Invalid training data. Predictors must be a numeric array, a datastore, or a table. For networks with sequence input, predictors can also be a cell array of sequences.

cvp = cvpartition(data.Var4,"Holdout",0.3);
dataTrain = data(training(cvp),:);
dataValidation = data(test(cvp),:);

%% Traning
%preprocess data

documentsTrain = preprocessText(dataTrain.Var3); %diagnosis
%%3 categoreis will be trained.
TTrain = categorical(dataTrain.Var4); % result
classNames = unique(TTrain); 
numObservations = numel(TTrain);%60

%% validation

documentsValidation = preprocessText(dataValidation.Var3); %diagnosis
TValidation = categorical(dataValidation.Var4);% result
enc = wordEncoding(documentsTrain);
numWords = enc.NumWords;
XTrain = doc2sequence(enc,documentsTrain);
XValidation = doc2sequence(enc,documentsValidation);
embeddingDimension = 100;
ngramLengths = [2 3 4 5];
numFilters = 200;

minLength = min(doclength(documentsTrain));
layers = [ 
    sequenceInputLayer(1,MinLength=minLength)
    wordEmbeddingLayer(embeddingDimension,numWords,Name="numwords")];
lgraph = layerGraph(layers);

numBlocks = numel(ngramLengths);
for j = 1:numBlocks
    N = ngramLengths(j);
    
    block = [
        convolution1dLayer(N,numFilters,Name="conv"+N,Padding="same")
        batchNormalizationLayer(Name="bn"+N)
        reluLayer(Name="soft"+N)
        dropoutLayer(0.2,Name="drop"+N)
        globalMaxPooling1dLayer(Name="max"+N)];
    
    lgraph = addLayers(lgraph,block);
    lgraph = connectLayers(lgraph,"numwords","conv"+N);
end


numClasses = numel(classNames);

layers = [
    concatenationLayer(1,numBlocks,Name="cat")
    fullyConnectedLayer(numClasses,Name="fc")
    softmaxLayer(Name="soft")
    classificationLayer(Name="classification")];

lgraph = addLayers(lgraph,layers);


for j = 1:numBlocks
    N = ngramLengths(j);
    lgraph = connectLayers(lgraph,"max"+N,"cat/in"+j);
end

figure
plot(lgraph)
title("New")

options = trainingOptions("adam", ...
    MiniBatchSize=128, ...
    ValidationData={XValidation,TValidation}, ...
    OutputNetwork="best-validation-loss", ...
    Plots="training-progress", ...
    Verbose=false);
net = trainNetwork (TTrain,XTrain,lgraph,options);

[neetha thomas]

Cant train the work

Answer 1

[Prasanna]

Hi Neeta,

The error message you are getting indicates training data provided to the ‘trainNetwork’ function is not in the expected format. Specifically, the predictors (input data) must be a numeric array, a datastore, or a table. Some steps to debug and resolve the issue are:

• Check the input data format and ensure that the ‘XTrain’ variable is a cell array of sequences. Each element of the cell array should be a numeric vector representing the sequence for a single observation.
• Correct the ‘trainNetwork’ function call in your script. The ‘trainNetwork’ function expects the input data (predictors) to be the first argument, followed by the target data (responses). In your code, you have swapped the order. The correct order is ‘trainNetwork(XTrain, TTrain, lgraph, options)’.

Also, ensure that the preprocessing steps, such as ‘preprocessText’ and ‘doc2sequence’, correctly transform the text data into sequences that can be input into the network. By following these steps and ensuring the correct data format and function usage, you should be able to resolve the error and successfully train your network. For more information regarding the error, refer the following documentations:

• trainNetwork: https://www.mathworks.com/help/deeplearning/ref/trainnetwork.html
• Error message resolution “Predictors must be a numeric array, a datastore, or a table”: https://www.mathworks.com/matlabcentral/answers/1908625-predictors-must-be-a-numeric-array-a-datastore-or-a-table
• Error message resolution “Invalid training data. Predictors must be a N-by-1 cell array of sequences, where N is the number of sequences.”: https://www.mathworks.com/matlabcentral/answers/393034-i-am-unable-to-resolve-this-error-invalid-training-data-predictors-must-be-a-n-by-1-cell-array-of

Hope this helps!