using datastores for two inputs to trainnet

We are trying to use datastores as input to trainnet. The data consists of cells in the form of 3x90 arrays of doubles and the labels which are simple one character strings. The structure of the program is:

clabels=categorical(labels,["A" "N" "V" "S" "F" "U"]);
datastoreclabels=arrayDatastore(clabels);

data=cell(numSets,1);
for i =1:numSets
    data(i)= 3x90 array of double
end
datastoredata=arrayDatastore(data);

datastoredataF=combine(datastoredata,datastoreclabels);
net=trainnet(datastoredataF,layers,"crossentropy",options);

and got the error below, but note that

net=trainnet(data,clabels,layers,"crossentropy",options);  

did work.

The error was:

Error using trainnet (line 54)

Error forming mini-batch for network input "sequenceinput". Data interpreted with format "SCTB". To specify a different format, use the InputDataFormats option.

I am assuming that this was because the conversion from clabels to datastoreclables was incorrect but I don't see a better way.

Can anyone suggest what is missing?

[Matt J]

What do the 3 and the 90 represent?

[Mark Hubelbank]

The raw data is the analog output from three sensors. It was sampled generating 90 values for each sensor so the result is an array of data with the dimension of 3x90. The "labels" are the manually created "true" labels that we would like the network to ultimately produce. A datastore is being used because once this works, more data from other sensors in a different format will be added.

[Jayanti]

Can you please provide the complete code (including model architecture) and data you used, so that I can reproduce the issue?

Answer 1

[Gayathri]

Hi @Mark Hubelbank,

Please refer to the following answer which solves a similar issue as mentioned in the question.

https://www.mathworks.com/matlabcentral/answers/2145634-error-forming-mini-batch-for-network-input-input-data-interpreted-with-format-sscb-to-specify

I can assist you better once you provide the complete code.

Hope this helps!