TensorFlow高阶操作之张量限幅
2020-02-04 本文已影响0人
酵母小木
【碎碎念】今天本来要看8节网课的，现在都下午4点了，才看了一节网课，其他时间都拿来想你了！fighting!!!
1. 【clip_by_value】：根据值进行限幅
-
tf.maximum(tensor, thread):下限幅，值必须大于阈值if data < thread， data = thread -
tf.minimum(tensor, thread):上限幅，值必须小于阈值if data > thread， data = thread -
tf.clip_by_value(tensor, down_thread, up_thread):值必须在阈值之间
In [49]: a = tf.range(10)
Out[50]: <tf.Tensor: id=81, shape=(10,), dtype=int32, numpy=array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])>
In [52]: tf.maximum(a, 2)
Out[52]: <tf.Tensor: id=83, shape=(10,), dtype=int32, numpy=array([2, 2, 2, 3, 4, 5, 6, 7, 8, 9])>
In [53]: tf.minimum(a, 8)
Out[53]: <tf.Tensor: id=85, shape=(10,), dtype=int32, numpy=array([0, 1, 2, 3, 4, 5, 6, 7, 8, 8])>
In [54]: tf.clip_by_value(a, 2, 8)
Out[54]: <tf.Tensor: id=89, shape=(10,), dtype=int32, numpy=array([2, 2, 2, 3, 4, 5, 6, 7, 8, 8])>
使用限幅函数实现RELU函数功能
In [55]: a = tf.range(10)
In [57]: a = a - 5
Out[58]: <tf.Tensor: id=95, shape=(10,), dtype=int32, numpy=array([-5, -4, -3, -2, -1, 0, 1, 2, 3, 4])>
//使用relu的可读性更强一些
In [59]: tf.nn.relu(a)
Out[59]: <tf.Tensor: id=96, shape=(10,), dtype=int32, numpy=array([0, 0, 0, 0, 0, 0, 1, 2, 3, 4])>
In [60]: tf.maximum(a, 0)
Out[60]: <tf.Tensor: id=98, shape=(10,), dtype=int32, numpy=array([0, 0, 0, 0, 0, 0, 1, 2, 3, 4])>
2. 【clip_by_norm】：根据范数进行数据裁剪，实现等比例放缩，不改变方向
In [61]: a = tf.random.normal([2, 2], mean=10)
Out[62]: <tf.Tensor: id=104, shape=(2, 2), dtype=float32, numpy=
array([[ 9.855437, 8.334126],
[ 9.349087, 10.561935]], dtype=float32)>
In [63]: tf.norm(a)
Out[63]: <tf.Tensor: id=109, shape=(), dtype=float32, numpy=19.11929>
In [64]: aa = tf.clip_by_norm(a, 15)
Out[66]: <tf.Tensor: id=126, shape=(2, 2), dtype=float32, numpy=
array([[7.7320633, 6.538522 ],
[7.334807 , 8.2863455]], dtype=float32)>
In [65]: tf.norm(aa)
Out[65]: <tf.Tensor: id=131, shape=(), dtype=float32, numpy=14.999999>
3.【Gradient clipping】：梯度下降
实现梯度下降存在两大障碍：梯度爆炸和梯度弥散。梯度爆炸是由于梯度太大，导致前进的步长太快；梯度弥散是由于梯度太小导致的。而【Gradient clipping】实现了整体梯度等比例缩放，但是梯度方向不变，在一定程度上抑制了梯度爆炸和梯度弥散
# 张量限幅实战
import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import datasets, layers, optimizers
print(tf.__version__)
(x, y), _ = datasets.mnist.load_data()
x = tf.convert_to_tensor(x, dtype=tf.float32) / 50.
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)
sample = next(iter(train_db))
print('sample:', sample[0].shape, sample[1].shape)
# print(x[0], y[0])
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):
# [b, 28, 28] => [b, 784]
x = tf.reshape(x, [-1, 784])
with tf.GradientTape() as tape:
# layer1.
h1 = x @ w1 + b1
h1 = tf.nn.relu(h1)
# layer2
h2 = h1 @ w2 + b2
h2 = tf.nn.relu(h2)
# output
out = h2 @ w3 + b3
# compute loss
# [b, 10] - [b, 10]
loss = tf.square(y - out)
# [b, 10] => [b]
loss = tf.reduce_mean(loss, axis=1)
# [b] => scalar
loss = tf.reduce_mean(loss)
# compute gradient
grads = tape.gradient(loss, [w1, b1, w2, b2, w3, b3])
# print('==before==')
# for g in grads:
# print(tf.norm(g))
grads, _ = tf.clip_by_global_norm(grads, 15)
# print('==after==')
# for g in grads:
# print(tf.norm(g))
# 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()
【注】在PyCharm中运行上述代码的时候，记得将ipython关闭，不然会出现GPU显存不足的情况
