How to structure MultiDataSet for mixed data / multi-input model?

Hi all,

I need your help on how MultiDataSet must be structured to work with models similar to this one:
Multi-input

That is each data instance would have 2 feature vectors of diffrent length and one label vector.

I tried the following approach, but I get java.lang.IllegalStateException: Ragged array detected:

public static final MultiDataSet TRAINING_DATA = new MultiDataSet(
            Nd4j.createFromArray(new Float[][][]{
                    {
                            {1f, 2f, 3f, 4f, 5f, 6f, 7f},
                            {9f, 8f, 1f, 1f, 3f, 4f, 5f},
                            {2f, 3f, 4f, 5f, 6f, 8f, 9f}
                    },
                    {
                            {1f, 2f, 3f, 4f},
                            {5f, 6f, 7f, 8f},
                            {9f, 1f, 3f, 4f}
                    }
            }),
            Nd4j.createFromArray(new Integer[][]{
                    {0, 0, 1},
                    {1, 0, 1},
                    {0, 1, 0}
            })
    );

You are using the MultiDataSet constructor with single features/labels input.

Instead you want to use the constructor with multiple inputs, i.e. the one that takes INDArray[] inputs.