tensorflow(十六)：张量的限幅（tf.clip_by_value()、 tf.clip_by_norm()、 tf.clip_by_global_norm()）

一、tf.clip_by_value()限幅

 

 

 

 二、tf.clip_by_norm()根据范数裁剪

• 等比例缩放，只改变模值大小，不改变方向！

 

 三、tf.clip_by_global_norm()梯度整体同比例缩放

 

  梯度爆炸：就是梯度值太大了，每一次前进的步长太长了，导致不停的来回震荡！
梯度消失：就是梯度的值太小了，每一次前进基本没什么变化，导致loss的值长时间不动。

四、实战

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import datasets, layers, optimizers
import os

os.environ['TF_CPP_MIN_LOG_LEVEL']='2'
print(tf.__version__)

(x, y), _ = datasets.mnist.load_data()
x = tf.convert_to_tensor(x, dtype=tf.float32) / 50.   #我们限制0~5范围内，比较大这时候有可能出现gradient exploding
y = tf.convert_to_tensor(y)
y = tf.one_hot(y, depth=10)
print("x: ", x.shape, "y: ", y.shape)

train_db = tf.data.Dataset.from_tensor_slices((x,y)).batch(128).repeat(30)

iteration = iter(train_db)
x,y =next(iteration)
print("sample: ", x.shape, y.shape)


def main():

    # 784 => 512
    w1, b1 = tf.Variable(tf.random.truncated_normal([784, 512], stddev=0.1)), tf.Variable(tf.zeros([512]))
    # 512 => 256
    w2, b2 = tf.Variable(tf.random.truncated_normal([512, 256], stddev=0.1)), tf.Variable(tf.zeros([256]))
    # 256 => 10
    w3, b3 = tf.Variable(tf.random.truncated_normal([256, 10], stddev=0.1)), tf.Variable(tf.zeros([10]))

    optimizer = optimizers.SGD(lr=0.01)

    for step, (x,y) in enumerate(train_db):
        # [batch_size, 28, 28]=> [batch_size, 784]
        x = tf.reshape(x, (-1, 784))

        with tf.GradientTape() as tape:
            # layer1
            h1 = x@w1 + b1
            h1 = tf.nn.relu(h1)
            # layer2
            h2=tf.nn.relu(h1@w2 + b2)
            # layer3
            out = tf.nn.relu(h2@w3 + b3)

            #computer loss
            # [batch_size, 10] - [batch_size, 10]
            loss = tf.square(out-y)
            # [batch_size, 10] => [batch_size]
            loss = tf.reduce_sum(loss, axis=1)
            # [batch_size] => scalar
            loss = tf.reduce_mean(loss)

        #compute gradient
        grads = tape.gradient(loss, [w1, b1, w2, b2, w3, b3])
        print('==before global_norm==')
        for g in grads:
            print(tf.norm(g))  #梯度的模，没有考虑方向

        grads, _ =tf.clip_by_global_norm(grads, 15) #梯度的值等比例缩放，都在15以内。

        # print('==after global_norm==')
        # for g in grads:
        #     print(tf.norm(g))  #梯度的模，没有考虑方向

        # update w' = w - lr*grad
        optimizer.apply_gradients(zip(grads, [w1, b1, w2, b2, w3, b3]))
        
        if step % 100 == 0:
            print(step, 'loss: ', float(loss))

if __name__ == '__main__':
    main()

 如果不加那一行，会出现