Playground / LeNet¶

Here we use LeNet as an example to show how to train a model in TensorFlow.

In [1]:

import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
from keras import layers, models
from keras.datasets import mnist
from keras.utils import to_categorical

import tensorflow as tf import numpy as np import matplotlib.pyplot as plt from keras import layers, models from keras.datasets import mnist from keras.utils import to_categorical

In [2]:

print(tf.config.experimental.list_physical_devices('GPU'))

print(tf.config.experimental.list_physical_devices('GPU'))

[]

In [3]:

(train_imgs, train_labels), (test_imgs, test_labels) = mnist.load_data()

print(train_imgs.shape)  # (60000, 28, 28)

(train_imgs, train_labels), (test_imgs, test_labels) = mnist.load_data() print(train_imgs.shape) # (60000, 28, 28)

(60000, 28, 28)

In [4]:

models = models.Sequential(
    [
        layers.Conv2D(
            filters=6,
            kernel_size=5,
            activation="sigmoid",
            padding="same",
        ),
        layers.AvgPool2D(pool_size=2, strides=2),
        layers.Conv2D(
            filters=16,
            kernel_size=5,
            activation="sigmoid",
            padding="valid",
        ),
        layers.AvgPool2D(pool_size=2, strides=2),
        layers.Flatten(),
        layers.Dense(120, activation="sigmoid"),
        layers.Dense(84, activation="sigmoid"),
        layers.Dense(10, activation="softmax"),
    ]
)

models = models.Sequential( [ layers.Conv2D( filters=6, kernel_size=5, activation="sigmoid", padding="same", ), layers.AvgPool2D(pool_size=2, strides=2), layers.Conv2D( filters=16, kernel_size=5, activation="sigmoid", padding="valid", ), layers.AvgPool2D(pool_size=2, strides=2), layers.Flatten(), layers.Dense(120, activation="sigmoid"), layers.Dense(84, activation="sigmoid"), layers.Dense(10, activation="softmax"), ] )

In [5]:

X = tf.random.uniform((1, 28, 28, 1))
for layer in models.layers:
    X = layer(X)
    print(layer.__class__.__name__, "output shape:\t", X.shape)

X = tf.random.uniform((1, 28, 28, 1)) for layer in models.layers: X = layer(X) print(layer.__class__.__name__, "output shape:\t", X.shape)

Conv2D output shape:	 (1, 28, 28, 6)
AveragePooling2D output shape:	 (1, 14, 14, 6)
Conv2D output shape:	 (1, 10, 10, 16)
AveragePooling2D output shape:	 (1, 5, 5, 16)
Flatten output shape:	 (1, 400)
Dense output shape:	 (1, 120)
Dense output shape:	 (1, 84)
Dense output shape:	 (1, 10)

In [6]:

# training:
train_imgs = train_imgs.reshape((60000, 28, 28, 1))
train_imgs = train_imgs.astype("float32") / 255
test_imgs = test_imgs.reshape((10000, 28, 28, 1))
test_imgs = test_imgs.astype("float32") / 255
train_labels_ = to_categorical(train_labels)
test_labels_ = to_categorical(test_labels)

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

history = models.fit(
    train_imgs,
    train_labels_,
    epochs=20,
    batch_size=128,
    validation_data=(test_imgs, test_labels_),
)

# training: train_imgs = train_imgs.reshape((60000, 28, 28, 1)) train_imgs = train_imgs.astype("float32") / 255 test_imgs = test_imgs.reshape((10000, 28, 28, 1)) test_imgs = test_imgs.astype("float32") / 255 train_labels_ = to_categorical(train_labels) test_labels_ = to_categorical(test_labels) models.compile( optimizer="adam", loss="categorical_crossentropy", metrics=["accuracy"], ) history = models.fit( train_imgs, train_labels_, epochs=20, batch_size=128, validation_data=(test_imgs, test_labels_), )

Epoch 1/20
469/469 ━━━━━━━━━━━━━━━━━━━━ 8s 15ms/step - accuracy: 0.2700 - loss: 2.0008 - val_accuracy: 0.8641 - val_loss: 0.5076
Epoch 2/20
469/469 ━━━━━━━━━━━━━━━━━━━━ 7s 14ms/step - accuracy: 0.8787 - loss: 0.4315 - val_accuracy: 0.9188 - val_loss: 0.2758
Epoch 3/20
469/469 ━━━━━━━━━━━━━━━━━━━━ 6s 14ms/step - accuracy: 0.9210 - loss: 0.2627 - val_accuracy: 0.9407 - val_loss: 0.1938
Epoch 4/20
469/469 ━━━━━━━━━━━━━━━━━━━━ 7s 14ms/step - accuracy: 0.9413 - loss: 0.1927 - val_accuracy: 0.9482 - val_loss: 0.1610
Epoch 5/20
469/469 ━━━━━━━━━━━━━━━━━━━━ 7s 14ms/step - accuracy: 0.9536 - loss: 0.1529 - val_accuracy: 0.9630 - val_loss: 0.1234
Epoch 6/20
469/469 ━━━━━━━━━━━━━━━━━━━━ 7s 14ms/step - accuracy: 0.9609 - loss: 0.1269 - val_accuracy: 0.9646 - val_loss: 0.1162
Epoch 7/20
469/469 ━━━━━━━━━━━━━━━━━━━━ 7s 14ms/step - accuracy: 0.9660 - loss: 0.1097 - val_accuracy: 0.9738 - val_loss: 0.0890
Epoch 8/20
469/469 ━━━━━━━━━━━━━━━━━━━━ 7s 15ms/step - accuracy: 0.9703 - loss: 0.0950 - val_accuracy: 0.9763 - val_loss: 0.0778
Epoch 9/20
469/469 ━━━━━━━━━━━━━━━━━━━━ 7s 16ms/step - accuracy: 0.9742 - loss: 0.0829 - val_accuracy: 0.9780 - val_loss: 0.0726
Epoch 10/20
469/469 ━━━━━━━━━━━━━━━━━━━━ 8s 17ms/step - accuracy: 0.9765 - loss: 0.0762 - val_accuracy: 0.9782 - val_loss: 0.0693
Epoch 11/20
469/469 ━━━━━━━━━━━━━━━━━━━━ 7s 16ms/step - accuracy: 0.9806 - loss: 0.0655 - val_accuracy: 0.9823 - val_loss: 0.0571
Epoch 12/20
469/469 ━━━━━━━━━━━━━━━━━━━━ 8s 17ms/step - accuracy: 0.9814 - loss: 0.0601 - val_accuracy: 0.9817 - val_loss: 0.0599
Epoch 13/20
469/469 ━━━━━━━━━━━━━━━━━━━━ 7s 15ms/step - accuracy: 0.9835 - loss: 0.0544 - val_accuracy: 0.9840 - val_loss: 0.0507
Epoch 14/20
469/469 ━━━━━━━━━━━━━━━━━━━━ 7s 16ms/step - accuracy: 0.9840 - loss: 0.0515 - val_accuracy: 0.9819 - val_loss: 0.0555
Epoch 15/20
469/469 ━━━━━━━━━━━━━━━━━━━━ 8s 17ms/step - accuracy: 0.9848 - loss: 0.0477 - val_accuracy: 0.9829 - val_loss: 0.0529
Epoch 16/20
469/469 ━━━━━━━━━━━━━━━━━━━━ 8s 17ms/step - accuracy: 0.9861 - loss: 0.0445 - val_accuracy: 0.9843 - val_loss: 0.0479
Epoch 17/20
469/469 ━━━━━━━━━━━━━━━━━━━━ 7s 15ms/step - accuracy: 0.9865 - loss: 0.0424 - val_accuracy: 0.9842 - val_loss: 0.0477
Epoch 18/20
469/469 ━━━━━━━━━━━━━━━━━━━━ 7s 15ms/step - accuracy: 0.9877 - loss: 0.0398 - val_accuracy: 0.9860 - val_loss: 0.0419
Epoch 19/20
469/469 ━━━━━━━━━━━━━━━━━━━━ 7s 15ms/step - accuracy: 0.9885 - loss: 0.0384 - val_accuracy: 0.9847 - val_loss: 0.0442
Epoch 20/20
469/469 ━━━━━━━━━━━━━━━━━━━━ 7s 15ms/step - accuracy: 0.9896 - loss: 0.0354 - val_accuracy: 0.9852 - val_loss: 0.0441

In [8]:

# plot:
plt.subplot(121)
plt.plot(history.history["accuracy"], label="accuracy")
plt.plot(history.history["val_accuracy"], label="val_accuracy")
plt.title("accuracy")
_ = plt.legend()
plt.subplot(122)
plt.plot(history.history["loss"], label="train_loss")
plt.plot(history.history["val_loss"], label="val_loss")
plt.title("loss")
_ = plt.legend()

# plot: plt.subplot(121) plt.plot(history.history["accuracy"], label="accuracy") plt.plot(history.history["val_accuracy"], label="val_accuracy") plt.title("accuracy") _ = plt.legend() plt.subplot(122) plt.plot(history.history["loss"], label="train_loss") plt.plot(history.history["val_loss"], label="val_loss") plt.title("loss") _ = plt.legend()

In [ ]: