Struct Trainer 

pub struct Trainer<'a> { /* private fields */ }

NeuralNetwork::train(..) builder. Allows optional setting of training hyperparameters.

Implementations

impl<'a> Trainer<'a>

pub fn evaluation_data( &mut self, evaluation_data: EvaluationData<'a>, ) -> &mut Trainer<'a>

Sets evaluation_data.

evaluation_data determines how to set the evaluation data.

pub fn cost(&mut self, cost: Cost) -> &mut Trainer<'a>

Sets cost.

cost determines cost function of network.

pub fn halt_condition( &mut self, halt_condition: HaltCondition, ) -> &mut Trainer<'a>

Sets halt_condition.

halt_condition sets after which Iteration/Duration or reached accuracy to stop training.

pub fn log_interval( &mut self, log_interval: MeasuredCondition, ) -> &mut Trainer<'a>

Sets log_interval.

log_interval sets some amount of Iterations/Duration to print the cost and accuracy of the neural net.

pub fn batch_size(&mut self, batch_size: Proportion) -> &mut Trainer<'a>

Sets batch_size.

pub fn learning_rate(&mut self, learning_rate: f32) -> &mut Trainer<'a>

Sets learning_rate.

pub fn l2(&mut self, lambda: f32) -> &mut Trainer<'a>

Sets lambda ($ \lambda $) for l2.

If $ \lambda $ set, implements L2 regularization with $ \lambda $ value.

pub fn early_stopping_condition( &mut self, early_stopping_condition: MeasuredCondition, ) -> &mut Trainer<'a>

Sets early_stopping_condition.

early_stopping_condition sets some amount of Iterations/Duration to stop after without notable cost improvement.

pub fn evaluation_min_change( &mut self, evaluation_min_change: Proportion, ) -> &mut Trainer<'a>

Sets evaluation_min_change.

Minimum change required to log positive evaluation change.

pub fn learning_rate_decay( &mut self, learning_rate_decay: f32, ) -> &mut Trainer<'a>

Sets learning_rate_decay.

learning_rate_decay is the mulipliers by which to decay the learning rate.

pub fn learning_rate_interval( &mut self, learning_rate_interval: MeasuredCondition, ) -> &mut Trainer<'a>

Sets learning_rate_interval.

pub fn checkpoint_interval( &mut self, checkpoint_interval: MeasuredCondition, ) -> &mut Trainer<'a>

Sets checkpoint_interval.

checkpoint_interval sets how often (if at all) to serialize and output neural network to .txt file.

pub fn name(&mut self, name: &'a str) -> &mut Trainer<'a>

Sets name

name sets what to pretend to checkpoint files. Used to differentiate between nets when checkpointing multiple.

pub fn tracking(&mut self) -> &mut Trainer<'a>

Sets tracking.

tracking determines whether to output percentage progress during backpropgation.

pub fn go(&mut self)

Begins training.