Deeplearning4j: libnd4j: add broadcast_to op

[AlexDBlack]

This is related to broadcast copy (which expects 1 input array and 1 output array), and broadcast_dynamic_shape (which expects 2 1d arrays representing shapes as input).
Here, we have a input 0 is array to be broadcast, and input 1 is the shape (represented as a 1d NDArray) that it should be broadcast to.

https://www.tensorflow.org/api_docs/python/tf/broadcast_to

[raver119]

We had tile_to_shape op. However it was broken and forgotten since long shapes migration.
It does exactly this: takes input array and shape, and returns output with values from input tiled to new shape.

p.s. we also should add option to take shape as second argument

[shyrma]

Done!
https://github.com/deeplearning4j/deeplearning4j/blob/master/libnd4j/include/ops/declarable/generic/shape/broadcast_to.cpp

[AlexDBlack]

Thanks! Mapped at java level and tests confirmed passing.

[lock[bot]]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.