Struct ReLUFunction 

pub struct ReLUFunction;

ReLU function: f(x) = max(0, x) Subgradient: ∂f(x) = {0 if x < 0, [0, 1] if x = 0, 1 if x > 0}

Trait Implementations

impl Debug for ReLUFunction

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

Formats the value using the given formatter.

impl DynFunction for ReLUFunction

fn name(&self) -> &str

Name of the function for debugging and profiling

fn is_differentiable(&self) -> bool

Whether this function is differentiable

fn memory_complexity(&self) -> MemoryComplexity

Memory complexity hint for optimization

fn computational_complexity(&self) -> ComputationalComplexity

Computational complexity hint for optimization

fn is_fusable(&self) -> bool

Whether this function can be fused with others

fn metadata(&self) -> FunctionMetadata

Get function metadata for optimization

impl SerializableFunction for ReLUFunction

fn serialize(&self) -> Result<Vec<u8>>

Serialize the function to bytes

fn deserialize(data: &[u8]) -> Result<Box<dyn SerializableFunction>>

where Self: Sized,

Deserialize the function from bytes

fn format_version(&self) -> u32

Get the function’s serialization format version

fn validate(&self) -> Result<()>

Validate the function after deserialization

impl SubgradientFunction for ReLUFunction

fn forward<T>( &self, _ctx: &mut FunctionContext, inputs: &[&dyn AutogradTensor<T>], ) -> Result<Vec<Box<dyn AutogradTensor<T>>>>

where T: TensorElement + Float,

Forward pass computation

fn subgradient<T>( &self, _ctx: &mut FunctionContext, grad_outputs: &[&dyn AutogradTensor<T>], ) -> Result<Vec<Option<SubgradientSet<T>>>>

where T: TensorElement + Float,

Subgradient computation for non-differentiable operations Returns a set of possible subgradients