Struct neuronika::nn::ModelStatus

pub struct ModelStatus { /* fields omitted */ }

A model’s components status.

This struct should be used when you are interested in keeping track of the statuses and the parameters of the components that are part of a neural network. There are many circumstances in which this can be useful, such as when you have more than one model in a pipeline.

This struct stores all the Learnable associated to a given model and the model’s status. It is suggested to perform the registration of the layers at the model construction.

Implementations

impl ModelStatus

pub fn parameters(&self) -> Vec<Param<'_>>

Returns a vector of Param linked to the learnable weights associated to a neural network.

Conceptually, this method behaves similarly to .parameters() when called on the differentiable variable outputted by the network. The key difference lies in that, while the differentiable variable’s .parameters() method would return all the differentiable leaves that took part in the computation of the output, possibly also the weights of another network, ModelStatus’s .parameters() method returns only the leaves that have been associated with it at the model’s instantiation.

Usually the result of this method is passed to an optimizer.

pub fn register<T: Register>(&mut self, component: T) -> T

Registers a component.

Arguments

component - layer to be registered.

pub fn train(&self)

Sets the status in training mode.

pub fn eval(&self)

Sets the status in inference mode.

Trait Implementations

impl Default for ModelStatus

fn default() -> Self

Returns a new ModelStatus with empty parameters and status set to train.

impl Eval for ModelStatus

fn train(&self)

Sets the status to train.

fn eval(&self)

Sets the status to eval.