How to change the dimensions of the original dataset to get the same forecast value? LTSM ONNX

I trained an LSTM model in matlab

The original data set is 56 * 3853

feature_dimension is 56

miniBatchSize = 50;

[net,YPred] = predictAndUpdateState(net,xtrain,'ExecutionEnvironment','cpu');

normal operation

then...

filename = "lstm.onnx";

exportONNXNetwork(net,filename)

save data.mat xtrain

in python...

import onnxruntime

onnx_model = onnxruntime.InferenceSession('lstm.onnx')

from scipy import io

mydata = io.loadmat('data.mat')

data=np.float32(mydata['xtrain'])

onnx_input = {onnx_model.get_inputs()[0].name: data}

outputs = onnx_model.run(None, onnx_input)

InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Invalid rank for input: input Got: 2 Expected: 3 Please fix either the inputs or the model.

How to change the dimensions of the original dataset to get the same forecast value? LTSM ONNX

thanks a lot !!!

Answer 1

[Sivylla Paraskevopoulou]

You should permute from the MATLAB ordering (CN) to the ONNX ordering (NC), where C is the number of features and N is the number of sequence observations. Permute the input data before saving them to data.mat.

xtrain = permute(xtrain,[2,1]);

For more information about dimension ordering for different deep learning platforms, see Input Dimension Ordering.