9. Tensor¶

9.1. what’s tensor¶

Very briefly, a tensor is an N-dimensional array containing the same type of data (int32, bool, etc.): All you need to describe a tensor fully is:

its data type and
the value of each of the N dimension.

9.2. class Tensor¶

In tf python API, Tensor is a class, <class ‘tensorflow.python.framework.ops.Tensor’>. The data flowing in the graph are instance of Tensor.

从class Tensor的官方文档, 可得对于tensor的理解，见如下代码的注释：

#c, d, and e are Tensor objects, a symbolic(象征性的) handle to one of the outputs of an Operation.
#c, d, and e do not hold the values of that operation's output, but build a dataflow connection between operations
c = tf.constant([[1.0, 2.0], [3.0, 4.0]])
d = tf.constant([[1.0, 1.0], [0.0, 1.0]])
e = tf.matmul(c, d)
#如果要得到“the value that `e` represents”，就得Execute the graph
sess = tf.Session() # Construct a `Session` to execute the graph.
result = sess.run(e) #result就不是Tensor对象了，而是numpy.ndarray对象

9.3. Tensor Type¶

Represents the type of the elements in a Tensor. tf.DType

9.4. tensor name¶

按照一定的规则，由 operation’s name 来决定。

A tensor name has the form “<OP_NAME>:<i>” where:

“<OP_NAME>” is the name of the operation that produces it.
“<i>” is an integer representing the index of that tensor among the operation’s outputs.

9.5. Tensor-like objects¶

tf中有5种 tensor-like objects ,它们在进入operation进行计算时，都将被转换为tf.Tensor

9.6. Manipulate the Shape¶

9.6.1. The shape of numpy’s array¶

https://stackoverflow.com/questions/22053050/difference-between-numpy-array-shape-r-1-and-r

9.6.2. tensor shape and tensor rank¶

rank	dimension	数学实例	python例子	shape
0	0-D	纯量 (只有大小)	s = 483	[]
1	1-D	向量(大小和方向)	v = [1.1, 2.2, 3.3]	[3]
2	2-D	矩阵(数据表)	m = [[1, 2, 3], [4, 5, 6], [7, 8, 9]]	[3,3]

9.6.3. Get the Size of Tensor¶

得到tensor所有元素的总个数， tf.size()

9.6.4. Change the Tensor Shape¶

tf.reshape()
tf.squeeze()，把tensor的shape中值为1的维度去掉

9.6.5. 返回tensor’s shape¶

t = tf.constant([[[1, 1, 1], [2, 2, 2]], [[3, 3, 3], [4, 4, 4]]])
tf.shape(t)  # [2, 2, 3]

9.6.6. 返回tensor’s rank¶

# shape of tensor 't' is [2, 2, 3]
t = tf.constant([[[1, 1, 1], [2, 2, 2]], [[3, 3, 3], [4, 4, 4]]])
tf.rank(t)  # 3

9.7. Manipulate the Type¶

9.7.1. Cast the Type of Tensor¶

tf.cast()
tf.to_int32()

x = tf.constant([1.8, 2.2], dtype=tf.float32)
tf.cast(x, tf.int32)  # [1, 2], dtype=tf.int32

9.8. Manipulate the Element¶

9.8.1. 获得tensor某个维度上的最值的index¶

tf.argmax(input_tensor, axis)

9.8.2. 从Tensor中摘取数据¶

tf.gather()，从一个tensor(params)的指定维度(axis)的指定位置(indices)获取element，组合成一个tensor

tf.where(

condition, x=None, y=None, name=None),从两个tensor,x,y,中选取element/row，组合成一个tensor

9.8.3. 矩阵相乘¶

tf.matmul(h_pool, W)

用于全连接的神经网络, e.g.CNN中的全连接层

9.8.4. 矩阵按元素相乘¶

1	matrix_1 * matrix_2

9.8.5. 矩阵按行/列求和¶

tf.reduce_sum(matrix, axis)

其中，

axis=0, 按列
axis=1, 按行
axis=None, all dimensions are reduced, and a tensor with a single element is returned.

9.8.6. element-wise 比较两个tensor¶

equal(
    x,
    y,
    name=None
)

9.8.7. 找出tensor中每行(last dimension)最大的k个数¶

tf.nn.top_k(), tutorial and examples

9.9. 求tensor的值(Evaluating)¶

Evaluating tensor和executing a graph是一个意思。

>>> a = tf.constant([10,20])
>>> a
<tf.Tensor 'Const_1:0' shape=(2,) dtype=int32>
>>> sess = tf.InteractivateSession()
>>> v = sess.run(a)
>>> v
array([10, 20])
>>> type(v)
<class 'numpy.ndarray'>