TensorFLow2基础
《HCIA-AI 华为认证人工智能工程师在线课程》
课程链接
本节课程分为8个单元，主要讲解了TensorFlow 2的基本语法.下面的8个单元内容分别包括：
1、常量与变量的创建方式
2、张量的切片与索引
3、张量的维度变化
4、张量的算术运算
5、张量的拼接与分割
6、张量排序
7、Eager Execution
8、AutoGraph
学完本节内容后，你能了解到TensorFlow 2的基本语法。
代码：
# -*- coding: utf-8 -*-
"""
Created on Mon Mar 15 15:00:00 2021
HCIA-AI 华为认证人工智能工程师在线课程《TensorFLow2基础》
@author: Administrator
"""
import tensorflow as tf
import numpy as np
const_a = tf.constant([[1,2,3,4]],shape=[2,2],dtype=tf.float32)
print(const_a)
zeros_b = tf.zeros(shape=[2,3],dtype=tf.int32)
print(zeros_b)
zeros_like_c = tf.zeros_like(const_a)
print(zeros_like_c)
fill_d = tf.fill([3,3],8)
print(fill_d.numpy())
random_e = tf.random.normal([5,5],mean=0,stddev=1.0)
print(random_e.numpy())
list_f = [1,2,3,4,5,6]
type(list_f)
tensor_f = tf.convert_to_tensor(list_f,dtype=tf.float32)
print(tensor_f)
var_l = tf.Variable(tf.ones([2,3]))
print(var_l)
var_value_l = [[1,2,3],[4,5,6]]
var_l.assign(var_value_l)
print(var_l)
var_l.assign_add(tf.ones([2,3]))
var_l.assign_add(tf.ones([2,3]))
print(var_l)
tensor_h = tf.random.normal([4,100,100,3])
#print(tensor_h)
print(tensor_h[0][19][39][1])
indices = [0,1,3]
#print(tf.gather(tensor_h,axis=0,indices=indices,batch_dims=1))
indices_nd = [[0,1,1,0],[1,2,2,0]]
#print(tf.gather_nd(tensor_h,indices=indices_nd))
#print(tensor_h[0,:,:,:])
const_d_l = tf.constant([[1,2,3,4]],shape=[2,2],dtype=tf.float32)
print(const_d_l.shape)
print(const_d_l.get_shape())
print(tf.shape(const_d_l))
reshape_l = tf.constant([[1,2,3],[4,5,6]])
print(reshape_l)
print(tf.reshape(reshape_l,(3,2)))
expand_sample_l = tf.random.normal([100,100,3])
print(expand_sample_l.shape)
print(tf.expand_dims(expand_sample_l,axis=0).shape)
print(tf.expand_dims(expand_sample_l,axis=1).shape)
print(tf.expand_dims(expand_sample_l,axis=-1).shape)
squeeze_sample_l = tf.random.normal([1,100,100,3])
print(squeeze_sample_l.shape)
print(tf.squeeze(squeeze_sample_l,axis=0).shape)
trans_sample_1 = tf.constant([1,2,3,4,5,6],shape=[2,3])
print(trans_sample_1.shape)
print(tf.transpose(trans_sample_1).shape)
trans_sample_2 = tf.random.normal([4,100,200,3])
print(trans_sample_2.shape)
print(tf.transpose(trans_sample_2,[0,2,1,3]).shape)
broadcast_sample_1 = tf.constant([1,2,3,4,5,6])
print(broadcast_sample_1.numpy())
print(tf.broadcast_to(broadcast_sample_1,shape=[4,6]).numpy())
a = tf.constant([[0,0,0],[10,10,10],[20,20,20],[30,30,30]])
b = tf.constant([1,2,3])
print(a+b)
a = tf.constant([[3,5],[4,8]])
b = tf.constant([[1,6],[2,9]])
print(tf.add(a,b))
print(tf.matmul(a, b))
argmax_sample_1 = tf.constant([[1,3],[2,5],[7,5]])
print("\r\n输入张量：\r\n",argmax_sample_1.numpy(),"\r\n大小：",argmax_sample_1.shape,"\r\n")
print(tf.argmax(argmax_sample_1,axis=0).numpy())
print(tf.argmax(argmax_sample_1,axis=1).numpy())
reduce_sample_1 = tf.constant([1,2,3,4,5,6],shape=[2,3])
print("\r\n原始数据：\r\n",reduce_sample_1.numpy(),"\r\n大小：",reduce_sample_1.shape,"\r\n")
print(tf.reduce_sum(reduce_sample_1,axis=None).numpy())
print(tf.reduce_sum(reduce_sample_1,axis=0).numpy())
print(tf.reduce_sum(reduce_sample_1,axis=1).numpy())
concat_sample_1 = tf.random.normal([4,100,100,3])
concat_sample_2 = tf.random.normal([40,100,100,3])
print("\r\n原始数据的尺寸分别为：\r\n",concat_sample_1.shape,"\r\n",concat_sample_1.shape)
print("拼接后数据的尺寸为：\r\n",tf.concat([concat_sample_1,concat_sample_2],axis=0).shape)
stack_sample_1 = tf.random.normal([100,100,3])
stack_sample_2 = tf.random.normal([100,100,3])
print("\r\n原始数据的尺寸分别为：\r\n",stack_sample_1.shape,"\r\n",stack_sample_2.shape)
stacked_sample_1 = tf.stack([stack_sample_1,stack_sample_2],axis=0)
print("拼接后数据的尺寸为：\r\n",stacked_sample_1.shape)
#print("分割后数据的尺寸为：\r\n",tf.unstack(stacked_sample_1,axis=0))
split_sample_1 = tf.random.normal([10,10,10,3])
print("\r\nsplit_sample_1原始数据的尺寸为：\r\n",split_sample_1.shape)
splited_sample_1 = tf.split(split_sample_1,num_or_size_splits=5,axis=0)
print("当num_or_size_splits=5，分割后数据的尺寸为：\r\n",np.shape(splited_sample_1))
splited_sample_2 = tf.split(split_sample_1,num_or_size_splits=[3,5,2],axis=0)
print("当num_or_size_splits=[[3,5,2]，分割后数据的尺寸为：\r\n",
np.shape(splited_sample_2[0]),
np.shape(splited_sample_2[1]),
np.shape(splited_sample_2[2]),)
sort_sample_1 =tf.random.shuffle(tf.range(10))
print("\r\n输入张量:",sort_sample_1.numpy())
sorted_sample_1 = tf.sort(sort_sample_1,direction="ASCENDING")
print("升序排列后的张量:",sorted_sample_1.numpy())
sorted_sample_2 = tf.argsort(sort_sample_1,direction="ASCENDING")
print("升序排列后，元素的索引:",sorted_sample_2.numpy())
values,index = tf.nn.top_k(sort_sample_1,5)
print("升序排列后的前5个数值：",values.numpy())
print("升序排列后的前5个数值的索引：",index.numpy())
x = tf.ones((2,2),dtype=tf.dtypes.float32)
y = tf.constant([[1,2],[3,4]],dtype=tf.dtypes.float32)
z = tf.matmul(x,y)
print("\r\nTensorFlow2.0不用Session可以直接运算:",z)
thre_1 = tf.random.uniform([],0,1)
x = tf.reshape(tf.range(0,4),[2,2])
print("\r\nTensorFlow2.0得益eager得到Tensor的numpy值thre1:\r\n",thre_1)
if thre_1.numpy() > 5:
y = tf.matmul(x,y)
else:
y = tf.add(x,x)
print("从而可以用条件判断计算出y:\r\n",y)
@tf.function
def simple_nn_layer(w,x,b):
print("\r\nAutoGraph静态图b:无法查看数值",b)
return tf.nn.relu(tf.matmul(w,x)+b)
w =tf.random.uniform((3,3))
x =tf.random.uniform((3,3))
b =tf.constant(0.5,dtype='float32')
simple_nn_layer(w,x,b)
import timeit
CNN_cell = tf.keras.layers.Conv2D(filters=100,kernel_size=2,strides=(1,1))
@tf.function
def CNN_fn(image):
return CNN_cell(image)
image = tf.zeros([100,200,200,3])
CNN_cell(image)
CNN_fn(image)
print("\r\neager execution模式下做一层CNN卷积层运算的实际：\r\n",timeit.timeit(lambda:CNN_cell(image),number=10))
print("graph模式下做一层CNN卷积层运算的实际：\r\n",timeit.timeit(lambda:CNN_fn(image),number=10))
运行输出：
tf.Tensor(
[[1. 2.]
[3. 4.]], shape=(2, 2), dtype=float32)
tf.Tensor(
[[0 0 0]
[0 0 0]], shape=(2, 3), dtype=int32)
tf.Tensor(
[[0. 0.]
[0. 0.]], shape=(2, 2), dtype=float32)
[[8 8 8]
[8 8 8]
[8 8 8]]
[[-0.80290383 -1.5259775 -0.6541044 0.56509346 0.12168431]
[-0.4033856 -0.52415013 1.8449188 0.420102 -2.3412702 ]
[ 0.8241716 0.6353277 1.2247677 1.3111122 0.32647747]
[-2.1685581 1.6500843 0.7036131 -0.79745334 -0.38055837]
[ 0.15103035 0.84089184 -0.546232 1.2970096 0.3392236 ]]
tf.Tensor([1. 2. 3. 4. 5. 6.], shape=(6,), dtype=float32)
<tf.Variable 'Variable:0' shape=(2, 3) dtype=float32, numpy=
array([[1., 1., 1.],
[1., 1., 1.]], dtype=float32)>
<tf.Variable 'Variable:0' shape=(2, 3) dtype=float32, numpy=
array([[1., 2., 3.],
[4., 5., 6.]], dtype=float32)>
<tf.Variable 'Variable:0' shape=(2, 3) dtype=float32, numpy=
array([[3., 4., 5.],
[6., 7., 8.]], dtype=float32)>
tf.Tensor(-0.16037415, shape=(), dtype=float32)
(2, 2)
(2, 2)
tf.Tensor([2 2], shape=(2,), dtype=int32)
tf.Tensor(
[[1 2 3]
[4 5 6]], shape=(2, 3), dtype=int32)
tf.Tensor(
[[1 2]
[3 4]
[5 6]], shape=(3, 2), dtype=int32)
(100, 100, 3)
(1, 100, 100, 3)
(100, 1, 100, 3)
(100, 100, 3, 1)
(1, 100, 100, 3)
(100, 100, 3)
(2, 3)
(3, 2)
(4, 100, 200, 3)
(4, 200, 100, 3)
[1 2 3 4 5 6]
[[1 2 3 4 5 6]
[1 2 3 4 5 6]
[1 2 3 4 5 6]
[1 2 3 4 5 6]]
tf.Tensor(
[[ 1 2 3]
[11 12 13]
[21 22 23]
[31 32 33]], shape=(4, 3), dtype=int32)
tf.Tensor(
[[ 4 11]
[ 6 17]], shape=(2, 2), dtype=int32)
tf.Tensor(
[[13 63]
[20 96]], shape=(2, 2), dtype=int32)
输入张量：
[[1 3]
[2 5]
[7 5]]
大小： (3, 2)
[2 1]
[1 1 0]
原始数据：
[[1 2 3]
[4 5 6]]
大小： (2, 3)
21
[5 7 9]
[ 6 15]
原始数据的尺寸分别为：
(4, 100, 100, 3)
(4, 100, 100, 3)
拼接后数据的尺寸为：
(44, 100, 100, 3)
原始数据的尺寸分别为：
(100, 100, 3)
(100, 100, 3)
拼接后数据的尺寸为：
(2, 100, 100, 3)
split_sample_1原始数据的尺寸为：
(10, 10, 10, 3)
当num_or_size_splits=5，分割后数据的尺寸为：
(5, 2, 10, 10, 3)
当num_or_size_splits=[[3,5,2]，分割后数据的尺寸为：
(3, 10, 10, 3) (5, 10, 10, 3) (2, 10, 10, 3)
输入张量: [8 9 4 5 0 3 2 6 7 1]
升序排列后的张量: [0 1 2 3 4 5 6 7 8 9]
升序排列后，元素的索引: [4 9 6 5 2 3 7 8 0 1]
升序排列后的前5个数值： [9 8 7 6 5]
升序排列后的前5个数值的索引： [1 0 8 7 3]
TensorFlow2.0不用Session可以直接运算: tf.Tensor(
[[4. 6.]
[4. 6.]], shape=(2, 2), dtype=float32)
TensorFlow2.0得益eager得到Tensor的numpy值thre1:
tf.Tensor(0.52014256, shape=(), dtype=float32)
从而可以用条件判断计算出y:
tf.Tensor(
[[0 2]
[4 6]], shape=(2, 2), dtype=int32)
AutoGraph静态图b:无法查看数值 Tensor("b:0", shape=(), dtype=float32)
eager execution模式下做一层CNN卷积层运算的实际：
13.16959130000032
graph模式下做一层CNN卷积层运算的实际：
7.5449627000016335
扫一扫
927
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
