Tensorflow一些常用基本概念与函数（1）

刺骨的言语ヽ痛彻心扉 2022-05-24 02:29 277阅读 0赞

摘要：本文主要对tf的一些常用概念与方法进行描述。

--------------------

### 1、tensorflow的基本运作 ###

为了快速的熟悉TensorFlow编程，下面从一段简单的代码开始：

import tensorflow as tf
     #定义‘符号’变量，也称为占位符
     a = tf.placeholder("float")
     b = tf.placeholder("float")
    
     y = tf.mul(a, b) #构造一个op节点
    
     sess = tf.Session()#建立会话
     #运行会话，输入数据，并计算节点，同时打印结果
     print sess.run(y, feed_dict={a: 3, b: 3})
     # 任务完成, 关闭会话.
     sess.close()

*  1
 *  2
 *  3
 *  4
 *  5
 *  6
 *  7
 *  8
 *  9
 *  10
 *  11
 *  12

其中tf.mul(a, b)函数便是tf的一个基本的算数运算，接下来介绍跟多的相关函数。

### 2、tf函数 ###

TensorFlow 将图形定义转换成分布式执行的操作, 以充分利用可用的计算资源(如 CPU 或 GPU。一般你不需要显式指定使用 CPU 还是 GPU, TensorFlow 能自动检测。如果检测到 GPU, TensorFlow 会尽可能地利用找到的第一个 GPU 来执行操作.
    并行计算能让代价大的算法计算加速执行，TensorFlow也在实现上对复杂操作进行了有效的改进。大部分核相关的操作都是设备相关的实现，比如GPU。下面是一些重要的操作/核：

*  1
 *  2
 *  3

#### TensorFlow的算术操作如下： ####

<table style="border-spacing:0px;width:852px;text-align:center;color:rgb(51,51,51);font-family:'-apple-system', 'SF UI Text', Arial, 'PingFang SC', 'Hiragino Sans GB', 'Microsoft YaHei', 'WenQuanYi Micro Hei', sans-serif, SimHei, SimSun;font-size:14px;background-color:rgb(255,255,255);"> 
 <thead> 
 <tr> 
 <th style="margin:0px;vertical-align:middle;">操作</th> 
 <th align="left" style="margin:0px;vertical-align:middle;">描述</th> 
 </tr> 
 </thead> 
 <tbody> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.add(x, y, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">求和</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.sub(x, y, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">减法</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.mul(x, y, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">乘法</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.div(x, y, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">除法</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.mod(x, y, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">取模</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.abs(x, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">求绝对值</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.neg(x, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">取负 (y = -x).</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.sign(x, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">返回符号 y = sign(x) = -1 if x &lt; 0; 0 if x == 0; 1 if x &gt; 0.</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.inv(x, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">取反</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.square(x, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">计算平方 (y = x * x = x^2).</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.round(x, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">舍入最接近的整数 # ‘a’ is [0.9, 2.5, 2.3, -4.4] tf.round(a) ==&gt; [ 1.0, 3.0, 2.0, -4.0 ]</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.sqrt(x, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">开根号 (y = \sqrt{x} = x^{1/2}).</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.pow(x, y, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">幂次方&nbsp; # tensor ‘x’ is [[2, 2], [3, 3]] # tensor ‘y’ is [[8, 16], [2, 3]] tf.pow(x, y) ==&gt; [[256, 65536], [9, 27]]</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.exp(x, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">计算e的次方</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.log(x, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">计算log，一个输入计算e的ln，两输入以第二输入为底</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.maximum(x, y, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">返回最大值 (x &gt; y ? x : y)</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.minimum(x, y, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">返回最小值 (x &lt; y ? x : y)</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.cos(x, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">三角函数cosine</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.sin(x, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">三角函数sine</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.tan(x, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">三角函数tan</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.atan(x, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">三角函数ctan</td> 
 </tr> 
 </tbody> 
</table>

--------------------

#### 张量操作Tensor Transformations ####

*  数据类型转换Casting

*  形状操作Shapes and Shaping

*  切片与合并（Slicing and Joining）

<table style="border-spacing:0px;width:852px;text-align:center;color:rgb(51,51,51);font-family:'-apple-system', 'SF UI Text', Arial, 'PingFang SC', 'Hiragino Sans GB', 'Microsoft YaHei', 'WenQuanYi Micro Hei', sans-serif, SimHei, SimSun;font-size:14px;background-color:rgb(255,255,255);"> 
 <thead> 
 <tr> 
 <th style="margin:0px;vertical-align:middle;">操作</th> 
 <th align="left" style="margin:0px;vertical-align:middle;">描述</th> 
 </tr> 
 </thead> 
 <tbody> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.slice(input_, begin, size, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">对tensor进行切片操作 其中size[i] = input.dim_size(i) - begin[i] 该操作要求 0 &lt;= begin[i] &lt;= begin[i] + size[i] &lt;= Di for i in [0, n] #’input’ is&nbsp; #[[[1, 1, 1], [2, 2, 2]],[[3, 3, 3], [4, 4, 4]],[[5, 5, 5], [6, 6, 6]]] tf.slice(input, [1, 0, 0], [1, 1, 3]) ==&gt; [[[3, 3, 3]]] tf.slice(input, [1, 0, 0], [1, 2, 3]) ==&gt;&nbsp; [[[3, 3, 3], [4, 4, 4]]] tf.slice(input, [1, 0, 0], [2, 1, 3]) ==&gt;&nbsp; [[[3, 3, 3]], [[5, 5, 5]]]</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.split(split_dim, num_split, value, name=’split’)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">沿着某一维度将tensor分离为num_split tensors # ‘value’ is a tensor with shape [5, 30] # Split ‘value’ into 3 tensors along dimension 1 split0, split1, split2 = tf.split(1, 3, value) tf.shape(split0) ==&gt; [5, 10]</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.concat(concat_dim, values, name=’concat’)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">沿着某一维度连结tensor t1 = [[1, 2, 3], [4, 5, 6]] t2 = [[7, 8, 9], [10, 11, 12]] tf.concat(0, [t1, t2]) ==&gt; [[1, 2, 3], [4, 5, 6], [7, 8, 9], [10, 11, 12]] tf.concat(1, [t1, t2]) ==&gt; [[1, 2, 3, 7, 8, 9], [4, 5, 6, 10, 11, 12]] 如果想沿着tensor一新轴连结打包,那么可以： tf.concat(axis, [tf.expand_dims(t, axis) for t in tensors]) 等同于tf.pack(tensors, axis=axis)</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.pack(values, axis=0, name=’pack’)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">将一系列rank-R的tensor打包为一个rank-(R+1)的tensor # ‘x’ is [1, 4], ‘y’ is [2, 5], ‘z’ is [3, 6] pack([x, y, z]) =&gt; [[1, 4], [2, 5], [3, 6]]&nbsp; # 沿着第一维pack pack([x, y, z], axis=1) =&gt; [[1, 2, 3], [4, 5, 6]] 等价于tf.pack([x, y, z]) = np.asarray([x, y, z])</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.reverse(tensor, dims, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">沿着某维度进行序列反转 其中dim为列表，元素为bool型，size等于rank(tensor) # tensor ‘t’ is&nbsp; [[[[ 0, 1, 2, 3], #[ 4, 5, 6, 7], #[ 8, 9, 10, 11]], #[[12, 13, 14, 15], #[16, 17, 18, 19], #[20, 21, 22, 23]]]] # tensor ‘t’ shape is [1, 2, 3, 4] # ‘dims’ is [False, False, False, True] reverse(t, dims) ==&gt; [[[[ 3, 2, 1, 0], [ 7, 6, 5, 4], [ 11, 10, 9, 8]], [[15, 14, 13, 12], [19, 18, 17, 16], [23, 22, 21, 20]]]]</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.transpose(a, perm=None, name=’transpose’)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">调换tensor的维度顺序 按照列表perm的维度排列调换tensor顺序， 如为定义，则perm为(n-1…0) # ‘x’ is [[1 2 3],[4 5 6]] tf.transpose(x) ==&gt; [[1 4], [2 5],[3 6]] # Equivalently tf.transpose(x, perm=[1, 0]) ==&gt; [[1 4],[2 5], [3 6]]</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.gather(params, indices, validate_indices=None, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">合并索引indices所指示params中的切片 <img src="https://img-blog.csdn.net/20160808174705034" alt="tf.gather" title="" style="margin-top:24px;"></td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.one_hot (indices, depth, on_value=None, off_value=None,&nbsp; axis=None, dtype=None, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">indices = [0, 2, -1, 1] depth = 3 on_value = 5.0&nbsp; off_value = 0.0&nbsp; axis = -1&nbsp; #Then output is [4 x 3]:&nbsp; output =&nbsp; [5.0 0.0 0.0] // one_hot(0)&nbsp; [0.0 0.0 5.0] // one_hot(2)&nbsp; [0.0 0.0 0.0] // one_hot(-1)&nbsp; [0.0 5.0 0.0] // one_hot(1)</td> 
 </tr> 
 </tbody> 
</table>

--------------------

#### 矩阵相关运算 ####

--------------------

#### 复数操作 ####

--------------------

#### 归约计算(Reduction) ####

<table style="border-spacing:0px;width:852px;text-align:center;color:rgb(51,51,51);font-family:'-apple-system', 'SF UI Text', Arial, 'PingFang SC', 'Hiragino Sans GB', 'Microsoft YaHei', 'WenQuanYi Micro Hei', sans-serif, SimHei, SimSun;font-size:14px;background-color:rgb(255,255,255);"> 
 <thead> 
 <tr> 
 <th style="margin:0px;vertical-align:middle;">操作</th> 
 <th align="left" style="margin:0px;vertical-align:middle;">描述</th> 
 </tr> 
 </thead> 
 <tbody> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.reduce_sum(input_tensor, reduction_indices=None,&nbsp; keep_dims=False, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">计算输入tensor元素的和，或者安照reduction_indices指定的轴进行求和 # ‘x’ is [[1, 1, 1] # [1, 1, 1]] tf.reduce_sum(x) ==&gt; 6 tf.reduce_sum(x, 0) ==&gt; [2, 2, 2] tf.reduce_sum(x, 1) ==&gt; [3, 3] tf.reduce_sum(x, 1, keep_dims=True) ==&gt; [[3], [3]] tf.reduce_sum(x, [0, 1]) ==&gt; 6</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.reduce_prod(input_tensor,&nbsp; reduction_indices=None,&nbsp; keep_dims=False, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">计算输入tensor元素的乘积，或者安照reduction_indices指定的轴进行求乘积</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.reduce_min(input_tensor,&nbsp; reduction_indices=None,&nbsp; keep_dims=False, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">求tensor中最小值</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.reduce_max(input_tensor,&nbsp; reduction_indices=None,&nbsp; keep_dims=False, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">求tensor中最大值</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.reduce_mean(input_tensor,&nbsp; reduction_indices=None,&nbsp; keep_dims=False, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">求tensor中平均值</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.reduce_all(input_tensor,&nbsp; reduction_indices=None,&nbsp; keep_dims=False, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">对tensor中各个元素求逻辑’与’ # ‘x’ is&nbsp; # [[True, True] # [False, False]] tf.reduce_all(x) ==&gt; False tf.reduce_all(x, 0) ==&gt; [False, False] tf.reduce_all(x, 1) ==&gt; [True, False]</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.reduce_any(input_tensor,&nbsp; reduction_indices=None,&nbsp; keep_dims=False, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">对tensor中各个元素求逻辑’或’</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.accumulate_n(inputs, shape=None,&nbsp; tensor_dtype=None, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">计算一系列tensor的和 # tensor ‘a’ is [[1, 2], [3, 4]] # tensor&nbsp;<code style="font-size:14px;line-height:22px;padding:4px 2px 0px;">b</code>&nbsp;is [[5, 0], [0, 6]] tf.accumulate_n([a, b, a]) ==&gt; [[7, 4], [6, 14]]</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">tf.cumsum(x, axis=0, exclusive=False,&nbsp; reverse=False, name=None)</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">求累积和 tf.cumsum([a, b, c]) ==&gt; [a, a + b, a + b + c] tf.cumsum([a, b, c], exclusive=True) ==&gt; [0, a, a + b] tf.cumsum([a, b, c], reverse=True) ==&gt; [a + b + c, b + c, c] tf.cumsum([a, b, c], exclusive=True, reverse=True) ==&gt; [b + c, c, 0]</td> 
 </tr> 
 <tr> 
 <td style="margin:0px;vertical-align:middle;">&nbsp;</td> 
 <td align="left" style="margin:0px;vertical-align:middle;">&nbsp;</td> 
 </tr> 
 </tbody> 
</table>

--------------------

#### 分割(Segmentation) ####

--------------------

#### 序列比较与索引提取(Sequence Comparison and Indexing) ####

--------------------

#### 神经网络(Neural Network) ####

*  激活函数（Activation Functions）

*  卷积函数（Convolution）

*  池化函数（Pooling）

*  数据标准化（Normalization）

*  损失函数（Losses）

*  分类函数（Classification）

*  符号嵌入（Embeddings）

*  循环神经网络（Recurrent Neural Networks）

> — tf.nn.rnn简要介绍—   
> cell: 一个RNNCell实例   
> inputs: 一个shape为\[batch\_size, input\_size\]的tensor   
> initial\_state: 为RNN的state设定初值，可选   
> sequence\_length：制定输入的每一个序列的长度，size为\[batch\_size\],值范围为\[0, T)的int型数据   
> 其中T为输入数据序列的长度   
> @   
> @针对输入batch中序列长度不同，所设置的动态计算机制   
> @对于在时间t，和batch的b行，有   
> (output, state)(b, t) = ? (zeros(cell.output\_size), states(b, sequence\_length(b) - 1)) : cell(input(b, t), state(b, t - 1))
> 
> --------------------

*  求值网络（Evaluation）

*  [监督候选采样网络（Candidate Sampling）][Candidate Sampling]

对于有巨大量的多分类与多标签模型，如果使用全连接softmax将会占用大量的时间与空间资源，所以采用候选采样方法仅使用一小部分类别与标签作为监督以加速训练。

# 保存与恢复变量 #

--------------------