TensorFlow 1.x学习（系列二 ：1）：基本概念TensorFlow的基本介绍，图，会话，会话中的run（），placeholder(),常见的报错

写在前边的话：之前发布过一个关于TensorFlow1.x的转载系列，自己将基本的TensorFlow操作敲了一遍，但是仍然有很多地方理解的不够深入。所以重开一个系列，跟着网上找到的教程边听边再敲一遍。最终的目的是实现一个新闻分类的demo,代码已有但是之前没有看懂。再往后应该会出一个pytorch的系列，最后目的是将tensorflow1.x的代码用pytorch再实现一遍。

1.基本介绍

tensorflow1.x 有其独有的语法体系，不同于Python代码的是，自定义的变量和函数无法直接输出结果，必须要在会话中完成该操作。

import tensorflow as tf

C:\Anaconda\envs\tensorflow16\lib\site-packages\tensorflow\python\framework\dtypes.py:517: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_qint8 = np.dtype([("qint8", np.int8, 1)])
C:\Anaconda\envs\tensorflow16\lib\site-packages\tensorflow\python\framework\dtypes.py:518: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_quint8 = np.dtype([("quint8", np.uint8, 1)])
C:\Anaconda\envs\tensorflow16\lib\site-packages\tensorflow\python\framework\dtypes.py:519: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_qint16 = np.dtype([("qint16", np.int16, 1)])
C:\Anaconda\envs\tensorflow16\lib\site-packages\tensorflow\python\framework\dtypes.py:520: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_quint16 = np.dtype([("quint16", np.uint16, 1)])
C:\Anaconda\envs\tensorflow16\lib\site-packages\tensorflow\python\framework\dtypes.py:521: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_qint32 = np.dtype([("qint32", np.int32, 1)])
C:\Anaconda\envs\tensorflow16\lib\site-packages\tensorflow\python\framework\dtypes.py:526: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  np_resource = np.dtype([("resource", np.ubyte, 1)])

# 实现一个加法运算
a = tf.constant(5.0)
b = tf.constant(6.0)
sum1 = tf.add(a,b)

# 查看结果
print(a,b)
print(sum1)

Tensor("Const:0", shape=(), dtype=float32) Tensor("Const_1:0", shape=(), dtype=float32)
Tensor("Add:0", shape=(), dtype=float32)

# 在会话中实现
with tf.Session() as sess:
    print(sess.run(sum1))

11.0

2.图的结构

一些概念：

tensor:张量
operation(op):专门运算的操作节点（注：所有操作都是一个op）
图(graph):你的整个程序的结构
会话(Session):运算程序的图

# 查看默认的图,根据打印的结果可以看到，图的在内存中的位置
graph = tf.get_default_graph()
graph

<tensorflow.python.framework.ops.Graph at 0x1c82a43df60>

# 打印各op的graph属性可以看到，地址都一样，都是该图的一部分
with tf.Session() as sess:
    print(sess.run(sum1))
    print(a.graph)
    print(sum1.graph)
    print(sess.graph)

11.0
<tensorflow.python.framework.ops.Graph object at 0x000001C82A43DF60>
<tensorflow.python.framework.ops.Graph object at 0x000001C82A43DF60>
<tensorflow.python.framework.ops.Graph object at 0x000001C82A43DF60>

# 一般一个程序里只有一个图，那么如何定义其他图呢
g = tf.Graph()
print(g)

<tensorflow.python.framework.ops.Graph object at 0x000001C82A48E588>

# 可以看到新定义的图和之前的图内存位置并不同
with g.as_default():
    c = tf.constant(11.0)
    print(c)
    print(c.graph)

Tensor("Const:0", shape=(), dtype=float32)
<tensorflow.python.framework.ops.Graph object at 0x000001C82A48E588>

总结：

创建一张图包含了一组op和tensor，上下文环境
op：只要使用tensorflow的API定义的函数都是OP
tensor:指代的就是数据

3.会话，会话的run方法

理解：

可以将tensorflow看做两部分：

前端系统：定义程序的图结构

后端系统：运算图结构

会话的作用：

1.运行图的结构

2.分配资源计算

3.掌握资源（变量的资源，队列，线程）

一次只能运行一个图，可以在会话中指定图去运行tf.Session(graph = )

# 一次只能运行一个图，可以在会话中指定图去运行tf.Session(graph = )
with tf.Session() as sess:
    print(sess.run(c))
    print(a.graph)
    print(sum1.graph)
    print(sess.graph)

# 会报错，因为c是别的图的一部分
# ValueError: Fetch argument <tf.Tensor 'Const:0' shape=() dtype=float32> cannot be interpreted as a Tensor. (Tensor Tensor("Const:0", shape=(), dtype=float32) is not an element of this graph.)

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

ValueError                                Traceback (most recent call last)

C:\Anaconda\envs\tensorflow16\lib\site-packages\tensorflow\python\client\session.py in __init__(self, fetches, contraction_fn)
    281         self._unique_fetches.append(ops.get_default_graph().as_graph_element(
--> 282             fetch, allow_tensor=True, allow_operation=True))
    283       except TypeError as e:


C:\Anaconda\envs\tensorflow16\lib\site-packages\tensorflow\python\framework\ops.py in as_graph_element(self, obj, allow_tensor, allow_operation)
   3458     with self._lock:
-> 3459       return self._as_graph_element_locked(obj, allow_tensor, allow_operation)
   3460 


C:\Anaconda\envs\tensorflow16\lib\site-packages\tensorflow\python\framework\ops.py in _as_graph_element_locked(self, obj, allow_tensor, allow_operation)
   3537       if obj.graph is not self:
-> 3538         raise ValueError("Tensor %s is not an element of this graph." % obj)
   3539       return obj


ValueError: Tensor Tensor("Const:0", shape=(), dtype=float32) is not an element of this graph.

​
During handling of the above exception, another exception occurred:

ValueError                                Traceback (most recent call last)

<ipython-input-10-c64ee74128b9> in <module>
      1 # 一次只能运行一个图
      2 with tf.Session() as sess:
----> 3     print(sess.run(c))
      4     print(a.graph)
      5     print(sum1.graph)


C:\Anaconda\envs\tensorflow16\lib\site-packages\tensorflow\python\client\session.py in run(self, fetches, feed_dict, options, run_metadata)
    903     try:
    904       result = self._run(None, fetches, feed_dict, options_ptr,
--> 905                          run_metadata_ptr)
    906       if run_metadata:
    907         proto_data = tf_session.TF_GetBuffer(run_metadata_ptr)


C:\Anaconda\envs\tensorflow16\lib\site-packages\tensorflow\python\client\session.py in _run(self, handle, fetches, feed_dict, options, run_metadata)
   1120     # Create a fetch handler to take care of the structure of fetches.
   1121     fetch_handler = _FetchHandler(
-> 1122         self._graph, fetches, feed_dict_tensor, feed_handles=feed_handles)
   1123 
   1124     # Run request and get response.


C:\Anaconda\envs\tensorflow16\lib\site-packages\tensorflow\python\client\session.py in __init__(self, graph, fetches, feeds, feed_handles)
    425     """
    426     with graph.as_default():
--> 427       self._fetch_mapper = _FetchMapper.for_fetch(fetches)
    428     self._fetches = []
    429     self._targets = []


C:\Anaconda\envs\tensorflow16\lib\site-packages\tensorflow\python\client\session.py in for_fetch(fetch)
    251         if isinstance(fetch, tensor_type):
    252           fetches, contraction_fn = fetch_fn(fetch)
--> 253           return _ElementFetchMapper(fetches, contraction_fn)
    254     # Did not find anything.
    255     raise TypeError('Fetch argument %r has invalid type %r' % (fetch,


C:\Anaconda\envs\tensorflow16\lib\site-packages\tensorflow\python\client\session.py in __init__(self, fetches, contraction_fn)
    287       except ValueError as e:
    288         raise ValueError('Fetch argument %r cannot be interpreted as a '
--> 289                          'Tensor. (%s)' % (fetch, str(e)))
    290       except KeyError as e:
    291         raise ValueError('Fetch argument %r cannot be interpreted as a '


ValueError: Fetch argument <tf.Tensor 'Const:0' shape=() dtype=float32> cannot be interpreted as a Tensor. (Tensor Tensor("Const:0", shape=(), dtype=float32) is not an element of this graph.)

指定图运行

# 指定图运行
with tf.Session(graph = g) as sess:
    print(sess.run(c))
    print(a.graph) # 这个能正常输出是因为在会话外就定义过
    print(sum1.graph) # 同上
    print(sess.graph)

11.0
<tensorflow.python.framework.ops.Graph object at 0x000001C82A43DF60>
<tensorflow.python.framework.ops.Graph object at 0x000001C82A43DF60>
<tensorflow.python.framework.ops.Graph object at 0x000001C82A48E588>

关于sess.run(fetches,feed_dict=None,graph = None):

1.作用：运行ops和计算tensor,相当于是启动整个图

2.sess.close():与sess.run()相对应，关闭资源。但是在使用上下文管理器的结构中（with tf.Session as sess:...）可以省略

其他：

3.tf.Session() 中还有另外一个参数config （tf.Session(config = tf.ConfigPorto(log_device_placement=True))）作用是显示你的op具体是在那个设备上运行的以及其他详细情况。

4.交互式：在命令行里使用，tf.interactiveSession(),方便调试，结合变量.eval()比较方便

例1:一次run多个

# 例1:一次run多个
with tf.Session() as sess:
    print(sess.run([a,b,sum1]))

[5.0, 6.0, 11.0]

2-1：run只能运行op和tensor

# 例2-1：run只能运行op和tensor
var1 = 2
var2 = 3
sum2 = var1 + var2
with tf.Session() as sess:
    print(sess.run(sum2))

# TypeError: Fetch argument 5 has invalid type <class 'int'>, must be a string or Tensor. (Can not convert a int into a Tensor or Operation.)

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

TypeError                                 Traceback (most recent call last)

C:\Anaconda\envs\tensorflow16\lib\site-packages\tensorflow\python\client\session.py in __init__(self, fetches, contraction_fn)
    281         self._unique_fetches.append(ops.get_default_graph().as_graph_element(
--> 282             fetch, allow_tensor=True, allow_operation=True))
    283       except TypeError as e:


C:\Anaconda\envs\tensorflow16\lib\site-packages\tensorflow\python\framework\ops.py in as_graph_element(self, obj, allow_tensor, allow_operation)
   3458     with self._lock:
-> 3459       return self._as_graph_element_locked(obj, allow_tensor, allow_operation)
   3460 


C:\Anaconda\envs\tensorflow16\lib\site-packages\tensorflow\python\framework\ops.py in _as_graph_element_locked(self, obj, allow_tensor, allow_operation)
   3547       raise TypeError("Can not convert a %s into a %s." % (type(obj).__name__,
-> 3548                                                            types_str))
   3549 


TypeError: Can not convert a int into a Tensor or Operation.

​
During handling of the above exception, another exception occurred:

TypeError                                 Traceback (most recent call last)

<ipython-input-13-31c14cd7002b> in <module>
      4 sum2 = var1 + var2
      5 with tf.Session() as sess:
----> 6     print(sess.run(sum2))


C:\Anaconda\envs\tensorflow16\lib\site-packages\tensorflow\python\client\session.py in run(self, fetches, feed_dict, options, run_metadata)
    903     try:
    904       result = self._run(None, fetches, feed_dict, options_ptr,
--> 905                          run_metadata_ptr)
    906       if run_metadata:
    907         proto_data = tf_session.TF_GetBuffer(run_metadata_ptr)


C:\Anaconda\envs\tensorflow16\lib\site-packages\tensorflow\python\client\session.py in _run(self, handle, fetches, feed_dict, options, run_metadata)
   1120     # Create a fetch handler to take care of the structure of fetches.
   1121     fetch_handler = _FetchHandler(
-> 1122         self._graph, fetches, feed_dict_tensor, feed_handles=feed_handles)
   1123 
   1124     # Run request and get response.


C:\Anaconda\envs\tensorflow16\lib\site-packages\tensorflow\python\client\session.py in __init__(self, graph, fetches, feeds, feed_handles)
    425     """
    426     with graph.as_default():
--> 427       self._fetch_mapper = _FetchMapper.for_fetch(fetches)
    428     self._fetches = []
    429     self._targets = []


C:\Anaconda\envs\tensorflow16\lib\site-packages\tensorflow\python\client\session.py in for_fetch(fetch)
    251         if isinstance(fetch, tensor_type):
    252           fetches, contraction_fn = fetch_fn(fetch)
--> 253           return _ElementFetchMapper(fetches, contraction_fn)
    254     # Did not find anything.
    255     raise TypeError('Fetch argument %r has invalid type %r' % (fetch,


C:\Anaconda\envs\tensorflow16\lib\site-packages\tensorflow\python\client\session.py in __init__(self, fetches, contraction_fn)
    284         raise TypeError('Fetch argument %r has invalid type %r, '
    285                         'must be a string or Tensor. (%s)' %
--> 286                         (fetch, type(fetch), str(e)))
    287       except ValueError as e:
    288         raise ValueError('Fetch argument %r cannot be interpreted as a '


TypeError: Fetch argument 5 has invalid type <class 'int'>, must be a string or Tensor. (Can not convert a int into a Tensor or Operation.)

例2-2 有重载的机制，默认会给运算符重载成op类型

# 例2-2 有重载的机制，默认会给运算符重载成op类型
var1 = 2.0
sum2 = a + var1
with tf.Session() as sess:
    print(sess.run(sum2))

7.0

4.placeholder

应用场景：训练模型时需要实时提供数据去训练

介绍：

1.placeholder是一个占位符，使用中充当feed_dict字典中的键

2.参数：placeholder(dtype,shape = None,name = None)

例1 ：placeholder的使用

# 例1 ：placeholder的使用
plt = tf.placeholder(tf.float32,[2,3])

with tf.Session(config = tf.ConfigProto(log_device_placement = True)) as sess:
    print(sess.run(plt,feed_dict = {plt:[[1,2,3],[4,5,6]]}))

[[1. 2. 3.]
 [4. 5. 6.]]

例2 ：placeholder的使用2（样本行数不固定）

# 例2 ：placeholder的使用2（样本行数不固定）
plt = tf.placeholder(tf.float32,[None,3]) # n行3列

with tf.Session(config = tf.ConfigProto(log_device_placement = True)) as sess:
    print(sess.run(plt,feed_dict = {plt:[[1,2,3],[4,5,6],[7,8,9]]}))

[[1. 2. 3.]
 [4. 5. 6.]
 [7. 8. 9.]]

5.返回值异常

RuntimeError: 如果它Session处于无效状态（例如已关闭）。

TypeError:如果fetches或feed_dict键是不合适的类型。

ValueError:如果fetches或feed_dict键无效或引用tensor不存在