Neural networks: Building a neural network that classifies MNIST images & time travel testing with that.

Author: DevOpsThinh

fundamentals/scientific_libraries/it_student_k19.csv renamed to fundamentals/datasets/it_student_k19.csv

@@ -17,9 +17,9 @@
 
 student_df = pd.DataFrame(student_dict)
 student_df.to_csv('it_student_k19.csv')
-my_data = pd.read_csv('it_student_k19.csv', index_col=0)
+my_data = pd.read_csv('../datasets/it_student_k19.csv', index_col=0)
 
-income_data = pd.read_csv('us-income-annual.csv', delimiter=';')
+income_data = pd.read_csv('../datasets/us-income-annual.csv', delimiter=';')
 print('DataFrame shape is: ', income_data.shape)
 print(income_data.keys())
 print('\nRevenue:\n', income_data['Revenue'])  # view a single column series

fundamentals/scientific_libraries/us-income-annual.csv renamed to fundamentals/datasets/us-income-annual.csv

@@ -9,6 +9,7 @@
 
 
 def classify(x, w):
+    """Predict the value of an unlabeled image"""
     y_hat = forward(x, w)
     labels = np.argmax(y_hat, axis=1)
     return labels.reshape(-1, 1)

fundamentals/scientific_libraries/pandas_lib.py

@@ -10,7 +10,7 @@
 import matplotlib.pyplot as plt
 import numpy as np
 
-from ml.util import one_hot_encoding
+from ml.util import prepend_bias, one_hot_encoding
 
 TRAIN_IMAGE = "../../../fundamentals/datasets/mnist/train-images-idx3-ubyte.gz"
 TRAIN_LABEL = "../../../fundamentals/datasets/mnist/train-labels-idx1-ubyte.gz"
@@ -26,22 +26,17 @@ def load_images(filename):
         return all_pixels.reshape(n_images, columns * rows)
 
 
-def prepend_bias(x):
-    """
-    Insert a column of 1s in the position 0 of X.
-    :param x: X examples - a matrix
-    :return: A new matrix
-    """
-    return np.insert(x, 0, 1, axis=1)
-
-
 # 60000 images, each 785 (1 bias + 28 * 28) elements
 X_train_data = prepend_bias(load_images(TRAIN_IMAGE))
 print(X_train_data.shape)
 # 10000 images, each 785 (1 bias + 28 * 28) elements
 X_test_data = prepend_bias(load_images(TEST_IMAGE))
 print(X_test_data.shape)
 
+# Neural networks #
+X_train = load_images(TRAIN_IMAGE)  # no bias - 784 elements
+X_test = load_images(TEST_IMAGE)  # no bias - 784 elements
+
 
 def load_labels(filename):
     """Loads MNIST labels into a Numpy array, then molds that array into a
@@ -85,10 +80,11 @@ def show_a_digit(num):
 print(Y_train_data.shape)
 
 Y_train_unencoded = load_labels(TRAIN_LABEL)
+# 60000 labels, each consisting of 10 one-hot encoded elements
 Y_train = one_hot_encoding(Y_train_unencoded, 10)
 
 # 10000 labels
 Y_test_data = encode_fives(load_labels(TEST_LABEL))
 print(Y_test_data.shape)
-
+# 10000 labels, each a single digit from 0 to 9
 Y_test = load_labels(TEST_LABEL)

ml/supervised_learning/classifications/base_classifier.py

@@ -0,0 +1,61 @@
+# Learner: Nguyen Truong Thinh
+# Contact me: [email protected] || +84393280504
+#
+# Topic: Supervised Learning: Classifications in Action.
+#           Neural Networks: Are leaps & bounds more powerful than perceptrons.
+#           The Network's classification: A Classifier's Answers
+
+import json
+
+import numpy as np
+
+from ml.supervised_learning.classifications import our_own_mnist_lib as mnist
+from ml.util import sigmoid, prepend_bias
+
+
+def loss(y, y_hat):
+    """The log loss - cross entropy loss for our binary classifier"""
+    return -np.sum(y * np.log(y_hat)) / y.shape[0]
+
+
+def softmax(logits):
+    """The Softmax function"""
+    exponentials = np.exp(logits)
+    return exponentials / np.sum(exponentials, axis=1).reshape(-1, 1)
+
+
+def forward(x, w1, w2):
+    """
+    Forward propagation.
+    Calculates the system's outputs from the system's inputs.
+    """
+    # Calculate the hidden layer
+    h = sigmoid(np.matmul(prepend_bias(x), w1))
+    # Calculate the output layer
+    y_hat = softmax(np.matmul(prepend_bias(h), w2))
+    return y_hat
+
+
+def classify(x, w1, w2):
+    """Predict the value of an unlabeled image"""
+    y_hat = forward(x, w1, w2)
+    labels = np.argmax(y_hat, axis=1)
+    return labels.reshape(-1, 1)
+
+
+def report(iteration, x_train, y_train, x_test, y_test, w1, w2):
+    """To check how well the system is learning"""
+    y_hat = forward(x_train, w1, w2)
+    training_loss = loss(y_train, y_hat)
+    classifications = classify(x_test, w1, w2)
+    accuracy = np.average(classifications == y_test) * 100.0
+    print("Iteration: %5d, Loss: %.6f, Accuracy: %.2f%%" % (iteration, training_loss, accuracy))
+
+
+# Time travel testing section #
+with open('weights.json') as f:
+    weights = json.load(f)
+
+weight1, weight2 = (np.array(weights[0]), np.array(weights[1]))
+report(0, mnist.X_train, mnist.Y_train, mnist.X_test, mnist.Y_test, weight1, weight2)
+# Iteration: 0, Loss: 2.221439, Accuracy: 43.19%

ml/supervised_learning/classifications/our_own_mnist_lib.py

@@ -27,6 +27,15 @@ def one_hot_encoding(y, a_number):
     return encoded_y
 
 
+def prepend_bias(x):
+    """
+    Insert a column of 1s in the position 0 of X.
+    :param x: X examples - a matrix
+    :return: A new matrix
+    """
+    return np.insert(x, 0, 1, axis=1)
+
+
 def test(x, _y, _w):
     total_examples = x.shape[0]
     correct_results = np.sum(classify(x, _w) == _y)

ml/supervised_learning/neural_networks/building_the_network.py

@@ -8,8 +8,7 @@
 
 # Get csv data
 income_data = pd.read_csv(
-    'D:/PythonInPractice/Coding-Deep-Learning/fundamentals/scientific_libraries/'
-    'us-income-annual.csv', delimiter=';')
+    '/fundamentals/datasets/us-income-annual.csv', delimiter=';')
 stock_prices = pd.read_csv('us-shareprices-daily.csv', delimiter=';')
 # Retrieve info
 print('Income data size is: ', income_data.shape)