我有一个堆叠的自动编码器网络,我想获得调谐网络的中间输出 . 我的输入数据是1620 x 3000,我的目标数据是12 x 3000.我的测试数据是1620 x 500.我有第一层有400个神经元,第二层有50个神经元,而softmax层到12个 . 我想要调谐后得到第二个自动编码器(50个输出)的输出 .
ann1 = trainAutoencoder(xTrain, 400, ...
'L2WeightRegularization', 0.004, ...
'SparsityRegularization', 4, ...
'SparsityProportion', 0.05, ...
'ScaleData', false)
out1 = encode(ann1, xTrain);
ann2 = trainAutoencoder(out1, 50, ...
'L2WeightRegularization', 0.004, ...
'SparsityRegularization', 4, ...
'SparsityProportion', 0.05, ...
'ScaleData', false)
out2 = encode(ann2, out1);
softNet = trainSoftmaxLayer(out2, yTrain);
deepNet = stack(ann1, ann2, softNet);
tunedDeepNet = train(deepNet, xTrain, yTrain);
outputs = tunedDeepNet(xTest);
% outputs is 12x500
% I would like to get the tuned outputs of the second autoencoder, 50x500
% This gives me 50x500, but it is before tuning.
shortNet = stack(ann1, ann2);
out50 = shortNet(xTest);