trainNetwork: Cell labels with file/signal data stores

Hi,

I have a datastore which consists of ~1000 files.

Each file is c*2 cell array, where c represents a number of events of intrest in the recorded data. The first colum of the cell array is a 9*12001 double of 9 features and the second column is a 1*1 categorical label for these events. (I will attach some sample data). This format seems to be suggested in Datastores for Deep Learning - MATLAB & Simulink - MathWorks Switzerland.

However, when I attempt to use the data to train a network in this format I get the following error;

Error using trainNetwork (line 184)
Invalid training data. For image, sequence-to-label, and feature classification tasks, responses
must be categorical.

I am asuiming this is because of the 1*1 categorical for each cell; however this is what seems to be suggested in Datastores for Deep Learning - MATLAB & Simulink - MathWorks Switzerland.

How can I get this data to read into the trainNetwork() function with either a signaldatastore() or filedatastore()?

Kind regards,

Christopher

[Shivam Malviya]

Hi Christopher,

I understand you want to load data through "datastores" for training a network using the "trainNetwork" function.

Could you share the following information for a better understanding of the issue;

• The file that creates the neural network to be trained.
• The precise steps to reproduce the issue.

Regards,

Shivam Malviya

[Christopher McCausland]

Hi Shivam,

I was able to solve the issue with a custom read function. If anyone reads this in the furture there were two issues;

1. Using the regular @load function via a handle loads the c*2 cell format in as a struct. This should have been expected, based on the behavour of the regular load command; however as the deep learning toolbox is expecting a cell array I was expecting it to bring in my data as a cell array. Writting a custom load function to bring in the prepocessed data is a soloution.
2. Number of categories, I have 36 labels, however not all are present in the data. The number of lables should match the total number of catigories avaliable in the categorical. Which in my case was 36. It's a good idea to remove any redundant categories too.

Kind Regards,

Christopher

[Shivam Malviya]

Glad to know the issue is resolved!