class

TanBackward

TanBackward

source edit

Python wrappers

pylucid.ops.tanfunction

Autograd node for element-wise tangent $y = \tan(x)$ .

Saves the input x so the backward pass can build $\cos^2(x)$ in the denominator. Defined for all real $x$ except the singular points $x = (k + \tfrac{1}{2})\pi$ where $\cos x = 0$ ; behaviour at those points is undefined.

Math

\frac{\partial y}{\partial x} = \sec^2(x) = \frac{1}{\cos^2(x)}, \qquad \frac{\partial \mathcal{L}}{\partial x} = \frac{1}{\cos^2(x)}\,\frac{\partial \mathcal{L}}{\partial y}.$$

Attributes

schema_v1OpSchema

Registered as "tan" with AmpPolicy::Promote.

Notes

Dispatch: Accelerate vvtanf/vvtan (CPU) / MLX tan (GPU).

Static methods

static

dispatch

int dispatch(backend::IBackend & be, const int & a, const int & s, Dtype dt)

source edit

Forward — calls IBackend::tan to compute $y = \tan x$ .

Methods

grad_formula

int grad_formula(const int & g)

source edit

Backward — $\partial y/\partial x = \sec^2 x$ , scaled by grad_out.

class

TanBackward

TanBackward

source edit

Python wrappers

pylucid.ops.tanfunction

Autograd node for element-wise tangent $y = \tan(x)$ .

Math

\frac{\partial y}{\partial x} = \sec^2(x) = \frac{1}{\cos^2(x)}, \qquad \frac{\partial \mathcal{L}}{\partial x} = \frac{1}{\cos^2(x)}\,\frac{\partial \mathcal{L}}{\partial y}.$$

Attributes

schema_v1OpSchema

Registered as "tan" with AmpPolicy::Promote.

Notes

Dispatch: Accelerate vvtanf/vvtan (CPU) / MLX tan (GPU).

Static methods

static

dispatch

int dispatch(backend::IBackend & be, const int & a, const int & s, Dtype dt)

source edit

Forward — calls IBackend::tan to compute $y = \tan x$ .

Methods

grad_formula

int grad_formula(const int & g)

source edit

Backward — $\partial y/\partial x = \sec^2 x$ , scaled by grad_out.