Trait neuronika::Forward

pub trait Forward {
    fn forward(&self);
    fn was_computed(&self) -> bool;
    fn reset_computation(&self);
}

Forward-propagation behavior.

This trait is implemented by all the internal forward components of Var and VarDiff.

The main method it provides is the .forward() method that is used to propagate computations from the leaf variables to the graph’s root.

The other two methods, namely .was_computed() and .reset_computation(), are used to perform caching during the forward pass. Caching is critical to avoid recomputing paths and to achieve good performance when a computational graph has more than one root, like the one, for instance, of a recurrent neural network.

Required methods

fn forward(&self)

Propagates the computations forwards.

It also defines the logic for the computation of the node.

fn was_computed(&self) -> bool

Returns true if the node was computed, false otherwise.

fn reset_computation(&self)

Reset the node’s flag, making it computable again.

Implementors