本文主要是介绍mxnet中ndarray*ndarray用来作为掩码进行与运算的用法，希望对大家解决编程问题提供一定的参考价值，需要的开发者们随着小编来一起学习吧！
def batch_loss(encoder, decoder, X, Y, loss):batch_size = X.shape[0]enc_state = encoder.begin_state(batch_size=batch_size)enc_outputs, enc_state = encoder(X, enc_state)# 初始化解码器的隐藏状态dec_state = decoder.begin_state(enc_state)# 解码器在最初时间步的输入是BOSdec_input = nd.array([out_vocab.token_to_idx[BOS]] * batch_size)# 我们将使用掩码变量mask来忽略掉标签为填充项PAD的损失mask, num_not_pad_tokens = nd.ones(shape=(batch_size,)), 0bflag1=1print("调用我的bflag1=",bflag1,"type(bflag1)=",type(bflag1))if(bflag1 == 1):print("type(mask)=",type(mask),"origin mask=",mask,"mask.shape=",mask.shape)bflag1 = bflag1 +1l = nd.array([0]) for y in Y.T:if(bflag1==2):print("y=",y)bflag1 = bflag1+1dec_output, dec_state = decoder(dec_input, dec_state, enc_outputs)l = l + (mask * loss(dec_output, y)).sum()dec_input = y # 使用强制教学num_not_pad_tokens += mask.sum().asscalar()# 当遇到EOS时，序列后面的词将均为PAD，相应位置的掩码设成0 mask = mask * (y != out_vocab.token_to_idx[EOS])if(bflag1==3):print("mask =",mask,"mask.shape=",mask.shape,"(y != out_vocab.token_to_idx[EOS])=",(y != out_vocab.token_to_idx[EOS]))bflag1 = bflag1+1return l / num_not_pad_tokens输出 注意mask 与 (y != out_vocab.token_to_idx[EOS])的结果都是ndarrayX=
[[ 9. 30. 27. 45. 4. 3. 1.][12. 5. 44. 37. 16. 4. 3.]]
<NDArray 2x7 @cpu(0)> Y=
[[ 7. 5. 31. 11. 18. 4. 3.][ 7. 5. 11. 27. 29. 4. 3.]]
<NDArray 2x7 @cpu(0)> Y.T=
[[ 7. 7.][ 5. 5.][31. 11.][11. 27.][18. 29.][ 4. 4.][ 3. 3.]]
<NDArray 7x2 @cpu(0)>
调用我的bflag1= 1 type(bflag1)= <class 'int'>
type(mask)= <class 'mxnet.ndarray.ndarray.NDArray'> origin mask=
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,)
y=
[7. 7.]
<NDArray 2 @cpu(0)>
mask =
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,) (y != out_vocab.token_to_idx[EOS])=
[1. 1.]
<NDArray 2 @cpu(0)>
调用我的bflag1= 1 type(bflag1)= <class 'int'>
type(mask)= <class 'mxnet.ndarray.ndarray.NDArray'> origin mask=
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,)
y=
[8. 9.]
<NDArray 2 @cpu(0)>
mask =
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,) (y != out_vocab.token_to_idx[EOS])=
[1. 1.]
<NDArray 2 @cpu(0)>
调用我的bflag1= 1 type(bflag1)= <class 'int'>
type(mask)= <class 'mxnet.ndarray.ndarray.NDArray'> origin mask=
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,)
y=
[8. 8.]
<NDArray 2 @cpu(0)>
mask =
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,) (y != out_vocab.token_to_idx[EOS])=
[1. 1.]
<NDArray 2 @cpu(0)>
调用我的bflag1= 1 type(bflag1)= <class 'int'>
type(mask)= <class 'mxnet.ndarray.ndarray.NDArray'> origin mask=
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,)
y=
[9. 8.]
<NDArray 2 @cpu(0)>
mask =
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,) (y != out_vocab.token_to_idx[EOS])=
[1. 1.]
<NDArray 2 @cpu(0)>
调用我的bflag1= 1 type(bflag1)= <class 'int'>
type(mask)= <class 'mxnet.ndarray.ndarray.NDArray'> origin mask=
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,)
y=
[7. 7.]
<NDArray 2 @cpu(0)>
mask =
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,) (y != out_vocab.token_to_idx[EOS])=
[1. 1.]
<NDArray 2 @cpu(0)>
调用我的bflag1= 1 type(bflag1)= <class 'int'>
type(mask)= <class 'mxnet.ndarray.ndarray.NDArray'> origin mask=
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,)
y=
[8. 8.]
<NDArray 2 @cpu(0)>
mask =
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,) (y != out_vocab.token_to_idx[EOS])=
[1. 1.]
<NDArray 2 @cpu(0)>
调用我的bflag1= 1 type(bflag1)= <class 'int'>
type(mask)= <class 'mxnet.ndarray.ndarray.NDArray'> origin mask=
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,)
y=
[7. 9.]
<NDArray 2 @cpu(0)>
mask =
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,) (y != out_vocab.token_to_idx[EOS])=
[1. 1.]
<NDArray 2 @cpu(0)>
调用我的bflag1= 1 type(bflag1)= <class 'int'>
type(mask)= <class 'mxnet.ndarray.ndarray.NDArray'> origin mask=
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,)
y=
[7. 9.]
<NDArray 2 @cpu(0)>
mask =
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,) (y != out_vocab.token_to_idx[EOS])=
[1. 1.]
<NDArray 2 @cpu(0)>
调用我的bflag1= 1 type(bflag1)= <class 'int'>
type(mask)= <class 'mxnet.ndarray.ndarray.NDArray'> origin mask=
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,)
y=
[9. 7.]
<NDArray 2 @cpu(0)>
mask =
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,) (y != out_vocab.token_to_idx[EOS])=
[1. 1.]
<NDArray 2 @cpu(0)>
调用我的bflag1= 1 type(bflag1)= <class 'int'>
type(mask)= <class 'mxnet.ndarray.ndarray.NDArray'> origin mask=
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,)
y=
[8. 9.]
<NDArray 2 @cpu(0)>
mask =
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,) (y != out_vocab.token_to_idx[EOS])=
[1. 1.]
<NDArray 2 @cpu(0)>
调用我的bflag1= 1 type(bflag1)= <class 'int'>
type(mask)= <class 'mxnet.ndarray.ndarray.NDArray'> origin mask=
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,)
y=
[8. 7.]
<NDArray 2 @cpu(0)>
mask =
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,) (y != out_vocab.token_to_idx[EOS])=
[1. 1.]
<NDArray 2 @cpu(0)>
调用我的bflag1= 1 type(bflag1)= <class 'int'>
type(mask)= <class 'mxnet.ndarray.ndarray.NDArray'> origin mask=
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,)
y=
[8. 9.]
<NDArray 2 @cpu(0)>
mask =
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,) (y != out_vocab.token_to_idx[EOS])=
[1. 1.]
<NDArray 2 @cpu(0)>
调用我的bflag1= 1 type(bflag1)= <class 'int'>
type(mask)= <class 'mxnet.ndarray.ndarray.NDArray'> origin mask=
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,)
y=
[9. 9.]
<NDArray 2 @cpu(0)>
mask =
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,) (y != out_vocab.token_to_idx[EOS])=
[1. 1.]
<NDArray 2 @cpu(0)>
调用我的bflag1= 1 type(bflag1)= <class 'int'>
type(mask)= <class 'mxnet.ndarray.ndarray.NDArray'> origin mask=
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,)
y=
[7. 8.]
<NDArray 2 @cpu(0)>
mask =
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,) (y != out_vocab.token_to_idx[EOS])=
[1. 1.]
<NDArray 2 @cpu(0)>
调用我的bflag1= 1 type(bflag1)= <class 'int'>
type(mask)= <class 'mxnet.ndarray.ndarray.NDArray'> origin mask=
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,)
y=
[7. 9.]
<NDArray 2 @cpu(0)>
mask =
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,) (y != out_vocab.token_to_idx[EOS])=
[1. 1.]
<NDArray 2 @cpu(0)>
调用我的bflag1= 1 type(bflag1)= <class 'int'>
type(mask)= <class 'mxnet.ndarray.ndarray.NDArray'> origin mask=
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,)
y=
[9. 7.]
<NDArray 2 @cpu(0)>
mask =
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,) (y != out_vocab.token_to_idx[EOS])=
[1. 1.]
<NDArray 2 @cpu(0)>
调用我的bflag1= 1 type(bflag1)= <class 'int'>
type(mask)= <class 'mxnet.ndarray.ndarray.NDArray'> origin mask=
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,)
y=
[7. 8.]
<NDArray 2 @cpu(0)>
mask =
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,) (y != out_vocab.token_to_idx[EOS])=
[1. 1.]
<NDArray 2 @cpu(0)>
调用我的bflag1= 1 type(bflag1)= <class 'int'>
type(mask)= <class 'mxnet.ndarray.ndarray.NDArray'> origin mask=
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,)
y=
[7. 8.]
<NDArray 2 @cpu(0)>
mask =
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,) (y != out_vocab.token_to_idx[EOS])=
[1. 1.]
<NDArray 2 @cpu(0)>
调用我的bflag1= 1 type(bflag1)= <class 'int'>
type(mask)= <class 'mxnet.ndarray.ndarray.NDArray'> origin mask=
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,)
y=
[8. 7.]
<NDArray 2 @cpu(0)>
mask =
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,) (y != out_vocab.token_to_idx[EOS])=
[1. 1.]
<NDArray 2 @cpu(0)>
调用我的bflag1= 1 type(bflag1)= <class 'int'>
type(mask)= <class 'mxnet.ndarray.ndarray.NDArray'> origin mask=
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,)
y=
[9. 8.]
<NDArray 2 @cpu(0)>
mask =
[1. 1.]
<NDArray 2 @cpu(0)> mask.shape= (2,) (y != out_vocab.token_to_idx[EOS])=
[1. 1.]
<NDArray 2 @cpu(0)>
(y != out_vocab.token_to_idx[EOS])= <class 'mxnet.ndarray.ndarray.NDArray'>
这篇关于mxnet中ndarray*ndarray用来作为掩码进行与运算的用法的文章就介绍到这儿，希望我们推荐的文章对编程师们有所帮助！
