trunc_op

→TensorImplPtr

int trunc_op(const int & a)

source edit

Python wrappers

pylucid.ops.truncfunction

Round toward zero: $y = \mathrm{trunc}(x) = \mathrm{sgn}(x) \lfloor |x| \rfloor$ .

Composite over greater_equal_op + where_op + floor_op + ceil_op. Picks floor for non-negative inputs and ceil for negatives. Both branches are piecewise constant, so the gradient contribution is zero (matches the reference framework's convention).

Math

y = \begin{cases} \lfloor x \rfloor & x \ge 0 \\ \lceil x \rceil & x < 0 \end{cases}, \qquad \frac{\partial y}{\partial x} = 0

See Also

floor_op, ceil_op, frac_op.

Parameters

aTensorImplPtr

Input tensor.

Returns

TensorImplPtr

Tensor of the same shape and dtype as a.