Struct finalfrontier::SGD

pub struct SGD<T> { /* fields omitted */ }

Stochastic gradient descent

This data type applies stochastic gradient descent on sentences.

Methods

impl<T> SGD<T> where
    T: Trainer, 

pub fn into_model(self) -> TrainModel<T>

pub fn new(model: TrainModel<T>) -> Self

Construct a new SGD instance,

pub fn model(&self) -> &TrainModel<T>

Get the training model associated with this SGD.

pub fn n_tokens_processed(&self) -> usize

Get the number of tokens that are processed by this SGD.

pub fn train_loss(&self) -> f32

Get the average training loss of this SGD.

This returns the average training loss over all instances seen by this SGD instance since its construction.

pub fn update_sentence<'b, S>(&mut self, sentence: &S, lr: f32) where
    S: ?Sized,
    T: TrainIterFrom<'b, S> + Trainer + NegativeSamples,
    &'a T::Focus: IntoIterator<Item = u64>,
    T::Focus: WordIdx, 

Update the model parameters using the given sentence.

This applies a gradient descent step on the sentence, with the given learning rate.

Trait Implementations

impl<T: Clone> Clone for SGD<T>

fn clone(&self) -> SGD<T>

Returns a copy of the value.

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

Performs copy-assignment from source.