input shape for 3d tensor

[gotostereo]

Hello, i have data in 3d array

tf.tensor3d

Tensor

    [[[0 , 0.5 ],

      [1 , 1.5 ],

      [2 , 2.5 ],

      [3 , 3.5 ],

      [4 , 4.5 ]],

     [[5 , 5.5 ],

      [6 , 6.5 ],

      [7 , 7.5 ],

      [8 , 8.5 ],

      [9 , 9.5 ]],

     [[10, 10.5],

      [11, 11.5],

      [12, 12.5],

      [13, 13.5],

      [14, 14.5]],

     [[15, 15.5],

      [16, 16.5],

      [17, 17.5],

      [18, 18.5],

      [19, 19.5]],

     [[20, 20.5],

      [21, 21.5],

      [22, 22.5],

      [23, 23.5],

      [24, 24.5]]]

I don't set input shape, try 

        this.model.add( tf.layers.dense( { 

            units: 5, 

            inputShape: [[2]],

            activation: 'linear'

        } ) );

Error: Tensor must have a shape comprised of positive integers but got shape [2,5].

How fix it?

[Mark Strefford]

Dense layers only take a 1D tensor, for example of shape [5]. I'm assuming in your example that you have 5 training examples there, each of shape [2, 5]. If this is the case, you'll need to flatten / reshape each example to be shape [10] and build your dense layer accordingly.

[Mark Strefford]

See https://js.tensorflow.org/api/latest/#layers.dense

• inputDim (number) If specified, defines inputShape as [inputDim].