在 MATLAB 中训练简单的神经网络时出现错误答案

【问题标题】：I get an error when training simple Neural Network in MATLAB在 MATLAB 中训练简单的神经网络时出现错误
【发布时间】：2019-10-18 23:13:14
【问题描述】：

我已经在MATLAB中构建了简单的神经网络，如下代码所示，输入和输出数据的维度为128x1，我将输入数据划分为21x1用于验证和107用于训练：

clear all; clc; clear;  

bits = randi([0 1], 1 , 128);                 %Original generated data 
data_temp = bi2de(reshape(bits,128/2,2));   
data_mod = (1/sqrt(2))*qammod(data_temp , 4);  

inputs = [real(data_mod); imag(data_mod)];      %the input data of input layer (128 x 1); 
outputs = bits.';                               %output data from output layer (128 x 1); 

inputs_len = 128;                    %length of input data 
outputs_len = 128;                   %length of output data 

IN_V1 = inputs(end-20:end);                     %Test validation input vector
OUT_V1 = outputs(end-20:end);                   %Test Validation output Vector 

Train_seq_in = inputs(22 : end);               %Training inpur Data
Train_seq_out = outputs(22:end);               %Training output Data 

%build Neural network
options1 = trainingOptions('adam','MaxEpochs',2000,'InitialLearnRate',0.001,'MiniBatchSize',10,'Shuffle','every','L2Regularization',0,'Plots','training-progress','ValidationData',{IN_V1,OUT_V1})
layers =[sequenceInputLayer(107) fullyConnectedLayer(1024) reluLayer() fullyConnectedLayer(outputs_len) regressionLayer];

%train network 
 Net1 = trainNetwork(Train_seq_in,Train_seq_out,layers,options1);                  %Training the Network real

当我运行上面的代码时，我得到一个错误：

Error using trainNetwork (line 165)
The validation sequences are of feature dimension 21 but the input layer expects sequences of feature
dimension 107.

Error in quetion (line 24)
 Net1 = trainNetwork(Train_seq_in,Train_seq_out,layers,options1);                  %Training the Network
 real

但是我认为这不合逻辑，因为验证数据应该小于测试数据！！

我尝试将验证数据更改为与输入数据相同，如下所示：

%build Neural network
options1 = trainingOptions('adam','MaxEpochs',2000,'InitialLearnRate',0.001,'MiniBatchSize',10,'Shuffle','every','L2Regularization',0,'Plots','training-progress','ValidationData',{IN_V1,OUT_V1})
layers =[sequenceInputLayer(21) fullyConnectedLayer(1024) reluLayer() fullyConnectedLayer(outputs_len) regressionLayer];

%train network 
 Net1 = trainNetwork(IN_V1,IN_V1,layers,options1);                  %Training the Network real

但在这种情况下，我会收到另一个奇怪的错误，如下所示：

Error using trainNetwork (line 165)
To RESHAPE the number of elements must not change.

Error in quetion (line 24)
 Net1 = trainNetwork(IN_V1,IN_V1,layers,options1);                  %Training the Network real

Caused by:
    Error using reshape
    To RESHAPE the number of elements must not change.

请任何人都可以帮助我解决这个问题？

谢谢

【问题讨论】：

标签： matlab machine-learning neural-network

【解决方案1】：

对于第二个错误：使用 trainNetwork 时出错（第 165 行）要 RESHAPE，元素的数量不能改变。

Error in quetion (line 24)
 Net1 = trainNetwork(IN_V1,IN_V1,layers,options1);                  %Training the Network real

Caused by:
    Error using reshape
    To RESHAPE the number of elements must not change.

当你修改输入层的大小时，你也应该修改输出层的大小，所以你应该替换这一行：

layers =[sequenceInputLayer(21) fullyConnectedLayer(1024) reluLayer() fullyConnectedLayer(outputs_len) regressionLayer];

通过这一行：

layers =[sequenceInputLayer(21) fullyConnectedLayer(1024) reluLayer() fullyConnectedLayer(21) regressionLayer];

【讨论】：