet_iter:dataSet_iter+1] = (slope*x+intercept) + (intercept*0.1)*(np.random.randn(x.size,1));

print('\n Plotting input data set :')

plt.plot(x,inputs,'o')

plt.xlabel('Pixel number')

plt.ylabel('Pixel value')

plt.title('Input data set')

plt.legend()

plt.show()

print('\n Create a dense layer, fully connected, of the class Layer_dense')

firstLayer = Layer_dense(num_pixels = num_pixels, num_neurons = num_neurons);

firstLayer.forward(inputs=inputs);

firstLayer.applyReluActivation();

print('\n Result = ')

print('\n Each column is result of the network per image, after activation')

Happiness