index_select_op

→TensorImplPtr

int index_select_op(const int & a, int dim, const int & indices)

source edit

Python wrappers

pylucid.ops.index_selectfunction

Pick indices.size slices along dim of a.

Composite over reshape_op + expand_op + gather_op. The 1-D index list is reshaped to rank a.ndim (size $k$ along dim, 1 elsewhere) and expanded to the source shape so gather_op's same-rank contract holds.

Math

For each output coordinate $(\ldots, j, \ldots)$ along dim:

y_{\ldots, j, \ldots} = a_{\ldots, \mathrm{idx}_j, \ldots}

See Also

take_op — flat indexing; narrow_op — contiguous window.

Parameters

aTensorImplPtr

Source tensor of rank

\ge 1

dimint

Axis to index along. Negative values wrap modulo a.ndim.

indicesTensorImplPtr

1-D integer tensor (int32 or int64) of length

k

Returns

TensorImplPtr

Tensor with the same shape as a except size $k$ along dim.

Raises

IndexError

If dim is out of range.

Failure

If indices is not 1-D or not an integer dtype.