TFLiteTensor.dims vs out_tensor.shape about tflite_beam HOT 2 CLOSED

Hi @mnishiguchi, I intend to also port TFLite in Erlang so I'll probably keep them for the completeness of the APIs for now. And I'll revisit this issue when the Erlang part is done. :)

from tflite_beam.

out_tensor.type
# vs
type = TFLite.TFLiteTensor.type(out_tensor)

I think this should be okay as we also have the same thing in :nx.

iex> tensor = Nx.iota({2,2})
iex> tensor.type
{:s, 64}
iex> Nx.type(tensor)
{:s, 64}

As for TFLiteTensor.dims/1 vs tensor.shape, I'm thinking that keeping both, but TFLiteTensor.dims/1 will behave the same as TFLite's C++ API, which gives tensor's dimension in an array, e.g., [1, 224, 224, 3]; while tensor.shape will follow :nx's convention, which would return the dimensions in a tuple, e.g., {1, 224, 224, 3}. Furthermore, I'll add a TFLiteTensor.shape/1 function to mirror what we have in :nx.

iex> tflite_tensor.type
{:s, 64}
iex> TFLiteTensor.type(tflite_tensor)
{:s, 64}
iex> tflite_tensor.shape
{1, 224, 224, 3}
iex> TFLiteTensor.shape(tflite_tensor)
{1, 224, 224, 3}
iex> TFLiteTensor.dims(tflite_tensor)
[1, 224, 224, 3]

from tflite_beam.