seq2seq+torch7聊天机器人bug处理

	[C]: in function 'error'
	...root1/torch/install/share/lua/5.2/rnn/recursiveUtils.lua:44: in function 'recursiveCopy'
	./seq2seq.lua:58: in function 'backwardConnect'
	./seq2seq.lua:78: in function 'train'
	train.lua:90: in main chunk
	[C]: in function 'dofile'
	...oot1/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:150: in main chunk
	[C]: in ?

出现这个bug应该是tensorflow更新问题。

修改seq2seq.lua文件

function Seq2Seq:backwardConnect()
  -- self.encoderLSTM.userNextGradCell =--
  --   nn.rnn.recursiveCopy(self.encoderLSTM.userNextGradCell, self.decoderLSTM.userGradPrevCell)
  -- self.encoderLSTM.gradPrevOutput =
  --   nn.rnn.recursiveCopy(self.encoderLSTM.gradPrevOutput, self.decoderLSTM.userGradPrevOutput)
  if (self.encoderLSTM.userNextGradCell ~= nil) then
    self.encoderLSTM.userNextGradCell =
    nn.rnn.recursiveCopy(self.encoderLSTM.userNextGradCell, self.decoderLSTM.userGradPrevCell)
  end
  if (self.encoderLSTM.gradPrevOutput ~= nil) then
    self.encoderLSTM.gradPrevOutput =
    nn.rnn.recursiveCopy(self.encoderLSTM.gradPrevOutput, self.decoderLSTM.userGradPrevOutput)
  end
end

训练完之后会出现无法对话，原因是没有cunn

所以必须安装cuda

有个地方修改后会无限出现我是小通

英文原版的几个作者写的也不尽相同，中文版有和他们不同，所以很坑