3/27/22, 7:19 PM mnist-CNN.

ipynb - Colaboratory

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers
from keras import Sequential

import numpy as np

mnist_dataset = tf.keras.datasets.mnist.load_data(path="mnist.npz")
 
(x_train, y_train), (x_test, y_test) = mnist_dataset
 
print(x_train.shape)
print(y_train.shape)
 
print(x_test.shape)
print(y_test.shape)
num_classes = 10
input_shape = (28, 28, 1)

x_train = x_train.astype("float32") / 255
x_test = x_test.astype("float32") / 255

# Make sure images have shape (28, 28, 1)
x_train = np.expand_dims(x_train, -1)
x_test = np.expand_dims(x_test, -1)

print("x_train shape:", x_train.shape)
print(x_train.shape[0], "train samples")
print(x_test.shape[0], "test samples")

# convert class vectors to binary class matrices
y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)
print(y_train)

https://colab.research.google.com/drive/17fvgPU817qzmm-UXrfDAKjHN1dpkPZcZ#printMode=true 1/5
3/27/22, 7:19 PM mnist-CNN.ipynb - Colaboratory

(60000, 28, 28)

(60000,)

(10000, 28, 28)

(10000,)

x_train shape: (60000, 28, 28, 1)

10000 test samples

[[0. 0. 0. ... 0. 0. 0.]

[1. 0. 0. ... 0. 0. 0.]

[0. 0. 0. ... 0. 0. 0.]

...

[0. 0. 0. ... 0. 0. 0.]

[0. 0. 0. ... 0. 0. 0.]

[0. 0. 0. ... 0. 1. 0.]]

model = keras.Sequential(

    [

        keras.Input(shape=input_shape),

        layers.Conv2D(32, kernel_size=(3, 3), activation="relu"),

        layers.MaxPooling2D(pool_size=(2, 2)),

        layers.Conv2D(64, kernel_size=(3, 3), activation="relu"),

        layers.MaxPooling2D(pool_size=(2, 2)),

        layers.Flatten(),

        layers.Dropout(0.5),

        layers.Dense(num_classes, activation="softmax"),

    ]

model.summary()

Model: "sequential"

_________________________________________________________________

Layer (type) Output Shape Param #

=================================================================

conv2d (Conv2D) (None, 26, 26, 32) 320

max_pooling2d (MaxPooling2D (None, 13, 13, 32) 0

https://colab.research.google.com/drive/17fvgPU817qzmm-UXrfDAKjHN1dpkPZcZ#printMode=true 2/5
3/27/22, 7:19 PM mnist-CNN.ipynb - Colaboratory

conv2d_1 (Conv2D) (None, 11, 11, 64)

max_pooling2d_1 (MaxPooling (None, 5, 5, 64) 0

2D)

flatten (Flatten) (None, 1600) 0

dropout (Dropout) (None, 1600) 0

dense (Dense) (None, 10) 16010

=================================================================

Total params: 34,826

Trainable params: 34,826

Non-trainable params: 0

_________________________________________________________________

batch_size = 128

epochs = 15

model.compile(loss="categorical_crossentropy", optimizer="adam", metrics=["accuracy"])

model.fit(x_train, y_train, batch_size=batch_size, epochs=15, validation_split=0.1)

Epoch 1/15

422/422 [==============================] - 43s 100ms/step - loss: 0.3684 - accuracy: 0.8887 - val_loss: 0.0835 - val_accuracy:
Epoch 2/15

422/422 [==============================] - 41s 98ms/step - loss: 0.1142 - accuracy: 0.9654 - val_loss: 0.0617 - val_accuracy: 0
Epoch 3/15

422/422 [==============================] - 41s 97ms/step - loss: 0.0863 - accuracy: 0.9733 - val_loss: 0.0518 - val_accuracy: 0
Epoch 4/15

422/422 [==============================] - 41s 97ms/step - loss: 0.0747 - accuracy: 0.9773 - val_loss: 0.0431 - val_accuracy: 0
Epoch 5/15

422/422 [==============================] - 41s 97ms/step - loss: 0.0637 - accuracy: 0.9804 - val_loss: 0.0398 - val_accuracy: 0
Epoch 6/15

422/422 [==============================] - 41s 98ms/step - loss: 0.0579 - accuracy: 0.9815 - val_loss: 0.0394 - val_accuracy: 0
Epoch 7/15

422/422 [==============================] - 41s 98ms/step - loss: 0.0535 - accuracy: 0.9830 - val_loss: 0.0348 - val_accuracy: 0
Epoch 8/15

https://colab.research.google.com/drive/17fvgPU817qzmm-UXrfDAKjHN1dpkPZcZ#printMode=true 3/5
3/27/22, 7:19 PM mnist-CNN.ipynb - Colaboratory

422/422 [==============================] - 42s 99ms/step - loss: 0.0490 - accuracy: 0.9850 - val_loss: 0.0352 - val_accuracy: 0
Epoch 9/15

422/422 [==============================] - 42s 99ms/step - loss: 0.0460 - accuracy: 0.9856 - val_loss: 0.0319 - val_accuracy: 0
Epoch 10/15

422/422 [==============================] - 42s 100ms/step - loss: 0.0416 - accuracy: 0.9871 - val_loss: 0.0334 - val_accuracy:
Epoch 11/15

422/422 [==============================] - 42s 100ms/step - loss: 0.0403 - accuracy: 0.9865 - val_loss: 0.0341 - val_accuracy:
Epoch 12/15

422/422 [==============================] - 42s 100ms/step - loss: 0.0400 - accuracy: 0.9872 - val_loss: 0.0341 - val_accuracy:
Epoch 13/15

422/422 [==============================] - 42s 99ms/step - loss: 0.0381 - accuracy: 0.9874 - val_loss: 0.0323 - val_accuracy: 0
Epoch 14/15

422/422 [==============================] - 42s 100ms/step - loss: 0.0344 - accuracy: 0.9888 - val_loss: 0.0317 - val_accuracy:
Epoch 15/15

422/422 [==============================] - 42s 99ms/step - loss: 0.0337 - accuracy: 0.9892 - val_loss: 0.0302 - val_accuracy: 0
<keras.callbacks.History at 0x7fc06ca2da10>

score = model.evaluate(x_test, y_test, verbose=10)
print(len(score))
print("Test loss:", score[0])
print("Test accuracy:", score[1])

Test loss: 0.025434585288167

Test accuracy: 0.9914000034332275

https://colab.research.google.com/drive/17fvgPU817qzmm-UXrfDAKjHN1dpkPZcZ#printMode=true 4/5
3/27/22, 7:19 PM mnist-CNN.ipynb - Colaboratory

https://colab.research.google.com/drive/17fvgPU817qzmm-UXrfDAKjHN1dpkPZcZ#printMode=true 5/5