深度学习之-01

深度学习近来火爆， 正好C站也有相关活动，参与了大神K同学的【21天学习挑战赛】K同学啊 邀你参加深度学习研讨班，趁下班时间练练手，把fashion服装数据集跟着博主抄了一下作业。

下载数据

import tensorflow as tf
from tensorflow.keras import datasets, layers, models
import matplotlib.pyplot  as plt
(train_images, train_labels), (test_images, test_labels) = datasets.fashion_mnist.load_data()

数据下载好后， 来看一下数据集的形状：

train_images.shape, train_labels.shape, test_images.shape, test_labels.shape

结果如下：
((60000, 28, 28), (60000,), (10000, 28, 28), (10000,))
目前的像素点值在0~255之间，下面将数据进行简单归一化

train_images, test_images = train_images/255.0, test_images/255.0

归一化后的像素点的数据值将在0~1之间， 适合做学度运算

下面给数据增加维度，以适合涨量运算

train_images = train_images.reshape((60000, 28, 28, 1))
test_images = test_images.reshape((10000, 28, 28, 1))
train_images.shape, train_labels.shape, test_images.shape, test_labels.shape

增加维芳后的数据形状如下：
((60000, 28, 28), (60000,), (10000, 28, 28), (10000,))
数据集中对应的服装名称如下表
class_names = [‘T_shirt/top’, ‘Trouser’, ‘Pullover’, ‘Dress’, ‘Coat’, ‘Sandal’, ‘Shirt’, ‘Sneaker’, ‘Bag’, ‘Ankle book’]
下面利用此表，结合matplotlit,打印一些图片出来

plt.figure(figsize=(20, 10))
for i in range(20):
    plt.subplot(5, 10, i+1)
    plt.xticks([])
    plt.yticks([])
    plt.grid(True)
    plt.imshow(train_images[i], cmap=plt.cm.binary)
    plt.xlabel(calass_names[train_labels[i]])
plt.show()

建议模型

model = models.Sequential([
    layers.Conv2D(32, (3, 3), activation='relu', input_shape=(28, 28, 1)),
    layers.MaxPool2D((2, 2)),
    layers.Conv2D(64, (3, 3), activation='relu'),
    layers.MaxPool2D((2, 2)),
    layers.Conv2D(64, (3, 3), activation='relu'),
    
    layers.Flatten(),
    layers.Dense(64, activation='relu'),
    layers.Dense(10)
])
model.summary()

模型信息如下：
Model: “sequential”

Layer (type) Output Shape Param #

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

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

conv2d_1 (Conv2D) (None, 11, 11, 64) 18496

max_pooling2d_1 (MaxPooling (None, 5, 5, 64) 0
2D)

conv2d_2 (Conv2D) (None, 3, 3, 64) 36928

flatten (Flatten) (None, 576) 0

dense (Dense) (None, 64) 36928

dense_1 (Dense) (None, 10) 650

=================================================================
Total params: 93,322
Trainable params: 93,322
Non-trainable params: 0

编译模型，设定超参数

model.compile(optimizer='adam',
             loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
             metrics=['accuracy'])

到些为止，模型与数据已准备好了，下面用前面准备好的训练数据，对模型进行训练：

history = model.fit(train_images, train_labels, epochs=10,
                   validation_data=(test_images, test_labels))

经过10轮训练后，模型的精度达到90.6%
Epoch 1/10
1875/1875 [] - 79s 42ms/step - loss: 0.5033 - accuracy: 0.8181 - val_loss: 0.3700 - val_accuracy: 0.8642
Epoch 2/10
1875/1875 [] - 84s 45ms/step - loss: 0.3234 - accuracy: 0.8823 - val_loss: 0.3330 - val_accuracy: 0.8831
Epoch 3/10
1875/1875 [] - 89s 48ms/step - loss: 0.2747 - accuracy: 0.9002 - val_loss: 0.2929 - val_accuracy: 0.8951
Epoch 4/10
1875/1875 [] - 96s 51ms/step - loss: 0.2432 - accuracy: 0.9101 - val_loss: 0.2674 - val_accuracy: 0.9014
Epoch 5/10
1875/1875 [] - 93s 50ms/step - loss: 0.2184 - accuracy: 0.9189 - val_loss: 0.2945 - val_accuracy: 0.8900
Epoch 6/10
1875/1875 [] - 93s 50ms/step - loss: 0.1997 - accuracy: 0.9258 - val_loss: 0.2577 - val_accuracy: 0.9090
Epoch 7/10
1875/1875 [] - 95s 51ms/step - loss: 0.1791 - accuracy: 0.9330 - val_loss: 0.2621 - val_accuracy: 0.9064
Epoch 8/10
1875/1875 [] - 89s 47ms/step - loss: 0.1666 - accuracy: 0.9371 - val_loss: 0.2994 - val_accuracy: 0.9002
Epoch 9/10
1875/1875 [] - 78s 41ms/step - loss: 0.1518 - accuracy: 0.9434 - val_loss: 0.2773 - val_accuracy: 0.9046
Epoch 10/10
1875/1875 [] - 78s 42ms/step - loss: 0.1411 - accuracy: 0.9476 - val_loss: 0.2913 - val_accuracy: 0.9065

可视化

将训练数据与测试数据的准确率放在同一图中进行对比

plt.plot(history.history['accuracy'], label='accuracy')
plt.plot(history.history['val_accuracy'], label='val_accuracy')
plt.xlabel('Epoch')
plt.ylabel('Accuracy')
plt.ylim([0.5, 1])
plt.legend(loc='lower right')
plt.show()

从图上可以看出， 蓝色的训练数据的精度随着训练次数的增加，精度在不断提升。而黄色线代表的测数数据的精度比较平稳，说明此模型比较好。