Tensorflow2.0中的Tensor维度操作

改变Tensor的维度

我想更改Tensor的维度，我发现了3种方法来实现这一点，如下所示

a = tf.constant([[1,2,3],[4,5,6]]) # shape (2,3)

# change dimention of a to (2,3,1)
b = tf.expand_dims(a,2)   # shape(2,3,1)
c = a[:,:,tf.newaxis]     # shape(2,3,1)
d = tf.reshape(a,(2,3,1)) # shape(2,3,1)

不同点

tf.expand_dims

tf.expand_dims(
    input,
    axis,
    name=None
)	#添加一个维度并且它的索引是可变的	

t = [[1, 2, 3],[4, 5, 6]] 	# shape   [2, 3] 
tf.expand_dims(t, 0) 		#shape=(1, 2, 3)
tf.expand_dims(t, 1) 		#shape=(2, 1, 3)
tf.expand_dims(t, 2) 		#shape=(2, 3, 1)
tf.expand_dims(t, -1) 		#shape=(2, 3, 1)

上例t的类型为list	 tf.expand_dims后类型变为tf.Tensor

tf.newaxis

a[:,:,tf.newaxis]			#可读性强	添加多个尺寸特别方便（而不是多次调用tf.expand_dims）

为什么可读性强强？ 例如

foo = tf.constant([[1,2,3], [4,5,6], [7,8,9]])

print(foo.shape)
print(foo[tf.newaxis, :, :]) # => [[[1,2,3], [4,5,6], [7,8,9]]]
print(foo[:, tf.newaxis, :]) # => [[[1,2,3]], [[4,5,6]], [[7,8,9]]]
print(foo[:, :, tf.newaxis]) # =>[[[1],[2],[3]], [[4],[5],[6]],[[7],[8],[9]]]

# 输出的4个的维度依次(3, 3)	(1, 3, 3)	(3, 1, 3)	(3, 3, 1)
# tf.newaxis表示添加的一个维度		:代表原来的维度		再比如
print(foo[:, :, tf.newaxis, tf.newaxis, tf.newaxis])#shape=(3, 3, 1, 1, 1)

foo = tf.constant([[1,2,3], [4,5,6], [7,8,9]])
print(foo)
print(foo[tf.newaxis, :, :])  # => [[[1,2,3], [4,5,6], [7,8,9]]]
print(foo[tf.newaxis, ...])  # => [[[1,2,3], [4,5,6], [7,8,9]]]
print(foo[tf.newaxis])  # => [[[1,2,3], [4,5,6], [7,8,9]]]
# shape=(3, 3)		(1, 3, 3)	(1, 3, 3)	(1, 3, 3)
# ...代表所以的维度

Tensor操作

去除最前和最后的2个元素

foo = tf.constant([1,2,3,4,5,6])
print(foo[2:-2])			# => [3,4]

跳过每隔一行并反转列的顺序

foo = tf.constant([[1,2,3], [4,5,6], [7,8,9]])
print(foo[::2,::-1])		# => [[3,2,1], [9,8,7]]

将标量张量用作两个维度的索引

foo = tf.constant([[1,2,3], [4,5,6], [7,8,9]])	#shape=(2, 3)
print(foo[tf.constant(0), tf.constant(2)])		# => 3	（第0维的第3个元素）

Masks

foo = tf.constant([[1,2,3], [4,5,6], [7,8,9]])

print(foo)
print("*"*50)	#分割线

print([foo > 2])
print("*"*50)	#分割线

print(foo[foo > 2])
print("*"*50)	#分割线

输出

tf.Tensor(
[[1 2 3]
 [4 5 6]
 [7 8 9]], shape=(3, 3), dtype=int32)
**************************************************
[<tf.Tensor: id=387, shape=(3, 3), dtype=bool, numpy=
array([[False, False,  True],
       [ True,  True,  True],
       [ True,  True,  True]])>]
**************************************************
tf.Tensor([3 4 5 6 7 8 9], shape=(7,), dtype=int32)
**************************************************