テンソルの添字
>>> t = torch.zeros(2, 3, 4)
>>> t
tensor([[[0., 0., 0., 0.],
[0., 0., 0., 0.],
[0., 0., 0., 0.]],
[[0., 0., 0., 0.],
[0., 0., 0., 0.],
[0., 0., 0., 0.]]])
>>> t.shape
torch.Size([2, 3, 4])
>>> t[0][1][2] = 1
>>> t
tensor([[[0., 0., 0., 0.],
[0., 0., 1., 0.],
[0., 0., 0., 0.]],
[[0., 0., 0., 0.],
[0., 0., 0., 0.],
[0., 0., 0., 0.]]])
t.shape が [2, 3, 4] であるということは、4つのまとまりが3つあるまとまりが2つある、みたいな意味
view による変形
>>> t = torch.Tensor([[[1,2,3], [4, 5, 6]], [[7, 8, 9], [10, 11, 12]]])
>>> t
tensor([[[ 1., 2., 3.],
[ 4., 5., 6.]],
[[ 7., 8., 9.],
[10., 11., 12.]]])
>>> t.shape
torch.Size([2, 2, 3])
>>> t2 = t.view(6, 2) # 2個のまとまりを6個に変換
>>> t2
tensor([[ 1., 2.],
[ 3., 4.],
[ 5., 6.],
[ 7., 8.],
[ 9., 10.],
[11., 12.]])
>>> t2.shape
torch.Size([6, 2])
>>> t3 = t.view(4, -1) # X個のまとまりを4個に変換
>>> t3
tensor([[ 1., 2., 3.],
[ 4., 5., 6.],
[ 7., 8., 9.],
[10., 11., 12.]])
>>> t3.shape
torch.Size([4, 3])
>>> t.view(9, -1)
RuntimeError: shape '[9, -1]' is invalid for input of size 12
テンソルの形を要素数が同じという条件下で変形する -1を使うと他に指定された数値からかってに埋めてくれる
squeeze
要素数が1の軸を削除してくれる
>>> t = torch.zeros(2, 1, 3, 1)
>>> t
tensor([[[[0.],
[0.],
[0.]]],
[[[0.],
[0.],
[0.]]]])
>>> t.squeeze() # [2, 1, 3, 1] を [2, 3] に整形してくれる
tensor([[0., 0., 0.],
[0., 0., 0.]])
>>> # 引数(dim)を指定すると dim番めの軸のみに着目
>>> t.squeeze(0) # 0番目は2なのでsqueezeされない
tensor([[[[0.],
[0.],
[0.]]],
[[[0.],
[0.],
[0.]]]])
>>> t.squeeze(1) # 1番目は1なので [2, 3, 1] に整形
tensor([[[0.],
[0.],
[0.]],
[[0.],
[0.],
[0.]]])
>>> t.squeeze(3) # 3番目は1なので[2, 1, 3]に整形
tensor([[[0., 0., 0.]],
[[0., 0., 0.]]])
unsqueeze
squeezeの逆バージョン 要素数が1の軸を足してくれる dimが必須
>>> t = torch.zeros(2, 3)
>>> t
tensor([[0., 0., 0.],
[0., 0., 0.]])
>>> t.unsqueeze(1)
tensor([[[0., 0., 0.]],
[[0., 0., 0.]]])
>>> t.unsqueeze(2)
tensor([[[0.],
[0.],
[0.]],
[[0.],
[0.],
[0.]]])
permute
>>> t = torch.Tensor([[[1,2,3], [4, 5, 6]], [[7, 8, 9], [10, 11, 12]]])
>>> t
tensor([[[ 1., 2., 3.],
[ 4., 5., 6.]],
[[ 7., 8., 9.],
[10., 11., 12.]]])
>>> t2 = t.permute(2, 0, 1)
>>> t2
tensor([[[ 1., 4.],
[ 7., 10.]],
[[ 2., 5.],
[ 8., 11.]],
[[ 3., 6.],
[ 9., 12.]]])
>>> t3 = t.permute(2, 1, 0)
>>> t3
tensor([[[ 1., 7.],
[ 4., 10.]],
[[ 2., 8.],
[ 5., 11.]],
[[ 3., 9.],
[ 6., 12.]]])
>>> t[1][0][2]
tensor(9.)
>>> t2[2][1][0]
tensor(9.)
>>> t3[2][0][1]
tensor(9.)
spacy と Multi30k
from torchtext.datasets import TranslationDataset, Multi30k
from torchtext.data import Field, BucketIterator
import spacy
!python -m spacy download en
spacy_en = spacy.load('en')
!python -m spacy download de
spacy_de = spacy.load('de')
l = spacy_en.tokenizer('This is a pen!')
type(l) # => spacy.tokens.doc.Doc
type(l[0]) # => spacy.tokens.token.Token
l[0].text # => 'This'
def tokenize_de(text):
"""
Tokenizes German text from a string into a list of strings (tokens) and reverses it
"""
return [tok.text for tok in spacy_de.tokenizer(text)]
def tokenize_en(text):
"""
Tokenizes English text from a string into a list of strings (tokens)
"""
return [tok.text for tok in spacy_en.tokenizer(text)]
SRC = Field(tokenize = tokenize_de,
init_token = '<sos>',
eos_token = '<eos>',
lower = True)
TRG = Field(tokenize = tokenize_en,
init_token = '<sos>',
eos_token = '<eos>',
lower = True)
train_data, valid_data, test_data = Multi30k.splits(exts = ('.de', '.en'),
fields = (SRC, TRG))
type(train_data) # => torchtext.datasets.translation.Multi30k
print(vars(train_data.examples[0]))
"""
=> {
'src': [
'zwei',
'junge',
'weiße',
'männer',
'sind',
'im',
'freien',
'in',
'der',
'nähe',
'vieler',
'büsche',
'.'
],
'trg': [
'two',
'young',
',',
'white',
'males',
'are',
'outside',
'near',
'many',
'bushes',
'.'
]
}
"""
SRC.build_vocab(train_data, min_freq = 2)
TRG.build_vocab(train_data, min_freq = 2)
BATCH_SIZE = 128
train_iterator, valid_iterator, test_iterator = BucketIterator.splits(
(train_data, valid_data, test_data),
batch_size = BATCH_SIZE,
device = device)
for b in train_iterator:
break
type(b) # => torchtext.data.batch.Batch
type(b.src) # => torch.Tensor
type(b.trg) # => torch.Tensor
b.src.shape # => [27, 128]
b.trg.shape # => [27, 128]
b.trg
"""
tensor([[ 2, 2, 2, ..., 2, 2, 2],
[ 4, 16, 4, ..., 4, 4, 4],
[ 9, 50, 4224, ..., 14, 9, 61],
...,
[ 1, 1, 1, ..., 1, 1, 1],
[ 1, 1, 1, ..., 1, 1, 1],
[ 1, 1, 1, ..., 1, 1, 1]], device='cuda:0')
"""
# b.trg[:, n] が n文目のはず
b.trg[:, 0]
"""
tensor([ 2, 4, 9, 11, 74, 589, 17, 6, 43, 12, 4, 59, 77, 5,
3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1],
device='cuda:0')
"""
for t in b.trg[:, 0]:
print(TRG.vocab.itos[t])
"""
<sos>
a
man
and
some
bicycles
are
in
front
of
a
large
building
.
<eos>
<pad>
<pad>
<pad>
<pad>
<pad>
<pad>
<pad>
<pad>
<pad>
<pad>
<pad>
<pad>
"""
