LSTM network error: Predictors and responses must have the same number of observations.

I am trying to use an LSTM neural network to output a number based on two separate time series. I give the time series in pairs, based on which the network should output a single number.

Here is a code example to reproduce the error message:

clc, clear all;

% Training data, series in pairs. There is a pair of series in each cell
% element.
seq = {[1 2 3; 2 3 4];...
       [0 1 2; 4 5 6]};

% 2 channels, since the series are in pairs.
numChannels = 2;
numHiddenUnits = 20;
% I would like to have a single response from the network.
numResponses = 1;

% Basic LSTM network architecture, taken from: 
% https://www.mathworks.com/help/deeplearning/ug/sequence-to-one-regression-using-deep-learning.html
layers = [sequenceInputLayer(numChannels, Normalization="zscore")
          lstmLayer(numHiddenUnits, OutputMode = "last")
          fullyConnectedLayer(numResponses)
          regressionLayer];

% Output targets for training. 2 outputs for 2 pairs of sequences.
target = [1 2];

% Arbitrarily chosen data for validation.
seqValidation = {[1 2 3; 2 3 4]*2; [0 1 2; 4 5 6]*3};
targetValidation = [2 6];

options = trainingOptions("adam", ...
    MaxEpochs = 250, ...
    ValidationData = {seqValidation targetValidation}, ...
    OutputNetwork = "best-validation-loss", ...
    InitialLearnRate = 0.005, ...
    SequenceLength = "shortest", ...
    Plots = "training-progress", ...
    Verbose = false);

net = trainNetwork(seq, target, layers, options);

When trying to run the script I am getting the following error (script is called LSTMtry.m):

Error using trainNetwork

Invalid training data. Predictors and responses must have the same number of observations.

Error in LSTMtry (line 34)

net = trainNetwork(seq, target, layers, options);

Answer 1

[Vinayak Choyyan]

Hello hoaradam0207,

As per my understanding, you are trying to train a neural network model and have 2 times series as input. I see that you have given this as

seq = {[1 2 3; 2 3 4];...
    [0 1 2; 4 5 6]};
size(seq)

ans = 1×2

2 1

This is a cell array of size 2x1. The output is a single number. You have given this as

target = [1 2];
size(target)

ans = 1×2

1 2

This is of size 1x2. Please refer to the below code. I have changed the shape of target and targetValidation and the error does not occur. The model does train.

clc, clear all;
% Training data, series in pairs. There is a pair of series in each cell
% element.
seq = {[1 2 3; 2 3 4];...
    [0 1 2; 4 5 6]};
% 2 channels, since the series are in pairs.
numChannels = 2;
numHiddenUnits = 20;
% I would like to have a single response from the network.
numResponses = 1;
% Basic LSTM network architecture, taken from: 
% https://www.mathworks.com/help/deeplearning/ug/sequence-to-one-regression-using-deep-learning.html
layers = [sequenceInputLayer(numChannels, Normalization="zscore")
    lstmLayer(numHiddenUnits, OutputMode = "last")
    fullyConnectedLayer(numResponses)
    regressionLayer];
% Output targets for training. 2 outputs for 2 pairs of sequences.
target = [1;2];
% Arbitrarily chosen data for validation.
seqValidation = {[1 2 3; 2 3 4]*2; [0 1 2; 4 5 6]*3};
targetValidation = [2;6];
options = trainingOptions("adam", ...
    MaxEpochs = 250, ...
    ValidationData = {seqValidation targetValidation}, ...
    OutputNetwork = "best-validation-loss", ...
    InitialLearnRate = 0.005, ...
    SequenceLength = "shortest", ...
    Plots = "training-progress", ...
    Verbose = false);
net = trainNetwork(seq, target, layers, options);

I hope this resolves the issue you were facing.