Two layers of autoencoder reconstruction error is bigger than a layer, is normal?-CodePudding

Introduction to deep learning small white help, MATLAB with function of a two layer autoencoder (encode - encode decode - decode), the training sample input x reconstruction error, than with a layer of (encode, decode), this kind of situation is normal? Or something wrong with the code?

The code is as follows:

% "Train" the first encoder

Autoenc1=trainAutoencoder (x, I, 'MaxEpochs', 400, 'DecoderTransferFunction' and 'purelin');

% Extract the encoded data for new images using the first autoencoder.

Features1=encode (autoenc1, x);

% "Train" the second encoder

Autoenc2=trainAutoencoder (features1, 10, 'MaxEpochs' 400, 'DecoderTransferFunction' and 'purelin');

% Extract the encoded data for new images using the second autoencoder.

Features2=encode (autoenc2 features1);

% Decode the encoded data from the autoencoder.

Regenerated2=decode (autoenc2 features2);

Regenerated=decode (autoenc1 regenerated2);

% Calculate the reconstruction error

Prefomance=SQRT (mse) (x - regenerated);

P.S. light spray why not use Python...