Deep Learning LSTM sequenceInputLayer - Data normalization on Test Data?

Question

Barry 2020-8-13

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/578994-deep-learning-lstm-sequenceinputlayer-data-normalization-on-test-data

评论： Barry 2020-8-17

Hi all,

when using the sequenceInputLayer option "Normalization", "zscore" (for example) will the same normalization be applied on the Testing Data when using the classify or predict function? My understanding is that i always have to use the same normalization on the Test Data as i used on the Training Data. Or am i missing something?

Regards,

Barry

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Raunak Gupta 2020-8-16

2
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/578994-deep-learning-lstm-sequenceinputlayer-data-normalization-on-test-data#answer_480468

Hi Barry,

The normalization is applied on every batch of the data that passes through any particular data input layer whether being sequenceInputLayer or imageInputLayer. So, when the training or testing happens it calls a forward function which invokes the batch normalization for that input layer with option like “zscore”, “zerocenter” etc.

No need to worry about saving any parameter about normalization because it varies from batch to batch but be sure that normalization applies to both training and testing data while passing it through the Network.