Error exporting trained neural network model using ONNX to onnx format

Hello everyone,

I have trained a model using the Deep Learning Toolbox in MATLAB 2020b. From the toolbox, I exported the following network to the workspace:

net =

    Neural Network
 
              name: 'Function Fitting Neural Network'
          userdata: (your custom info)
 
    dimensions:
 
         numInputs: 1
         numLayers: 2
        numOutputs: 1
    numInputDelays: 0
    numLayerDelays: 0
 numFeedbackDelays: 0
 numWeightElements: 157
        sampleTime: 1
 
    connections:
 
       biasConnect: [1; 1]
      inputConnect: [1; 0]
      layerConnect: [0 0; 1 0]
     outputConnect: [0 1]
 
    subobjects:
 
             input: Equivalent to inputs{1}
            output: Equivalent to outputs{2}
 
            inputs: {1x1 cell array of 1 input}
            layers: {2x1 cell array of 2 layers}
           outputs: {1x2 cell array of 1 output}
            biases: {2x1 cell array of 2 biases}
      inputWeights: {2x1 cell array of 1 weight}
      layerWeights: {2x2 cell array of 1 weight}
 
    functions:
 
          adaptFcn: 'adaptwb'
        adaptParam: (none)
          derivFcn: 'defaultderiv'
         divideFcn: 'dividerand'
       divideParam: .trainRatio, .valRatio, .testRatio
        divideMode: 'sample'
           initFcn: 'initlay'
        performFcn: 'mse'
      performParam: .regularization, .normalization
          plotFcns: {'plotperform', 'plottrainstate', 'ploterrhist',
                    'plotregression', 'plotfit'}
        plotParams: {1x5 cell array of 5 params}
          trainFcn: 'trainbr'
        trainParam: .showWindow, .showCommandLine, .show, .epochs,
                    .time, .goal, .min_grad, .max_fail, .mu, .mu_dec,
                    .mu_inc, .mu_max
 
    weight and bias values:
 
                IW: {2x1 cell} containing 1 input weight matrix
                LW: {2x2 cell} containing 1 layer weight matrix
                 b: {2x1 cell} containing 2 bias vectors
 
    methods:
 
             adapt: Learn while in continuous use
         configure: Configure inputs & outputs
            gensim: Generate Simulink model
              init: Initialize weights & biases
           perform: Calculate performance
               sim: Evaluate network outputs given inputs
             train: Train network with examples
              view: View diagram
       unconfigure: Unconfigure inputs & outputs
 
    evaluate:       outputs = net(inputs)
    
    

To export the above network to ONNX format, I am using the following add-on:

https://www.mathworks.com/matlabcentral/fileexchange/67296-deep-learning-toolbox-converter-for-onnx-model-format

However, when I try to export the model, it throws the following error:

    Error using nnet.internal.cnn.onnx.exportONNXNetwork>iValidateNetwork (line 62)
First argument must be a SeriesNetwork, DAGNetwork, dlnetwork, or layerGraph.

Error in nnet.internal.cnn.onnx.exportONNXNetwork>iValidateInputs (line 53)
NNTNetwork  = iValidateNetwork(NNTNetwork);

Error in nnet.internal.cnn.onnx.exportONNXNetwork (line 29)
[NNTNetwork, Filename, NetworkName, OpsetVersion] = iValidateInputs(NNTNetwork, Filename, defaultOpset, varargin{:});

Error in exportONNXNetwork (line 40)
nnet.internal.cnn.onnx.exportONNXNetwork(Network, filename, varargin{:});

It seems that the add-on does not recognize the format of the network, even though the network should be a series network since it is a simple multi-layer perceptron. Is there any workaround this? I do not understand how else to export the model otherwise. I am trying to export it to ONNX format so that it can be used in Python.

Any help would be much appreciated,

Thank You.

Answer 1

[Shashank Gupta]

Hi faraz,

you are using shallow neural network and this network does not belongs to any of the above network datatype and thus you are getting the error. MATLAB as of now does not have a straightforward way to convert shallow neural networks to ONNX. exportONNXNetwork function only applies for deep networks APIs. But there are some workaround I can suggest. My first suggestion would be to re-create the shallow neural network using trainNetwork workflow, then give that network to exportONNXNetwork function. Secondly, if you are looking for deployment purpose only then you there are some other reference you can use too. It is not specific to ONNX. Link is here.

I hope this helps you.

Cheers.