featureinputlayer and convolution1dlayer

featureinputlayer matches fullyconnectedlayer but not matches convolution1dlayer

Answer 1

[yanqi liu]

layers = [
    featureInputLayer(100,'Name','input')
    fullyConnectedLayer(5, 'Name','fc')
    softmaxLayer('Name','sm')
    classificationLayer('Name','classification')]

layers = 
  4×1 Layer array with layers:

     1   'input'            Feature Input           100 features
     2   'fc'               Fully Connected         5 fully connected layer
     3   'sm'               Softmax                 softmax
     4   'classification'   Classification Output   crossentropyex

layers2 = [
    featureInputLayer(100,'Name','input')
    convolution1dLayer(50, 3,'Stride',1, 'Name','cnn1d')
    fullyConnectedLayer(5, 'Name','fc')
    softmaxLayer('Name','sm')
    classificationLayer('Name','classification')]

layers2 = 
  5×1 Layer array with layers:

     1   'input'            Feature Input           100 features
     2   'cnn1d'            Convolution             3 50 convolutions with stride 1 and padding [0  0]
     3   'fc'               Fully Connected         5 fully connected layer
     4   'sm'               Softmax                 softmax
     5   'classification'   Classification Output   crossentropyex

[san su]

Thank you for your answer. Many thanks.

My friend run your code, and add "analyzeNetwork(layers2)". There still exist one error.

layers2 = [

featureInputLayer(100,'Name','input')

convolution1dLayer(50, 3,'Stride',1, 'Name','cnn1d')

fullyConnectedLayer(5, 'Name','fc')

softmaxLayer('Name','sm')

classificationLayer('Name','classification')];

analyzeNetwork(layers2)

（非常感谢回答！我朋友跑了下这段程序，添加了analyzeNetwork(layers2)，还是报错。见图片。他用featureInputLayer和convolution1dLayer的时候就是这样报错。当时是将多个通道合并，matlab不能合并多个sequenceinputlayer，只能用featureInputLayer。报错显示合并后的维度不满足convolution1dLayer对维度的要求。感觉用的挺别扭。）

[yanqi liu]

您好，方便的话可以发一下mat数据，有时候网络提示异常不影响训练

[san su]

您好！matlab这个数据要通过arrayDatastore和combine处理下才能训练，也不太好发数据。后来用imageinputlayer代替了featureinputlayer就解决了，不过想要在fullyconnectedlayer后面接convolution1dlayer是不行的。和上面图中的报错一样，程序根本不走。

[yanqi liu]

是的，如果是做分类，则一般fullyconnectedlayer全连接后就要进入classify了，卷积层一般在中间