narrow_op

→TensorImplPtr

int narrow_op(const int & a, int dim, int start, int length)

source edit

Python wrappers

pylucid.ops.narrowfunction

Slice a contiguous window [start, start + length) along dim.

Composite over split_at_op. Returns the input unchanged when the window covers the full axis (zero-cost fast path). Gradient flows through SplitSliceBackward.

Math

y_{\ldots, j, \ldots} = a_{\ldots, j + \mathrm{start}, \ldots}, \qquad 0 \le j < \mathrm{length}

See Also

split_at_op — underlying primitive.

Parameters

aTensorImplPtr

Source tensor of rank

\ge 1

dimint

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

startint64

Inclusive start index along dim. Must be non-negative.

lengthint64

Window length. Must satisfy start + length <= a.shape[dim].

Returns

TensorImplPtr

Tensor with the same shape as a except size length along dim. When the window covers the full axis, the input is returned by identity.

Raises

IndexError

If dim is out of range or the window falls outside a.