My Custom Deep learning Layer is giving me an error

Question

Muhammad Nabi 2019-9-15

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/480450-my-custom-deep-learning-layer-is-giving-me-an-error

回答： Lucas Teng 2020-8-7

I am using CNN of time series data from 8 channels with 500 time samples. My Deep learning architecture consists of the following layers. In the First step i perform 1D convolution across the channels (input size [500 8 1 ])so I get 16, [500 x 1 ] arrays and the output size becomes [500 x 1 x 16].

I wrote a custom function to reshape its size to [500 x 16 ] but it gives me an error when i train the network.

please just see the pictures and help me out. I have tried my best to explain it as far as I can. Thanks

When I Analyzed the network using analyzeNetwork(layers). It showed that my Architecture is correct Here are the results of it

But When I start training the Network It gives me the following error

Here is the code to the Reshape Layer which I wrote:

classdef reshapeLayer < nnet.layer.Layer
    properties
    end
    
    methods
        function layer = reshapeLayer(Name) 
            layer.Name = Name; 
        end
        
        function [Z] = predict(layer, X)
            sz = [size(X,1) size(X,2)*size(X,3) size(X,4)];
            Z  = reshape(X , sz);
        end
        
        function [dldx]  = backward(layer,~,~,dldz,~)
            sz = [size(dldz,1) , 1 , size(dldz,2) , size(dldz,3)];
            dldx = reshape(dldz , sz);
        end 
    end
end