Struct LearningRateFinder 

pub struct LearningRateFinder {
    pub history: Vec<(f64, f64)>,
    /* private fields */
}

Learning rate finder callback using the LR range test.

This callback implements the learning rate range test proposed by Leslie N. Smith. It gradually increases the learning rate from a minimum to a maximum value over a specified number of iterations/epochs and tracks the loss at each step.

The optimal learning rate is typically found just before the loss starts to increase.

Example

use tensorlogic_train::{LearningRateFinder, CallbackList};

let mut callbacks = CallbackList::new();
callbacks.add(Box::new(LearningRateFinder::new(
    1e-7,   // start_lr
    10.0,   // end_lr
    100,    // num_steps
)));

Fields

history: Vec<(f64, f64)>

History of (lr, loss) pairs.

Implementations

impl LearningRateFinder

pub fn new(start_lr: f64, end_lr: f64, num_steps: usize) -> Self

Create a new learning rate finder.

Arguments

• start_lr - Starting learning rate (e.g., 1e-7)
• end_lr - Ending learning rate (e.g., 10.0)
• num_steps - Number of steps to test

pub fn with_exponential_scaling(self) -> Self

Enable exponential scaling (recommended, default).

pub fn with_linear_scaling(self) -> Self

Enable linear scaling.

pub fn with_smoothing(self, smoothing: f64) -> Self

Set loss smoothing factor (0.0-1.0).

Recommended: 0.9 for noisy losses, 0.0 for smooth losses.

pub fn suggest_lr(&self) -> Option<f64>

Find the suggested optimal learning rate.

Returns the learning rate with the steepest negative gradient (fastest decrease in loss).

pub fn print_results(&self)

Print the LR finder results.

Trait Implementations

impl Callback for LearningRateFinder

fn on_batch_end( &mut self, _batch: usize, state: &TrainingState, ) -> TrainResult<()>

Called at the end of a batch.

fn should_stop(&self) -> bool

Check if training should stop early.

fn on_train_begin(&mut self, _state: &TrainingState) -> TrainResult<()>

Called at the beginning of training.

fn on_train_end(&mut self, _state: &TrainingState) -> TrainResult<()>

Called at the end of training.

fn on_epoch_begin( &mut self, _epoch: usize, _state: &TrainingState, ) -> TrainResult<()>

Called at the beginning of an epoch.

fn on_epoch_end( &mut self, _epoch: usize, _state: &TrainingState, ) -> TrainResult<()>

Called at the end of an epoch.

fn on_batch_begin( &mut self, _batch: usize, _state: &TrainingState, ) -> TrainResult<()>

Called at the beginning of a batch.

fn on_validation_end(&mut self, _state: &TrainingState) -> TrainResult<()>

Called after validation.