Struct GradientCheckConfig 

pub struct GradientCheckConfig {
    pub epsilon: f64,
    pub rtol: f64,
    pub atol: f64,
    pub check_elementwise: bool,
    pub use_central_difference: bool,
    pub max_samples: Option<usize>,
    pub random_seed: Option<u64>,
}

Configuration for numerical gradient checking

Fields

epsilon: f64

Epsilon for finite difference computation

rtol: f64

Relative tolerance for gradient comparison

atol: f64

Absolute tolerance for gradient comparison

check_elementwise: bool

Check gradients element-wise (slower but more detailed)

use_central_difference: bool

Use central differences (more accurate but 2x slower)

max_samples: Option<usize>

Maximum number of samples to check (for large tensors)

random_seed: Option<u64>

Random seed for sampling (if max_samples is set)

Implementations

impl GradientCheckConfig

pub fn strict() -> Self

Create a configuration for strict checking (tighter tolerances)

pub fn relaxed() -> Self

Create a configuration for relaxed checking (looser tolerances)

pub fn fast() -> Self

Create a configuration for fast checking (forward differences, sampling)

Trait Implementations

impl Clone for GradientCheckConfig

fn clone(&self) -> Self

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for GradientCheckConfig

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for GradientCheckConfig

fn default() -> Self

Returns the “default value” for a type.