y[:,:1]与y[:,:-1]

import torch
import torch.nn as nn
import torch
from torch.autograd import *
input = torch.randn(4,4)

print(input)
#m = nn.LayerNorm(input.size()[1:])
#m = nn.LayerNorm(input.size()[1:], elementwise_affine=False)
#m = nn.LayerNorm([2, 2])
#m = nn.LayerNorm(2)
#y = input
#y = y.long()
#_,y= torch.max(input,0)
y = input
print(y.eq(0.))
y = (1. - y.eq(0.).float())
print(3.eq(0.))
print(y)

'''
decoder_inputs = torch.cat([Variable(torch.ones(y[:, :1].size()) * 2).long(), y[:, :-1]], dim=-1)  # 2:<S>
de = torch.unsqueeze(torch.arange(0, decoder_inputs.size()[1]), 0).repeat(decoder_inputs.size(0), 1).long()
dec = torch.arange(0, decoder_inputs.size()[1])
k = torch.arange(0,7)
print(k)
print(dec)
print(de)
print(de.shape)
print(decoder_inputs)
print(decoder_inputs.shape)
print(y[:,:1].shape)
print(y[:,:-1])
print(y[:,:1])'''