FullMNISTExample.py

# First, install TensorFlow (https://www.tensorflow.org/) and Keras (https://keras.io/).

from keras.datasets import mnist
import numpy as np

(training_images, training_labels), (test_images, test_labels) = mnist.load_data()

#normalize image values
training_images = training_images / 255
test_images = test_images / 255

#flatten image arrays
training_images = training_images.reshape(len(training_images),28*28)
test_images = test_images.reshape(len(test_images),28*28)

#convert labels into arrays
zeroArray = [0,0,0,0,0,0,0,0,0,0]
def transformLabels(labels):
    transformed = []
    for i in range(len(labels)):
        label = labels[i]
        transformed.append(zeroArray[0:label] + [1] + zeroArray[label+1:])
    return transformed
training_labels = transformLabels(training_labels)
test_labels = transformLabels(test_labels)

np.random.seed(10)
weights = np.random.rand(784,10)
alpha = 0.01

def neural_network(input, weights):
    return np.dot(input, weights)

#training phase
num_of_training_images = 40000
for iteration in range(num_of_training_images):
    image = training_images[iteration]
    label = training_labels[iteration]
    
    pred = neural_network(image, weights)
    
    error = [0,0,0,0,0,0,0,0,0,0]
    delta = [0,0,0,0,0,0,0,0,0,0]
    
    for i in range(len(goal_pred)):
        error[i] = (pred[i] - label[i]) ** 2
        delta[i] = pred[i] - label[i]
    
    weight_deltas = np.outer(image, delta)
    
    for i in range(len(weights)):
        for j in range(len(weights[i])):
            weights[i][j] -= alpha * weight_deltas[i][j]

#testing phase
num_correct = 0
for i in range(len(test_images)):
    image = test_images[i]
    label = test_labels[i]
    
    pred = neural_network(image, weights)
    
    if (np.argmax(label) == np.argmax(pred)):
        num_correct += 1

print (str(num_correct) + "/" + str(len(test_images)))