因为论文问题，陆续学了huhuhuang老师的多个课程，课程质量都很高。TF2是完全自己要用，课程内容完全可以满足自己所需。教程＋挑战结合，深入浅出。

huhuhang老师的课程质量都还不错～近期我要把自己代码里 1.x 的接口都升级成 2.x 的

hard

难

牛逼

学习了TensorFlow2.0的新特性，不需要1.X版本的API那样麻烦，只需要像python一样编程就好了。

感觉入门还行吧

学到了迁移学习 写的很好

有所收获，很好！ 最终预测结果86%

val_accuracy: 0.9364

对one-hot和label encoding有了直观的理解 (label encoder实验出来的效果不好)

不错，复习了

很好

不错

开心心

!wget -nc "http://labfile.oss.aliyuncs.com/courses/1211/fashion-mnist.zip" !mkdir -p "/root/.keras/datasets/fashion-mnist/" !unzip -o "fashion-mnist.zip" -d "/root/.keras/datasets/fashion-mnist" import tensorflow as tf import numpy as np

(X_train, y_train), (X_test, y_test) = tf.keras.datasets.fashion_mnist.load_data()

对特征进行归一化处理

X_train = X_train / 255 X_test = X_test / 255

对标签进行独热编码

y_train = np.eye(10)[y_train] y_test = np.eye(10)[y_test]

X_train.shape, X_test.shape, y_train.shape, y_test.shape """挑战：按要求构建 LeNet-5 顺序模型""" model_vgg = tf.keras.Sequential([ tf.keras.layers.Conv2D(6, (5, 5), input_shape=(32, 32, 1), activation='relu'), tf.keras.layers.AveragePooling2D(pool_size=(2, 2), strides=(2, 2)), tf.keras.layers.Conv2D(16, (5, 5), activation='relu'), tf.keras.layers.AveragePooling2D(pool_size=(2, 2), strides=(2, 2)), tf.keras.layers.Flatten(), tf.keras.layers.Dense(120, activation='relu'), tf.keras.layers.Dense(84, activation='relu'), tf.keras.layers.Dense(10, activation='softmax') ])

model_vgg.summary() """挑战：完成模型训练过程""" X_train = tf.reshape(X_train, [-1, 28, 28, 1]) X_test = tf.reshape(X_test, [-1, 28, 28, 1]) X_train.shape, X_test.shape X_train_lite = tf.image.resize(X_train, (32, 32)) X_test_lite = tf.image.resize(X_test, (32, 32)) X_train_lite.shape,X_test_lite.shape model_vgg.compile(optimizer=tf.optimizers.Adam(), loss='categorical_crossentropy', metrics=['accuracy'])

训练模型

history = model_vgg.fit(X_train_lite, y_train, epochs=10, batch_size=32, validation_data=(X_test_lite, y_test))