TF构造八股文

• Import: 导入相关的包
• train and test: 处理好训练与测试数据集,包括乱序等操作
• models.Sequential: 搭建网络模型
• model.compile: 设置优化算法,配置训练方法
• model.fit: 设置训练时的学习率,迭代次数,batch_size,对模型进行训练;
• model.summary：打印出网络结构和参数统计

简单实现

# 1. import
import tensorflow as tf

# 2. train and test
w = tf.Variable(tf.constant(5, dtype=tf.float32))

# 3. model.sequential

# 4. model.compile and 5. model.fit
lr = 0.2
epochs = 40

for epoch in range(epochs):
    with tf.GradientTape() as tape:
        loss = tf.square(w + 1)        # 定义loss函数
    grads = tape.gradient(loss, w)     # 求loss对w的梯度

    w.assign_sub(lr * grads)
    print("After %s epoch, w is %f, loss is %f" % (epoch, w.numpy(), loss))     # 6. model.summary()
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19