The CIFAR-10 dataset (Canadian Institute For Advanced Research) is a collection of images that are commonly used to train machine learning and computer vision algorithms. It is one of the most widely used datasets for machine learning research. The CIFAR-10 dataset contains 60,000 32x32 color images in 10 different classes. The 10 different classes represent airplanes, cars, birds, cats, deer, dogs, frogs, horses, ships, and trucks. There are 6,000 images of each class.
import numpy as np
import pandas as pd
from packaging import version
from sklearn.metrics import confusion_matrix, classification_report
from sklearn.metrics import accuracy_score
from sklearn.metrics import mean_squared_error as MSE
from sklearn.model_selection import train_test_split
import matplotlib.pyplot as plt
import seaborn as sns
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import models, layers
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Conv2D, MaxPool2D, BatchNormalization, Dropout, Flatten, Dense
from tensorflow.keras.callbacks import ModelCheckpoint, EarlyStopping
from tensorflow.keras.preprocessing import image
from tensorflow.keras.utils import to_categorical
%matplotlib inline
np.set_printoptions(precision=3, suppress=True)
print("This notebook requires TensorFlow 2.0 or above")
print("TensorFlow version: ", tf.__version__)
assert version.parse(tf.__version__).release[0] >=2
This notebook requires TensorFlow 2.0 or above TensorFlow version: 2.10.0
print("Keras version: ", keras.__version__)
Keras version: 2.10.0
# from google.colab import drive
# drive.mount('/content/gdrive')
def get_three_classes(x, y):
def indices_of(class_id):
indices, _ = np.where(y == float(class_id))
return indices
indices = np.concatenate([indices_of(0), indices_of(1), indices_of(2)], axis=0)
x = x[indices]
y = y[indices]
count = x.shape[0]
indices = np.random.choice(range(count), count, replace=False)
x = x[indices]
y = y[indices]
y = tf.keras.utils.to_categorical(y)
return x, y
def show_random_examples(x, y, p):
indices = np.random.choice(range(x.shape[0]), 10, replace=False)
x = x[indices]
y = y[indices]
p = p[indices]
plt.figure(figsize=(10, 5))
for i in range(10):
plt.subplot(2, 5, i + 1)
plt.imshow(x[i])
plt.xticks([])
plt.yticks([])
col = 'green' if np.argmax(y[i]) == np.argmax(p[i]) else 'red'
plt.xlabel(class_names_preview[np.argmax(p[i])], color=col)
plt.show()
def plot_history(history):
losses = history.history['loss']
accs = history.history['accuracy']
val_losses = history.history['val_loss']
val_accs = history.history['val_accuracy']
epochs = len(losses)
plt.figure(figsize=(16, 4))
for i, metrics in enumerate(zip([losses, accs], [val_losses, val_accs], ['Loss', 'Accuracy'])):
plt.subplot(1, 2, i + 1)
plt.plot(range(epochs), metrics[0], label='Training {}'.format(metrics[2]))
plt.plot(range(epochs), metrics[1], label='Validation {}'.format(metrics[2]))
plt.legend()
plt.show()
def print_validation_report(y_test, predictions):
print("Classification Report")
print(classification_report(y_test, predictions))
print('Accuracy Score: {}'.format(accuracy_score(y_test, predictions)))
print('Root Mean Square Error: {}'.format(np.sqrt(MSE(y_test, predictions))))
def plot_confusion_matrix(y_true, y_pred):
mtx = confusion_matrix(y_true, y_pred)
fig, ax = plt.subplots(figsize=(8,8))
sns.heatmap(mtx, annot=True, fmt='d', linewidths=.75, cbar=False, ax=ax,cmap='Blues',linecolor='white')
# square=True,
plt.ylabel('true label')
plt.xlabel('predicted label')
The CIFAR-10 dataset consists of 60000 32x32 colour images in 10 classes, with 6000 images per class. There are 50000 training images and 10000 test images.
The dataset is divided into five training batches and one test batch, each with 10000 images. The test batch contains exactly 1000 randomly-selected images from each class. The training batches contain the remaining images in random order, but some training batches may contain more images from one class than another. Between them, the training batches contain exactly 5000 images from each class.
(x_train, y_train), (x_test, y_test) = keras.datasets.cifar10.load_data()
print('train_images:\t{}'.format(x_train.shape))
print('train_labels:\t{}'.format(y_train.shape))
print('test_images:\t\t{}'.format(x_test.shape))
print('test_labels:\t\t{}'.format(y_test.shape))
train_images: (50000, 32, 32, 3) train_labels: (50000, 1) test_images: (10000, 32, 32, 3) test_labels: (10000, 1)
print("First ten labels training dataset:\n {}\n".format(y_train[0:10]))
print("This output the numeric label, need to convert to item description")
First ten labels training dataset: [[6] [9] [9] [4] [1] [1] [2] [7] [8] [3]] This output the numeric label, need to convert to item description
(train_images, train_labels),(test_images, test_labels)= tf.keras.datasets.cifar10.load_data()
x_preview, y_preview = get_three_classes(train_images, train_labels)
x_preview, y_preview = get_three_classes(test_images, test_labels)
class_names_preview = ['aeroplane', 'car', 'bird']
show_random_examples(x_preview, y_preview, y_preview)
The labels are an array of integers, ranging from 0 to 9. These correspond to the class of clothing the image represents:
| Label | Class_ |
|---|---|
| 0 | airplane |
| 1 | automobile |
| 2 | bird |
| 3 | cat |
| 4 | deer |
| 5 | dog |
| 6 | frog |
| 7 | horse |
| 8 | ship |
| 9 | truck |
class_names = ['airplane'
,'automobile'
,'bird'
,'cat'
,'deer'
,'dog'
,'frog'
,'horse'
,'ship'
,'truck']
x_train_split, x_valid_split, y_train_split, y_valid_split = train_test_split(x_train
,y_train
,test_size=.1
,random_state=42
,shuffle=True)
print(x_train_split.shape, x_valid_split.shape, x_test.shape)
(45000, 32, 32, 3) (5000, 32, 32, 3) (10000, 32, 32, 3)
The images are 28x28 NumPy arrays, with pixel values ranging from 0 to 255
x_train_norm = x_train_split/255
x_valid_norm = x_valid_split/255
x_test_norm = x_test/255
We use a Sequential class defined in Keras to create our model. The first 9 layers Conv2D MaxPooling handle feature learning. The last 3 layers, handle classification
model = Sequential([
Flatten(input_shape=x_train_norm.shape[1:]),
Dense(units=2000,activation=tf.nn.softmax),
Dropout(0.3),
Dense(units=2000,activation=tf.nn.softmax),
Dropout(0.3),
Dense(units=10, activation=tf.nn.softmax)
])
model.summary()
Model: "sequential_1"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
flatten_1 (Flatten) (None, 3072) 0
dense_3 (Dense) (None, 2000) 6146000
dropout (Dropout) (None, 2000) 0
dense_4 (Dense) (None, 2000) 4002000
dropout_1 (Dropout) (None, 2000) 0
dense_5 (Dense) (None, 10) 20010
=================================================================
Total params: 10,168,010
Trainable params: 10,168,010
Non-trainable params: 0
_________________________________________________________________
keras.utils.plot_model(model, "CIFAR10.png", show_shapes=True)
You must install pydot (`pip install pydot`) and install graphviz (see instructions at https://graphviz.gitlab.io/download/) for plot_model to work.
In addition to setting up our model architecture, we also need to define which algorithm should the model use in order to optimize the weights and biases as per the given data. We will use stochastic gradient descent.
We also need to define a loss function. Think of this function as the difference between the predicted outputs and the actual outputs given in the dataset. This loss needs to be minimised in order to have a higher model accuracy. That's what the optimization algorithm essentially does - it minimises the loss during model training. For our multi-class classification problem, categorical cross entropy is commonly used.
Finally, we will use the accuracy during training as a metric to keep track of as the model trains.
model.compile(optimizer='adam',
loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
metrics=['accuracy'])
history = model.fit(x_train_norm
,y_train_split
,epochs=200
,batch_size=64
,validation_data=(x_valid_norm, y_valid_split)
,callbacks=[
tf.keras.callbacks.ModelCheckpoint("CNN_model.h5",save_best_only=True,save_weights_only=False)
,tf.keras.callbacks.EarlyStopping(monitor='val_accuracy', patience=3),
]
)
Epoch 1/200 704/704 [==============================] - 139s 196ms/step - loss: 0.7440 - accuracy: 0.7428 - val_loss: 0.8154 - val_accuracy: 0.7152 Epoch 2/200 704/704 [==============================] - 135s 192ms/step - loss: 0.7170 - accuracy: 0.7510 - val_loss: 0.8042 - val_accuracy: 0.7278 Epoch 3/200 704/704 [==============================] - 136s 193ms/step - loss: 0.6900 - accuracy: 0.7610 - val_loss: 0.7930 - val_accuracy: 0.7250 Epoch 4/200 704/704 [==============================] - 137s 194ms/step - loss: 0.6659 - accuracy: 0.7687 - val_loss: 0.7645 - val_accuracy: 0.7368 Epoch 5/200 704/704 [==============================] - 137s 194ms/step - loss: 0.6529 - accuracy: 0.7727 - val_loss: 0.7673 - val_accuracy: 0.7418 Epoch 6/200 704/704 [==============================] - 133s 189ms/step - loss: 0.6322 - accuracy: 0.7798 - val_loss: 0.7811 - val_accuracy: 0.7318 Epoch 7/200 704/704 [==============================] - 141s 200ms/step - loss: 0.6154 - accuracy: 0.7852 - val_loss: 0.7865 - val_accuracy: 0.7336 Epoch 8/200 704/704 [==============================] - 234s 332ms/step - loss: 0.5954 - accuracy: 0.7934 - val_loss: 0.7545 - val_accuracy: 0.7360
In order to ensure that this is not a simple "memorization" by the machine, we should evaluate the performance on the test set. This is easy to do, we simply use the evaluate method on our model.
model = tf.keras.models.load_model("CNN_model.h5")
print(f"Test acc: {model.evaluate(x_test_norm, y_test)[1]:.3f}")
313/313 [==============================] - 14s 42ms/step - loss: 0.7693 - accuracy: 0.7432 Test acc: 0.743
preds = model.predict(x_test_norm)
print('shape of preds: ', preds.shape)
313/313 [==============================] - 14s 42ms/step shape of preds: (10000, 10)
We use Matplotlib to create 2 plots--displaying the training and validation loss (resp. accuracy) for each (training) epoch side by side.
history_dict = history.history
history_dict.keys()
dict_keys(['loss', 'accuracy', 'val_loss', 'val_accuracy'])
history_df=pd.DataFrame(history_dict)
history_df.tail().round(3)
| loss | accuracy | val_loss | val_accuracy | |
|---|---|---|---|---|
| 3 | 0.666 | 0.769 | 0.765 | 0.737 |
| 4 | 0.653 | 0.773 | 0.767 | 0.742 |
| 5 | 0.632 | 0.780 | 0.781 | 0.732 |
| 6 | 0.615 | 0.785 | 0.786 | 0.734 |
| 7 | 0.595 | 0.793 | 0.754 | 0.736 |
plot_history(history)
Using both sklearn.metrics. Then we visualize the confusion matrix and see what that tells us.
pred1= model.predict(x_test_norm)
pred1=np.argmax(pred1, axis=1)
313/313 [==============================] - 15s 46ms/step
print_validation_report(y_test, pred1)
Classification Report
precision recall f1-score support
0 0.77 0.78 0.78 1000
1 0.87 0.86 0.86 1000
2 0.68 0.62 0.65 1000
3 0.56 0.58 0.57 1000
4 0.73 0.68 0.70 1000
5 0.63 0.67 0.65 1000
6 0.74 0.84 0.78 1000
7 0.82 0.78 0.80 1000
8 0.84 0.83 0.83 1000
9 0.82 0.80 0.81 1000
accuracy 0.74 10000
macro avg 0.74 0.74 0.74 10000
weighted avg 0.74 0.74 0.74 10000
Accuracy Score: 0.7432
Root Mean Square Error: 2.1067985190805505
plot_confusion_matrix(y_test,pred1)
model = tf.keras.models.load_model('CNN_model.h5')
preds = model.predict(x_test_norm)
313/313 [==============================] - 14s 42ms/step
preds.shape
(10000, 10)
cm = sns.light_palette((260, 75, 60), input="husl", as_cmap=True)
df = pd.DataFrame(preds[0:20], columns = ['airplane', 'automobile', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck'])
df.style.format("{:.2%}").background_gradient(cmap=cm)
| airplane | automobile | bird | cat | deer | dog | frog | horse | ship | truck | |
|---|---|---|---|---|---|---|---|---|---|---|
| 0 | 0.07% | 0.00% | 0.14% | 90.68% | 0.09% | 6.29% | 2.18% | 0.02% | 0.51% | 0.02% |
| 1 | 0.19% | 68.41% | 0.00% | 0.00% | 0.00% | 0.00% | 0.00% | 0.00% | 31.32% | 0.08% |
| 2 | 14.15% | 6.17% | 0.25% | 0.68% | 0.06% | 0.05% | 0.03% | 0.36% | 75.80% | 2.46% |
| 3 | 69.21% | 1.61% | 0.36% | 1.27% | 0.04% | 0.00% | 0.00% | 0.01% | 27.48% | 0.01% |
| 4 | 0.00% | 0.00% | 0.65% | 5.31% | 8.84% | 0.12% | 85.08% | 0.00% | 0.00% | 0.00% |
| 5 | 0.01% | 0.02% | 0.09% | 0.90% | 0.62% | 0.65% | 97.23% | 0.43% | 0.01% | 0.03% |
| 6 | 0.95% | 60.11% | 0.03% | 1.86% | 0.00% | 10.45% | 15.15% | 1.14% | 0.01% | 10.30% |
| 7 | 0.32% | 0.00% | 18.37% | 7.52% | 26.52% | 11.35% | 34.28% | 1.62% | 0.00% | 0.03% |
| 8 | 0.04% | 0.00% | 0.53% | 95.70% | 1.02% | 1.11% | 1.10% | 0.46% | 0.03% | 0.01% |
| 9 | 2.12% | 73.87% | 0.11% | 0.06% | 0.13% | 0.01% | 0.06% | 0.03% | 0.21% | 23.41% |
| 10 | 70.61% | 0.21% | 3.73% | 5.32% | 7.09% | 1.70% | 0.12% | 1.97% | 9.02% | 0.25% |
| 11 | 0.00% | 0.41% | 0.00% | 0.00% | 0.00% | 0.00% | 0.00% | 0.00% | 0.00% | 99.59% |
| 12 | 0.00% | 0.02% | 7.60% | 25.75% | 3.07% | 61.98% | 0.44% | 1.08% | 0.04% | 0.01% |
| 13 | 0.00% | 0.04% | 0.01% | 0.14% | 0.11% | 0.14% | 0.01% | 99.55% | 0.00% | 0.01% |
| 14 | 0.00% | 0.05% | 0.00% | 0.00% | 0.00% | 0.00% | 0.00% | 0.01% | 0.00% | 99.92% |
| 15 | 1.37% | 0.46% | 0.46% | 1.64% | 1.39% | 0.59% | 66.84% | 0.00% | 27.22% | 0.02% |
| 16 | 0.00% | 0.03% | 0.05% | 5.64% | 0.01% | 94.04% | 0.16% | 0.03% | 0.00% | 0.04% |
| 17 | 0.23% | 0.01% | 1.20% | 53.60% | 0.69% | 5.78% | 0.51% | 37.32% | 0.07% | 0.57% |
| 18 | 0.50% | 7.69% | 0.01% | 0.02% | 0.00% | 0.01% | 0.00% | 0.01% | 88.61% | 3.14% |
| 19 | 0.00% | 0.00% | 0.02% | 0.07% | 0.03% | 0.02% | 99.86% | 0.00% | 0.00% | 0.00% |
(_,_), (test_images, test_labels) = tf.keras.datasets.cifar10.load_data()
img = test_images[2000]
img_tensor = image.img_to_array(img)
img_tensor = np.expand_dims(img_tensor, axis=0)
class_names = ['airplane'
,'automobile'
,'bird'
,'cat'
,'deer'
,'dog'
,'frog'
,'horse'
,'ship'
,'truck']
plt.imshow(img, cmap='viridis')
plt.axis('off')
plt.show()
# Extracts the outputs of the top 8 layers:
layer_outputs = [layer.output for layer in model.layers[:8]]
# Creates a model that will return these outputs, given the model input:
activation_model = models.Model(inputs=model.input, outputs=layer_outputs)
activations = activation_model.predict(img_tensor)
len(activations)
1/1 [==============================] - 1s 669ms/step
8
layer_names = []
for layer in model.layers:
layer_names.append(layer.name)
layer_names
['conv2d_2', 'max_pooling2d_2', 'dropout_2', 'conv2d_3', 'max_pooling2d_3', 'dropout_3', 'flatten_1', 'dense_1']