MLP实现fashion_mnist数据集分类(1)-模型构建、训练、保存与加载(tensorflow)

1、查看tensorflow版本

import tensorflow as tf

print('Tensorflow Version:{}'.format(tf.__version__))
print(tf.config.list_physical_devices())

2、fashion_mnist数据集下载与展示

(train_image,train_label),(test_image,test_label) = tf.keras.datasets.fashion_mnist.load_data()
print(train_image.shape)
print(train_label.shape)
print(test_image.shape)
print(test_label.shape)

import matplotlib.pyplot as plt
# plt.imshow(train_image[0])  # 此处为啥是彩色的？

def plot_images_lables(images,labels,start_idx,num=5):
    fig = plt.gcf()
    fig.set_size_inches(12,14)
    for i in range(num):
        ax = plt.subplot(1,num,1+i)
        ax.imshow(images[start_idx+i],cmap='binary')
        title = 'label=' + str(labels[start_idx+i])
        ax.set_title(title,fontsize=10)
        ax.set_xticks([])
        ax.set_yticks([])
    plt.show()
plot_images_lables(train_image,train_label,0,5)
# plot_images_lables(test_image,test_label,0,5)

3、数据预处理

X_train,X_test = tf.cast(train_image/255.0,tf.float32),tf.cast(test_image/255.0,tf.float32) # 归一化
y_train,y_test = train_label,test_label # 此处对y没有做onehot处理，需要使用稀疏交叉损失函数

4、模型构建

from keras import Sequential
from keras.layers import Flatten,Dense,Dropout
from keras import Input

model = Sequential()
model.add(Input(shape=(28,28)))
model.add(Flatten())
model.add(Dense(units=256,kernel_initializer='normal',activation='relu'))
model.add(Dropout(rate=0.1))
model.add(Dense(units=64,kernel_initializer='normal',activation='relu'))
model.add(Dropout(rate=0.1))
model.add(Dense(units=10,kernel_initializer='normal',activation='softmax'))
model.summary()

5、模型配置

model.compile(optimizer='adam',
              loss='sparse_categorical_crossentropy',
              metrics=['acc'])

6、模型训练

H = model.fit(x=X_train,
              y=y_train,
              validation_split=0.2,
              # validation_data=(X_test,y_test),
              epochs=10,
              batch_size=128,
              verbose=1)

plt.plot(H.epoch, H.history['loss'], label='loss')
plt.plot(H.epoch, H.history['val_loss'], label='val_loss')
plt.legend()

plt.plot(H.epoch, H.history['acc'], label='acc')
plt.plot(H.epoch, H.history['val_acc'], label='val_acc')
plt.legend()

7、模型评估

model.evaluate(X_test,y_test)

8、模型预测

import numpy as np
import matplotlib.pyplot as plt

def pred_plot_images_lables(images,labels,start_idx,num=5):
    # 预测
    res = model.predict(images[start_idx:start_idx+num])
    res = np.argmax(res,axis=1)

    # 画图
    fig = plt.gcf()
    fig.set_size_inches(12,14)
    for i in range(num):
        ax = plt.subplot(1,num,1+i)
        ax.imshow(images[start_idx+i],cmap='binary')
        title = 'label=' + str(labels[start_idx+i]) + ', pred=' + str(res[i])
        ax.set_title(title,fontsize=10)
        ax.set_xticks([])
        ax.set_yticks([])
    plt.show()
pred_plot_images_lables(X_test,y_test,0,5)

9、模型保存与加载

import numpy as np

tf.keras.models.save_model(model,"model.keras")
loaded_model = tf.keras.models.load_model("model.keras")
# assert np.allclose(model.predict(X_test[:5]), loaded_model.predict(X_test[:5]))
print(np.argmax(model.predict(X_test[:5]),axis=1))
print(np.argmax(loaded_model.predict(X_test[:5]),axis=1))