Struct wyrm::ParameterNode

pub struct ParameterNode { /* fields omitted */ }

Parameter node, holds the optimizable parameters of the model.

Methods

impl ParameterNode

fn shared(value: Arc<HogwildParameter>) -> Variable<Self>

Create a parameter node that shares its parameter values with other parameter nodes via the HogwildParameter object.

fn new(value: Arr) -> Variable<Self>

Create a new parameter node. The parameters held by this node cannot be shared and optimized in parallel.

fn zero_gradient(&self)

Zero the accumulated gradients of this node.

Trait Implementations

impl Debug for ParameterNode

fn fmt(&self, __arg_0: &mut Formatter) -> Result

Formats the value using the given formatter.

impl Node for ParameterNode

type Value = Arr

Type of the node's value.

type InputGradient = Arr

Type of the input gradient the node receives during backpropagation.

type OutputGradient = ()

Type of the gradient the node passes down to its ancestors during backpropagation.

fn forward(&self)

Perform the forward step. Should recursively call the forward methods of its ancestors.

fn backward(&self, gradient: &Ref<Self::InputGradient>)

Perform the backward step. Should recursively call the backward methods of its ancestors.

fn value(&self) -> Bor<Self::Value>

Return the value of the node.

fn needs_gradient(&self) -> bool

If the node needs to be used in the backward step.