Use Convolution network for 1D data - regression to regression

Hello

I am trying to use Convolution neural network for time series regression problem using MATLAB.

Below is my problem definition

Two time series inputs (signals)

A(t) = vector (1XN)

B(t) = vector (1XN)

2. One time series output (signal)

C(t) = vector (1XN)

3. 1000 sample inputs/outputs

A1(t), A2(t), A3(t), ……………………… A1000(t)

B1(t), B2(t), B3(t), ……………………… B1000(t)

C1(t), C2(t), C3(t), ……………………… C1000(t)

4. trainedNet = trainNetwork( X, Y, layers, options)

How do I set up X and Y arrays.

Per Matlab help “X is specified as a 4-D numeric array. The first three dimensions are the height, width, and channels, and the last dimension indexes the individual images.”

Matlab recommends use of reshape command but which array can be reshaped.

Height = 2;
Width = N; 
Channels = 1;
Sample = M; 

X_tmp = ……
X = reshapre(X_tmp, [height, width, channels, sampleSize])  ;   % but how to setup X_tmp

How to set up Y

N-by-1 cell array of numeric sequences, where N is the number of observations. The sequences are matrices with r rows, where r is the number of responses. Not clear from the description.

How about layers to be used for network --

layers =

    convolution2dLayer(R,S,'Padding','same')
    reluLayer
      ............
    maxPooling2dLayer(2,'Stride',2)   
    fullyConnectedLayer(1)
    regressionLayer];

Answer 1

[Walter Roberson]

X = cat(4, A{:});

However, I doubt that the tool will be willing to work with 1D data.