tensorflow知识点

最新推荐文章于 2023-01-03 18:46:12 发布

薛定谔的智能

最新推荐文章于 2023-01-03 18:46:12 发布

阅读量682

点赞数

分类专栏： tensorflow 文章标签： tensorflow

本文链接：https://blog.csdn.net/fanzonghao/article/details/81277087

版权

tensorflow 专栏收录该内容

17 篇文章 0 订阅

订阅专栏

一.bazel编译tensorflow注意版本号:

在/tensorflow/tensorflow/configure.py 查看bazel版本号

https://github.com/tensorflow/tensorflow

https://github.com/bazelbuild/bazel/releases?after=0.26.1

https://tensorflow.google.cn/

二，基础知识点

1.打印出与训练变量相关的信息

"""
slim.model_analyzer.analyze_vars打印出与训练变量相关的信息
"""
import tensorflow as tf
import tensorflow.contrib.slim as slim
x1=tf.Variable(tf.constant(1,shape=[1],dtype=tf.float32,name='x1'))
x2=tf.Variable(tf.random_normal(shape=[2,1],dtype=tf.float32,name='x2'))
y=tf.trainable_variables()
for i in y:
    print(6666)
    print(i)
slim.model_analyzer.analyze_vars(y,print_info=True)
print(88888888)

2. tf.concat拼接

"""
tf.concat拼接
"""
import tensorflow as tf
t1=tf.constant([[1,2,3],[4,5,6]])
t2=tf.constant([[7,8,9],[10,11,12]])
t3=tf.concat([t1,t2],0)
t4=tf.concat([t1,t2],1)
print('t1={}'.format(t1))
print('t2={}'.format(t2))
print('t3={}'.format(t3))
print('t4={}'.format(t4))

"""
tf.concat拼接
"""
import tensorflow as tf
t1=tf.constant([[[1,2,3],[4,5,6]]])
t2=tf.constant([[[7,8,9],[10,11,12]]])
t3=tf.concat([t1,t2],0)
t4=tf.concat([t1,t2],1)
t5=tf.concat([t1,t2],-1)
print('t1={}'.format(t1))
print('t2={}'.format(t2))
print('t3={}'.format(t3))
print('t4={}'.format(t4))
print('t5={}'.format(t5))

3.tensorboard调用

graphs:计算图显示

tf.summary.FileWritter(path,sess.graph)

import tensorflow as tf
input1=tf.constant([1.0,2.0,3.0],name='input_1')
input2=tf.constant([2.0,5.0,8.0],name='input_2')
output=tf.add(input1,input2,name='add')
with tf.Session() as sess:
    writer=tf.summary.FileWriter('./data/', sess.graph)
    sess.run(tf.global_variables_initializer())
    print(sess.run(output))
writer.close()

GRPHS显示了网络结构信息。

4. 池化和卷积的‘SAME’和‘VALID’

一维：

"VALID" = without padding:

   inputs:         1  2  3  4  5  6  7  8  9  10 11 (12 13)
                  |________________|                dropped
                                 |_________________|

valid采用丢弃
"SAME" = with zero padding:

               pad|                                      |pad
   inputs:      0 |1  2  3  4  5  6  7  8  9  10 11 12 13|0  0
               |________________|
                              |_________________|
                                             |________________|

SAME为向上取整。ceil

卷积过程：

For the SAME padding, the output height and width are computed as:

out_height = ceil(float(in_height) / float(strides[1]))

out_width = ceil(float(in_width) / float(strides[2]))

And

For the VALID padding, the output height and width are computed as:

out_height = ceil(float(in_height - filter_height + 1) / float(strides[1]))

out_width = ceil(float(in_width - filter_width + 1) / float(strides[2]))

5.tf.variable_scope和tf.name_scope

tf.variable_scope可以让变量有相同的命名，包括tf.get_variable得到的变量，还有tf.Variable的变量

tf.name_scope可以让变量有相同的命名，但只是限于tf.Variable的变量

import tensorflow as tf

with tf.variable_scope('V1'):
    a1 = tf.get_variable(name='a1', shape=[1], initializer=tf.constant_initializer(1))
    a2 = tf.Variable(tf.random_normal(shape=[2, 3], mean=0, stddev=1), name='a2')
with tf.variable_scope('V2'):
    a3 = tf.get_variable(name='a1', shape=[1], initializer=tf.constant_initializer(1))
    a4 = tf.Variable(tf.random_normal(shape=[2, 3], mean=0, stddev=1), name='a2')

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    print(a1.name)
    print(a2.name)
    print(a3.name)
    print(a4.name)

import tensorflow as tf

with tf.name_scope('V1'):
    a1 = tf.get_variable(name='a1', shape=[1], initializer=tf.constant_initializer(1))
    a2 = tf.Variable(tf.random_normal(shape=[2, 3], mean=0, stddev=1), name='a2')
with tf.name_scope('V2'):
    a3 = tf.get_variable(name='a1', shape=[1], initializer=tf.constant_initializer(1))
    a4 = tf.Variable(tf.random_normal(shape=[2, 3], mean=0, stddev=1), name='a2')

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    print(a1.name)
    print(a2.name)
    print(a3.name)
    print(a4.name)

import tensorflow as tf

with tf.name_scope('V1'):
    # a1 = tf.get_variable(name='a1', shape=[1], initializer=tf.constant_initializer(1))
    a2 = tf.Variable(tf.random_normal(shape=[2, 3], mean=0, stddev=1), name='a2')
with tf.name_scope('V2'):
    # a3 = tf.get_variable(name='a1', shape=[1], initializer=tf.constant_initializer(1))
    a4 = tf.Variable(tf.random_normal(shape=[2, 3], mean=0, stddev=1), name='a2')

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    # print(a1.name)
    print(a2.name)
    # print(a3.name)
    print(a4.name)

6. tf.pad，二维情况

import tensorflow as tf
t = tf.constant([[1, 2, 3], [4, 5, 6]])
#向上两行 向下一行 向左三行 向右两行
paddings = tf.constant([[2, 1,], [3, 2]])
#CONSTAN代表补零
t_pad_constant=tf.pad(t,paddings,mode='CONSTANT')

with tf.Session() as sess:
    print(sess.run(t_pad_constant))

import tensorflow as tf
t = tf.constant([[1, 2, 3], [4, 5, 6]])
#向上两行 向下一行 向左三行 向右两行
paddings = tf.constant([[1, 1,], [2, 2]])
#CONSTAN代表补零
t_pad_constant=tf.pad(t,paddings,mode='CONSTANT')
#REFLECT代表镜像对称不包含轴
t_pad_reflect=tf.pad(t,paddings,mode='REFLECT')

with tf.Session() as sess:
    print(sess.run(t_pad_constant))
    print(sess.run(t_pad_reflect))

import tensorflow as tf
t = tf.constant([[1, 2, 3], [4, 5, 6]])
#向上两行 向下一行 向左三行 向右两行
paddings = tf.constant([[1, 1,], [2, 2]])
#CONSTAN代表补零
t_pad_constant=tf.pad(t,paddings,mode='CONSTANT')
#REFLECT代表镜像对称不包含轴
t_pad_reflect=tf.pad(t,paddings,mode='REFLECT')
#SYMMETRIC代表镜像对称 包含轴
t_pad_symmetric=tf.pad(t,paddings,mode='SYMMETRIC')

with tf.Session() as sess:
    print(sess.run(t_pad_constant))
    print(sess.run(t_pad_reflect))
    print(sess.run(t_pad_symmetric))

三维情况：

import tensorflow as tf
t = tf.constant([[
                   [[1, 2, 3],
                   [4, 5, 6]]
                  ,[[1, 2, 3],
                    [4, 5, 6]]
                  ,[[1, 2, 3],
                    [4, 5, 6]]
                  ]])
#向上两行 向下一行 向左三行 向右两行
paddings = tf.constant([[0,0],[1, 1,], [1, 1],[0,0]])
#CONSTAN代表补零
t_pad_constant=tf.pad(t,paddings,mode='CONSTANT')
t_pad_constant=tf.pad(t,paddings,mode='CONSTANT')
t_pad_constant=tf.pad(t,paddings,mode='CONSTANT')
t_pad_constant=tf.pad(t,paddings,mode='CONSTANT')
#REFLECT代表镜像对称不包含轴
t_pad_reflect=tf.pad(t,paddings,mode='REFLECT')
# #SYMMETRIC代表镜像对称 包含轴
t_pad_symmetric=tf.pad(t,paddings,mode='SYMMETRIC')

with tf.Session() as sess:
    print(sess.run(t))
    print(sess.run(t).shape)
    print(sess.run(t_pad_constant))
    print(sess.run(t_pad_constant).shape)
    print(sess.run(t_pad_reflect))
    print(sess.run(t_pad_reflect).shape)
    print(sess.run(t_pad_symmetric))
    print(sess.run(t_pad_symmetric).shape)

打印结果：

7.学习率衰减

tf.train.exponential_decay的返回值
decayed_learning_rate = learning_rate * decay_rate ^ (global_step / decay_steps)

其中，decayed_learning_rate为每一轮优化时使用的学习率；learning_rate为事先设定的初始学习率； decay_rate为衰减系数；decay_steps为衰减速度。

下面代码：返回值=0.1×0.96^ (global_step / 100)，其中staircase=True代表取整，故打印的结果是蓝线，staircase=Fasle代表取小数，故打印的结果是红线

import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt

learning_rate = 0.1
decay_rate = 0.96
global_steps = 1000
decay_steps = 100

global_ = tf.Variable(tf.constant(0))
c = tf.train.exponential_decay(learning_rate, global_, decay_steps, decay_rate, staircase=True)
d = tf.train.exponential_decay(learning_rate, global_, decay_steps, decay_rate, staircase=False)

T_C = []
F_D = []

with tf.Session() as sess:
    for i in range(global_steps):
        T_c = sess.run(c, feed_dict={global_: i})
        T_C.append(T_c)
        F_d = sess.run(d, feed_dict={global_: i})
        F_D.append(F_d)

plt.figure()
plt.plot(range(global_steps), F_D, 'r-')
plt.plot(range(global_steps), T_C, 'b-')
plt.show()

打印结果：

8. tf.assign

#tf.assign实现值的转换
import tensorflow as tf
a = tf.Variable(tf.constant(0.0), dtype=tf.float32)
init=tf.global_variables_initializer()
with tf.Session() as sess:
    sess.run(init)
    print('a={}'.format(sess.run(a)))
    print('a={}'.format(sess.run(tf.assign(a, 1))))

9. tf.nn.moments

#tf求均值和方差
import tensorflow as tf
W = tf.constant([[1.,2.,3.],[4.,5.,6.]])
mean,var = tf.nn.moments(W, axes = [0])
init=tf.global_variables_initializer()
with tf.Session() as sess:
    sess.run(init)
    Mean = sess.run(mean)
    print(Mean)
    Var = sess.run(var)
    print(Var)

#tf求均值和方差
import tensorflow as tf
W = tf.constant([[1.,2.,3.],[4.,5.,6.]])
mean,var = tf.nn.moments(W, axes = [1])
init=tf.global_variables_initializer()
with tf.Session() as sess:
    sess.run(init)
    Mean = sess.run(mean)
    print(Mean)
    Var = sess.run(var)
    print(Var)

#tf求均值和方差
import tensorflow as tf
W = tf.constant([[1.,2.,3.],[4.,5.,6.]])
mean,var = tf.nn.moments(W, axes = [0,1])
init=tf.global_variables_initializer()
with tf.Session() as sess:
    sess.run(init)
    Mean = sess.run(mean)
    print(Mean)
    Var = sess.run(var)
    print(Var)

# tf求均值和方差
import tensorflow as tf
W = tf.constant([[[[1., 2., 3.], [4., 5., 6.]]]])
mean, var = tf.nn.moments(W, axes=[0,1,2])
y=W-mean
init = tf.global_variables_initializer()
with tf.Session() as sess:
    sess.run(init)
    Mean = sess.run(mean)
    print(Mean)
    print(sess.run(y))

求每一个feature map的均值，把每个feature map当成一个神经元，故轴方向取[0,1,2]，而对于二位矩阵，需要计算神经元的均值和方差故轴取的是0。

10. tensorflow队列操作

队列是一种先入先出的线性数据结构，队尾增加数据，队首输出和删除数据。

队列的创建：

import tensorflow as tf
with tf.Session() as sess:
    #队列先入先出  存放三个数据
    q=tf.FIFOQueue(3,'float')
    #将上文创建的FIFOQueue函数 填充3个数据 做预备工作
    init=q.enqueue_many(([0.1,0.2,0.3],))

    #真正填充三个数据
    sess.run(init)
    #获取长度
    quelen=sess.run(q.size())
    for i in range(quelen):
        #将元素从队列移出
        print(sess.run(q.dequeue()))

注意：每次做元素填充队列和弹出都需要sess run

import tensorflow as tf
with tf.Session() as sess:
    #队列先入先出  存放三个数据
    q=tf.FIFOQueue(3,'float')
    #将上文创建的FIFOQueue函数 填充3个数据 做预备工作
    init=q.enqueue_many(([0.1,0.2,0.3],))

    #弹出一个元素  准备工作 后面需要 sess run
    init2=q.dequeue()
    #将元素放入队列 准备工作 后面需要 sess run
    init3=q.enqueue(1.)

    #真正填充三个数据
    sess.run(init)
    sess.run(init2)
    sess.run(init3)
    #获取长度
    quelen=sess.run(q.size())
    for i in range(quelen):
        #将元素从队列移出
        print(sess.run(q.dequeue()))

上述的程序队列的操作主要是在session中进行，优点不易阻塞，好找bug，缺点，效率低。下面用队列管理器QueueRunner解决异步操作问题，创建一系列线程在主线程内操作，数据读取与操作（训练模型）是同步的，提升效率。

import tensorflow as tf
with tf.Session() as sess:
    # 队列先入先出  存放三个数据
    q=tf.FIFOQueue(1000, 'float')
    counter=tf.Variable(0.0)
    #counter=counter+1.0
    add_op=tf.assign_add(counter,tf.constant(1.0))
    # 将元素放入队列 准备工作 后面需要 sess run
    enqueueData_op=q.enqueue(counter)
    qr=tf.train.QueueRunner(q,enqueue_ops=[add_op,enqueueData_op]*2)
    sess.run(tf.global_variables_initializer())
    qr.create_threads(sess,start=True)

    for i in range(10):
        print(sess.run(q.dequeue()))

首先正常执行，最后队列管理器QueueRunner报错。原因是多线程虽然方便了在一个session下共同工作，并行地相互执行，但是这种同步会造成某个线程想要关闭session时，session被强行关闭而未完成的线程也被强行关闭。

故为了解决多线程的同步和处理问题，提供了Coordinator和QueueRunner函数来对线程进行控制与协调。

import tensorflow as tf
with tf.Session() as sess:
    # 队列先入先出  存放三个数据
    q=tf.FIFOQueue(1000, 'float')
    counter=tf.Variable(0.0)
    #counter=counter+1.0
    add_op=tf.assign_add(counter,tf.constant(1.0))
    # 将元素放入队列 准备工作 后面需要 sess run
    enqueueData_op=q.enqueue(counter)
    #调用线程
    qr=tf.train.QueueRunner(q,enqueue_ops=[add_op,enqueueData_op]*2)
    sess.run(tf.global_variables_initializer())
    
    coord=tf.train.Coordinator()
    # 开启线程 启动入队线程  coord线程协调器 启动线程后负责对所有线程接受和处理，故当一个线程结束时，其会对所有线程发出通知，协调完毕。
    enqueue_threads = qr.create_threads(sess, coord=coord,start=True)

    for i in range(10):
        print(sess.run(q.dequeue()))
        
    coord.request_stop()
    coord.join(enqueue_threads)

11. minimize里的 global_steps

x = tf.placeholder(tf.float32, shape=[None, 1], name='x')
y = tf.placeholder(tf.float32, shape=[None, 1], name='y')
w = tf.Variable(tf.constant(0.0))

# global_steps = tf.Variable(0, trainable=False)
global_steps = tf.train.get_or_create_global_step()
# learning_rate = tf.train.exponential_decay(0.1, global_steps, 10, 2, staircase=False)
loss = tf.pow(w*x - y, 2)

train_step = tf.train.GradientDescentOptimizer(0.01).minimize(loss, global_step=global_steps)

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    for i in range(5):
        sess.run(train_step, feed_dict={x: np.linspace(1, 2, 10).reshape([10, 1]),
                                        y: np.linspace(1, 2, 10).reshape([10, 1])})
        # print sess.run(learning_rate)
        print sess.run(global_steps)

可看出，global_steps每次自动加1

12. tf.add_to_collection

tf.add_to_collection(‘list_name’, element)：将元素element添加到列表list_name中

tf.get_collection(‘list_name’)：返回名称为list_name的列表

tf.add_n(list)：将列表元素相加并返回

tf.add_to_collection('losses', tf.constant(1.2))
tf.add_to_collection('losses', tf.constant(5.))
with tf.Session() as sess:
    print(sess.run(tf.get_collection('losses')))
    print(sess.run(tf.add_n(tf.get_collection('losses'))))

13. tf.nn.depthwise_conv2d，其输出的通道数是卷积核的输入与输出channel相乘

https://blog.csdn.net/mao_xiao_feng/article/details/78003476

import tensorflow as tf
img1 = tf.constant(value=[[[[1],[2],[3],[4]],[[1],[2],[3],[4]],[[1],[2],[3],[4]],[[1],[2],[3],[4]]]],dtype=tf.float32)
img2 = tf.constant(value=[[[[1],[1],[1],[1]],[[1],[1],[1],[1]],[[1],[1],[1],[1]],[[1],[1],[1],[1]]]],dtype=tf.float32)
img = tf.concat(values=[img1,img2],axis=3)
print(img1.shape)
print('img.shape={}'.format(img.shape))
filter1 = tf.constant(value=0, shape=[3,3,1,1],dtype=tf.float32)
filter2 = tf.constant(value=1, shape=[3,3,1,1],dtype=tf.float32)
filter3 = tf.constant(value=2, shape=[3,3,1,1],dtype=tf.float32)
filter4 = tf.constant(value=3, shape=[3,3,1,1],dtype=tf.float32)
filter_out1 = tf.concat(values=[filter1,filter2],axis=2)
filter_out2 = tf.concat(values=[filter3,filter4],axis=2)
filter = tf.concat(values=[filter_out1,filter_out2,filter_out2],axis=3)
print(filter_out1.shape)
print(filter_out2.shape)
print('filter.shape={}'.format(filter.shape))

out_img = tf.nn.conv2d(input=img, filter=filter, strides=[1,1,1,1], padding='VALID')
print('out_img.shape={}'.format(out_img.shape))

#
t_img = tf.nn.depthwise_conv2d(input=img, filter=filter, strides=[1,1,1,1], rate=[1,1], padding='VALID')
print('t_img.shape={}'.format(t_img.shape))

input：
指需要做卷积的输入图像，要求是一个4维Tensor，具有[batch, height, width, in_channels]这样的shape，具体含义是[训练时一个batch的图片数量, 图片高度, 图片宽度, 图像通道数]
filter：
相当于CNN中的卷积核，要求是一个4维Tensor，具有[filter_height, filter_width, in_channels, channel_multiplier]这样的shape，具体含义是[卷积核的高度，卷积核的宽度，输入通道数，输出卷积乘子]，同理这里第三维in_channels，就是参数value的第四维
strides：
卷积的滑动步长。
结果返回一个Tensor，shape为[batch, out_height, out_width, in_channels * channel_multiplier]，注意这里输出通道变成了in_channels * channel_multiplier

14. tf.expand_dims扩充维度

import tensorflow as tf

a=tf.Variable(tf.zeros(shape=[2,3,4]))
b=tf.expand_dims(a,axis=-1)
c=tf.expand_dims(a,axis=0)
d=a[:,:,0]
e=tf.expand_dims(a[:,:,0],axis=0)
f=tf.expand_dims(a[:,:,0],axis=-1)
# tf.reset_default_graph()
with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())

    print(a)
    print(b)
    # print(sess.run(b))
    print(c)
    print(d)
    print(e)
    print(f)

15. tf.gradients，求梯度

https://blog.csdn.net/taoyanqi8932/article/details/77602721

import tensorflow as tf

w1 = tf.Variable([[1,2]])#a1 a2
w2 = tf.Variable([[3,4]])

res = tf.matmul(w1, [[2],[1]])#2*a1+a2

grads = tf.gradients(res,[w1])#求梯度

a = tf.constant(0.)
b=2*a
g1 = tf.gradients(a + b, [a, b], stop_gradients=[a, b])
g2 = tf.gradients(b, [a, b])
with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    print(sess.run(res))
    print(sess.run(grads))
    print(sess.run(g1))
    print(sess.run(g2))

16. tf.trainable_variables()与tf.global_variables()

tf.trainable_variables返回的是需要训练的变量列表

tf.global_variables返回的是所有变量的列表

import tensorflow as tf

v = tf.Variable(tf.constant(0.0, shape=[1], dtype=tf.float32), name='v')
v1 = tf.Variable(tf.constant(5, shape=[1], dtype=tf.float32), name='v1')

global_step = tf.Variable(tf.constant(5, shape=[1], dtype=tf.float32), name='global_step', trainable=False)
ema = tf.train.ExponentialMovingAverage(0.99, global_step)

for i in tf.trainable_variables():
    print(i)
print('===============')
for i in tf.global_variables():
    print(i)

17. tf.clip_by_global_norm

https://blog.csdn.net/u013713117/article/details/56281715

18. tf.reduce_sum中的reduction_indices和axis一样的用法，keepdims=True表示维持维度

x = tf.constant([[1, 1, 1], [1, 1, 1]])
a=tf.reduce_sum(x,keepdims=True)  # 6
b=tf.reduce_sum(x, 0,keepdims=True)  # [2, 2, 2]
c=tf.reduce_sum(x, reduction_indices=0)  # [2, 2, 2]
d=tf.reduce_sum(x, 1,keepdims=True)  # [2, 2, 2]
e=tf.reduce_sum(x, reduction_indices=1)  # [2, 2, 2]
with tf.Session() as sess:
    print(sess.run(a))
    print(sess.run(b))
    print(sess.run(c))
    print(sess.run(d))
    print(sess.run(e))

19. tf.app.flags.FLAGS

命令行解析

import tensorflow  as tf

FLAGS = tf.app.flags.FLAGS
tf.app.flags.DEFINE_float(
    'flag_float', 0.01, 'input a float')
tf.app.flags.DEFINE_integer(
    'flag_int', 400, 'input a int')
tf.app.flags.DEFINE_boolean(
    'flag_bool', True, 'input a bool')
tf.app.flags.DEFINE_string(
    'flag_string', 'yes', 'input a string')

print(FLAGS.flag_float)
print(FLAGS.flag_int)
print(FLAGS.flag_bool)
print(FLAGS.flag_string)

20. tensorflow读图片和opencv输出图片

def tf_read_image():
    path='./img_size.jpg'
    with tf.gfile.FastGFile(path, 'r') as f:
        image_data = f.read()

    with tf.Session() as sess:
        image_data = tf.image.decode_jpeg(image_data)
        image = sess.run(image_data)
        """第一种方式"""
        r, g, b = cv2.split(image)
        image = cv2.merge([b, g, r])
        cv2.imwrite('img_size_out.jpg',image)
        """第二种方式"""
        # plt.imshow(image)
        # plt.show()
        # print(image.shape)

由于opencv读图片是b,g,r故在decode成image时，拆分成r,g,b在融合成b,g,r给opencv输出，否则直接输出会造成r通道，却是b通道的值，整张图片现蓝色，而用plot直接show就行。故在用opencv读图片时，记得加下面这句话。

cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

21. 不同计算图之间的联系，用来隔离tensor和计算

#定义计算图g1
g1=tf.Graph()
with g1.as_default():
    v=tf.get_variable('v',shape=[1],initializer=tf.zeros_initializer)
# 定义计算图g2
g2 = tf.Graph()
with g2.as_default():
    v = tf.get_variable('v', shape=[1], initializer=tf.ones_initializer)
#在计算图g1中读取变量v的值
with tf.Session(graph=g1) as sess:
    sess.run(tf.global_variables_initializer())
    with tf.variable_scope('',reuse=True):
        print(sess.run(tf.get_variable('v')))

# 在计算图g2中读取变量v的值
with tf.Session(graph=g2) as sess:
    sess.run(tf.global_variables_initializer())
    with tf.variable_scope('', reuse=True):
        print(sess.run(tf.get_variable('v')))

22. tf.greater,tf.where

v1 = tf.constant([1.0, 2.0, 3.0, 4.0])
v2 = tf.constant([4.0, 3.0, 2.0, 1.0])
with tf.Session() as sess:
    print(sess.run(tf.greater(v1,v2)))

tf.where两种用法

where(condition, x=None, y=None,name=None)

如果x,y为空，返回condition中值为True的位置的Tensor

labels=tf.constant([[[1],[2],[3]],
                    [[4], [5], [6]]],dtype=tf.float32)
ignore_label=-1
a=tf.squeeze(labels)
b=tf.not_equal(a, ignore_label)
c=tf.where(b)
with tf.Session() as sess:
    print(sess.run(a))
    print(sess.run(b))
    print(sess.run(c))

如果x,y不为空，返回值和x、y有相同的形状，如果condition对应位置值为True那么返回Tensor对应位置为x的值，否则为y的值

v1 = tf.constant([1.0, 2.0, 3.0, 4.0])
v2 = tf.constant([4.0, 3.0, 2.0, 1.0])
with tf.Session() as sess:
    print(sess.run(tf.greater(v1,v2)))
    #true选v1,False选2
    print(sess.run(tf.where(tf.greater(v1,v2),v1,v2)))

23. shape,set_shape,reshape,get_shape，注意看feed的是x1,还是x2

x1 = tf.placeholder(tf.float32, shape=[2, 2])
print(tf.shape(x1))
print(x1.get_shape())

可看出，shape返回值是一个tensor,而get_shape返回的是一个tuple

x1 = tf.placeholder(tf.int32)
x2 = tf.reshape(x1, [2, 2])
print(tf.shape(x1))

with tf.Session() as sess:
    print(sess.run(tf.shape(x2), feed_dict={x1: [0, 1, 2, 3]}))

可看出：reshape生成新的shape,创造一个新的tensor以供我们使用

x1 = tf.placeholder(tf.int32)
x1 = tf.reshape(x1, [2, 2])  # use tf.reshape()
print(tf.shape(x1))

sess = tf.Session()
print(sess.run(tf.shape(x1), feed_dict={x1:[0,1,2,3]}))

reshape改变了x1的shape，此时传入就会报错

x1=tf.placeholder(tf.float32)
print(x1.get_shape())

with tf.Session() as sess:
    print(sess.run(tf.shape(x1), feed_dict={x1: [[0, 1], [2, 3]]}))

未加set_shape，没有更新图信息。

x1 = tf.placeholder(tf.int32)
x1.set_shape([2,2])
print(x1.get_shape())

with tf.Session() as sess:
    print(sess.run(tf.shape(x1), feed_dict={x1:[0,1,2,3]}))
    # print(sess.run(tf.shape(x1), feed_dict={x1: [[0, 1], [2, 3]]}))

set_shape更新了图信息

x1 = tf.placeholder(tf.int32)
x1.set_shape([2,2])
print(x1.get_shape())

with tf.Session() as sess:
    print(sess.run(tf.shape(x1), feed_dict={x1:[0,1,2,3]}))
    # print(sess.run(tf.shape(x1), feed_dict={x1: [[0, 1], [2, 3]]}))

图中最开始没有shape的x1在使用了set_shape后，它的图中的信息已经改变了，但是却不能改变tensor的shape,传入了和图不符合的参数就会报错。

x1 = tf.Variable([[0, 1], [2, 3]])
print(x1.get_shape())

x1 = tf.reshape(x1, [4, 1]) 
print(x1.get_shape())

x1 = tf.Variable([[0, 1], [2, 3]])
print(x1.get_shape())

x1 = x1.set_shape([4, 1])
print(x1.get_shape())

可见创建新的tensor或者动态地改变原有tensor的shape的时候可以使用reshape；而当我们只是想更新图中某个tensor的shape或者补充某个tensor的shape信息可以使用set_shape来进行更新

tf.reshape(,-1)变为一行

labels=tf.constant([[[1],[2],[3]],
                    [[4], [5], [6]]],dtype=tf.float32)
d = tf.reshape(labels, [-1])
with tf.Session() as sess:
    print(sess.run(d))

24. 不同计算graph

def different_graph():
    #定义计算图g1
    g1=tf.Graph()
    with g1.as_default():
        v=tf.get_variable('v',shape=[1],initializer=tf.zeros_initializer)
    # 定义计算图g2
    g2 = tf.Graph()
    with g2.as_default():
        v = tf.get_variable('v', shape=[1,2], initializer=tf.ones_initializer)
    #在计算图g1中读取变量v的值
    with tf.Session(graph=g1) as sess:
        sess.run(tf.global_variables_initializer())
        with tf.variable_scope('',reuse=True):
            print('g1_V',sess.run(tf.get_variable('v')))

    # 在计算图g2中读取变量v的值
    with tf.Session(graph=g2) as sess:
        sess.run(tf.global_variables_initializer())
        with tf.variable_scope('', reuse=True):
            print('g2_V',sess.run(tf.get_variable('v')))
            a=tf.get_variable('v')
            print('g2_V.shape',sess.run(tf.shape(a)))

25. tf.boolean_mask，用于找出需要的元素

with tf.Session() as sess:    
    tensor = [0, 1, 2, 3]
    mask = np.array([True, False, True, False])
    print(sess.run(tf.boolean_mask(tensor, mask)))
    #2D
    tensor = [[1, 2], [3, 4], [5, 6]]
    mask = np.array([True, False, True])
    print(sess.run(tf.boolean_mask(tensor, mask)))

26. tf.slice与tf.gather切片

http://www.360doc.com/content/17/0115/14/10408243_622618137.shtml

tf.slice(input_, begin, size, name=None)：按照指定的下标范围抽取连续区域的子集,可以从图片中截取指定的像素点

tf.gather(params, indices, validate_indices=None, name=None)：按照指定的下标集合从axis=0中抽取子集，适合抽取不连续区域的子集

input = tf.constant([[[1, 1, 1], [2, 2, 2]],
                     [[3, 3, 3], [4, 4, 4]],
                     [[5, 5, 5], [6, 6, 6]]],dtype=tf.float32)
a=tf.slice(input, [1, 0, 0], [1, 1, 3])
b=tf.slice(input, [1, 0, 0], [1, 2, 3])
c=tf.slice(input, [1, 0, 0], [2, 1, 3])
with tf.Session() as sess:
    print(sess.run(a))
    print(sess.run(b))
    print(sess.run(c))

input = tf.constant([[[1, 1, 1], [2, 2, 2]],
                     [[3, 3, 3], [4, 4, 4]],
                     [[5, 5, 5], [6, 6, 6]]],dtype=tf.float32)
a=tf.gather(input, [0,1])
with tf.Session() as sess:
    print(sess.run(a))

用于语义分割中，可用于踢掉不需要的ground truth,比如下面示例是6分类，不需要用到6

raw_gt = np.array([0,1,2, 3, 4, 5, 6])
#6 class
less_euqal=tf.less_equal(raw_gt, 6 - 1)
where_index=tf.where(less_euqal)
indices=tf.squeeze(where_index,axis=1)
gt=tf.gather(raw_gt, indices)

然后预测的结果也用类似处理

raw_prediction = tf.reshape(raw_output, [-1, self.conf.num_classes])
prediction = tf.gather(raw_prediction, indices)

# Pixel-wise softmax_cross_entropy loss
loss = tf.nn.sparse_softmax_cross_entropy_with_logits(logits=prediction, labels=gt)

然后在做交叉熵。

完整示例：

raw_gt = tf.reshape(label_proc, [-1,])
indices = tf.squeeze(tf.where(tf.less_equal(raw_gt, self.conf.num_classes - 1)), 1)
gt = tf.cast(tf.gather(raw_gt, indices), tf.int32)
raw_prediction = tf.reshape(raw_output, [-1, self.conf.num_classes])
prediction = tf.gather(raw_prediction, indices)

# Pixel-wise softmax_cross_entropy loss
loss = tf.nn.sparse_softmax_cross_entropy_with_logits(logits=prediction, labels=gt)

27. tf.group,tf.tuple

w = tf.Variable(1)
mul = tf.multiply(w, 2)
add = tf.add(w, 2)
group = tf.group(mul, add)
tuple = tf.tuple([mul, add])
print(group)
print(tuple)
with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    print(sess.run(group))
    print(sess.run(tuple))

# sess.run(group)和sess.run(tuple)都会求Tensor(add)
#Tensor(mul)的值。区别是，tf.group()返回的是`op`
#tf.tuple()返回的是list of tensor。
#这样就会导致，sess.run(tuple)的时候，会返回 Tensor(mul),Tensor(add)的值.
#而 sess.run(group)不会

28. tf.metrics.true_positives

https://blog.csdn.net/jyzhang_cvml/article/details/82694631

a = tf.Variable([0, 1, 1, 0], tf.bool)
b = tf.Variable([0, 1, 0, 1], tf.bool)
tp, tp_update = tf.metrics.true_positives(predictions=a, labels=b)

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    # tp_update 是保存在 tf.local_variables（）中
    sess.run(tf.local_variables_initializer())
    sess.run(tp_update)
    print(sess.run(tp))

29. tf.one_hot

classes = 3
labels = tf.constant([0, 1, 2])  # 输入的元素值最小为0，最大为2
output = tf.one_hot(labels, classes)
with tf.Session() as sess:
    output = sess.run(output)
    print("output of one-hot is : ", output)

classes = 3
labels = tf.constant([[0, 1, 2],
                      [1,2,0]])  # 输入的元素值最小为0，最大为2
output = tf.one_hot(labels, classes)
with tf.Session() as sess:
    output = sess.run(output)
    print("output of one-hot is : ", output)

薛定谔的智能

关注

0
点赞
踩
3

收藏

觉得还不错? 一键收藏
0
评论
tensorflow知识点

https://tensorflow.google.cn/一，打印出与训练变量相关的信息"""slim.model_analyzer.analyze_vars打印出与训练变量相关的信息"""import tensorflow as tfimport tensorflow.contrib.slim as slimx1=tf.Variable(tf.constant(1,shape...
复制链接

扫一扫