8
0

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?

More than 5 years have passed since last update.

PyTorch で tensor をある軸に従って flip する方法

Posted at

##目的
seq2seqなどのrnnモデルで学習する時、答えのシーケンスを逆転する必要があります。普通のnumpyでやればa[::-1]でできるが、pytorch がマイナスインデクシングをまだ実装してないのでやり方をまとめます。

##numpyの場合

>>> import numpy as np
>>> a=np.random.rand(2,3,4)
>>> a
array([[[0.77035401, 0.24779618, 0.76642706, 0.05111546],
        [0.93068699, 0.63876274, 0.37042163, 0.09062927],
        [0.91439172, 0.91490527, 0.82236844, 0.74023026]],

       [[0.0294182 , 0.15135825, 0.22691598, 0.64636446],
        [0.58836392, 0.97549164, 0.39324753, 0.66243478],
        [0.38082428, 0.20068817, 0.42776323, 0.49329102]]])
>>> a[:,:,::-1]
array([[[0.05111546, 0.76642706, 0.24779618, 0.77035401],
        [0.09062927, 0.37042163, 0.63876274, 0.93068699],
        [0.74023026, 0.82236844, 0.91490527, 0.91439172]],

       [[0.64636446, 0.22691598, 0.15135825, 0.0294182 ],
        [0.66243478, 0.39324753, 0.97549164, 0.58836392],
        [0.49329102, 0.42776323, 0.20068817, 0.38082428]]])

pytorch の場合

torch.arange(3,-1,-1) を使うことで、インデックスの逆転を通じてテンサーの軸を逆転します。

>>> a = torch.rand(2,3,4)
>>> a
tensor([[[0.7077, 0.0939, 0.3385, 0.8469],
         [0.1563, 0.1077, 0.5764, 0.2502],
         [0.3548, 0.1774, 0.8615, 0.1192]],

        [[0.0571, 0.6192, 0.9620, 0.3761],
         [0.3884, 0.6482, 0.3567, 0.4959],
         [0.6242, 0.9567, 0.0100, 0.9072]]])
>>> a[:,:,torch.arange(3,-1,-1)]
tensor([[[0.8469, 0.3385, 0.0939, 0.7077],
         [0.2502, 0.5764, 0.1077, 0.1563],
         [0.1192, 0.8615, 0.1774, 0.3548]],

        [[0.3761, 0.9620, 0.6192, 0.0571],
         [0.4959, 0.3567, 0.6482, 0.3884],
         [0.9072, 0.0100, 0.9567, 0.6242]]])
8
0
0

Register as a new user and use Qiita more conveniently

  1. You get articles that match your needs
  2. You can efficiently read back useful information
  3. You can use dark theme
What you can do with signing up
8
0

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?