I’m making a Reshape layer, and in the constructor, one parameter is
shape: The name to use for the 1D shape tensor of the layer
So I’m creating a separate tensor, which this layer will reference by name (right?)
Does this mean, it’ll use that tensors shape for the reshaping?
Or, will it use it as a data container, and pull the desired output shape out of the tensor’s inner numerical data?
Like if I’m going from [1,16,4,4] input, to [1, 256] output, I should be creating a one dimensional Tensor with two elements, 1 and 256?
Or should I create a [1, 256] tensor filled with zeroes?