Python wrappers
Public Python APIs implemented by this engine symbol.Collapse a contiguous range of axes into a single dimension.
Folds dimensions [start_axis, end_axis] (inclusive on both ends) into
one axis whose size is the product of the collapsed dimensions. Delegates
to reshape_op after computing the flattened size, so the input
must be contiguous along the affected range.
Shape
(D_0, ..., D_{s-1}, D_s, ..., D_e, D_{e+1}, ..., D_{n-1}) becomes
(D_0, ..., D_{s-1}, \prod_{d=s}^{e} D_d, D_{e+1}, ..., D_{n-1})
with s = start_axis and e = end_axis.
See Also
reshape_op, squeeze_all_op.
Parameters
aTensorImplPtrInput tensor.
start_axisintFirst axis in the range to collapse. Supports negative indexing.
end_axisintLast axis in the range (inclusive). Supports negative indexing. Must satisfy
start_axis <= end_axis after normalisation.Returns
TensorImplPtrTensor with rank ndim - (end_axis - start_axis) and a flattened dimension of size \prod_{d=start_axis}^{end_axis} a.shape[d] at position start_axis.
Raises
ShapeMismatchIf the collapsed range cannot be reshape-merged because of non-contiguous strides.