Transposes
a
. Permutes the dimensions according to perm.The returned tensor's dimension i will correspond to the input dimension
perm[i]
. Ifperm
is not given, it is set to (n-1...0), where n is the rank of the input tensor. Hence by default, this operation performs a regular matrix transpose on 2-D input Tensors.
但它仍然是一个有点不清楚我,我应该如何切分的输入张量。例如。从文档太:
tf.transpose(x, perm=[0, 2, 1]) ==> [[[1 4]
[2 5]
[3 6]]
[[7 10]
[8 11]
[9 12]]]
为什么是它perm=[0,2,1]
产生1x3x2张量?
一些试验和错误后:
twothreefour = np.array([ [[1,2,3,4], [5,6,7,8], [9,10,11,12]] ,
[[13,14,15,16], [17,18,19,20], [21,22,23,24]] ])
twothreefour
[出]:
array([[[ 1, 2, 3, 4],
[ 5, 6, 7, 8],
[ 9, 10, 11, 12]],
[[13, 14, 15, 16],
[17, 18, 19, 20],
[21, 22, 23, 24]]])
如果我转它:
fourthreetwo = tf.transpose(twothreefour)
with tf.Session() as sess:
init = tf.initialize_all_variables()
sess.run(init)
print (fourthreetwo.eval())
我得到一个4x3x2的2x3x4和这听起来很合理。
[出]:
[[[ 1 13]
[ 5 17]
[ 9 21]]
[[ 2 14]
[ 6 18]
[10 22]]
[[ 3 15]
[ 7 19]
[11 23]]
[[ 4 16]
[ 8 20]
[12 24]]]
但是,当我使用perm
参数输出,我不知道什么我真的越来越:
twofourthree = tf.transpose(twothreefour, perm=[0,2,1])
with tf.Session() as sess:
init = tf.initialize_all_variables()
sess.run(init)
print (threetwofour.eval())
[出]:
[[[ 1 5 9]
[ 2 6 10]
[ 3 7 11]
[ 4 8 12]]
[[13 17 21]
[14 18 22]
[15 19 23]
[16 20 24]]]
为什么perm=[0,2,1]
从2x3x4?返回2x4x3矩阵?
与perm=[1,0,2]
再次尝试它:
threetwofour = tf.transpose(twothreefour, perm=[1,0,2])
with tf.Session() as sess:
init = tf.initialize_all_variables()
sess.run(init)
print (threetwofour.eval())
[出]:
[[[ 1 2 3 4]
[13 14 15 16]]
[[ 5 6 7 8]
[17 18 19 20]]
[[ 9 10 11 12]
[21 22 23 24]]]
为什么perm=[1,0,2]
回报从2x3x4一个3x2x4?
这是否意味着perm
参数考虑我np.shape
,并根据我的数组形状调换基于元素的张量?
I.e. :
_size = (2, 4, 3, 5)
randarray = np.random.randint(5, size=_size)
shape_idx = {i:_s for i, _s in enumerate(_size)}
randarray_t_func = tf.transpose(randarray, perm=[3,0,2,1])
with tf.Session() as sess:
init = tf.initialize_all_variables()
sess.run(init)
tranposed_array = randarray_t_func.eval()
print (tranposed_array.shape)
print (tuple(shape_idx[_s] for _s in [3,0,2,1]))
[出]:
(5, 2, 3, 4)
(5, 2, 3, 4)
我想是这样,但我不知道。如果你可以链接到repo中的特定代码行或更清晰的文档,这将会有很大的帮助!提前致谢! – alvas
我认为文档中的例子给出了一个足够好的例子来说明发生了什么。维0是内矩阵,并且它们通过置换不变,维1是内矩阵的行,维2是列,并且它们通过置换来切换。因此每个内部矩阵的第1行进入相同内部矩阵的第1列。这是否有意义? – maxymoo
阅读起来有点令人困惑,但在尝试用更多类似于我发布的代码的例子说服自己之后,我得到了它。 – alvas