Repeat the contents of a tensor along a single axis.

Produces an output whose axis dimension is repeats times larger than the input's; every other dimension is unchanged. Each element along axis is replicated as a contiguous block of repeats copies in the output (this is interleaved repetition, distinct from tile_op).

Math

$y_{[\ldots, i_a, \ldots]} = x_{[\ldots, \lfloor i_a / r \rfloor, \ldots]}$ where $r =$ repeats and $i_a$ is the index along axis.

Shape

out.shape[axis] = a.shape[axis] * repeats; all other dims unchanged.

See Also

tile_op — multi-axis tiling; different element ordering.