Load LSTM.onnx in Java

[brital]

I have created a lstm network in Matlab and converted it to onnx using matlabs exportONNXNetwork.
My network is a SeriesNetwork with
Layers: [5x1 nnet.cnn.layer.Layer]
layers =
Sequence Input Sequence input with 6 dimensions (numberOfFeatures)
LSTM LSTM with 50 hidden units
Fully connected 2 fully connected layers
Softmax softmax
Clasification Output crossentropyex

Sequence length is 24.

Creating Ortsession from this model:
{sequenceinput=NodeInfo(name=sequenceinput,info=TensorInfo(javaType=FLOAT,onnxType=ONNX_TENSOR_ELEMENT_DATA_TYPE_FLOAT,shape=[-1, -1, 6]))}

The OnnxTensor is created from float[nOfSequences][sequenceLength][sequenceInput=6 dim] :

var testData = new float[nOfSequences][sequenceLength][nOfFeatures];
OrtEnvironment env = OrtEnvironment.getEnvironment();
OnnxTensor onnxTensor = OnnxTensor.createTensor(env, testData);

Map<String, OnnxTensor> container = new HashMap<>();
SessionOptions options = new SessionOptions();
OrtSession session = env.createSession("lstm.onnx", options);
Map<String, NodeInfo> inMeta = session.getInputInfo();
String inputName = session.getInputNames().iterator().next();
container.put(inputName, onnxTensor);

try (Result result = session.run(container)) {}

This returns an onnxruntimeerror:
Non-zero status code returned while running LSTM node. Name:'lstm' Status Message: Input initial_h must have shape {1,24,50}. Actual:{1,1,50}

Where can I change the shape of initial_h?

[Craigacp]

If the only reported input is sequenceinput that sounds like it might be an issue with how the model was exported from MATLAB. Try running the ONNX model checker on it, and looking at the graph with Netron to see if there are any issues there.