Struct Ranger 

pub struct Ranger<T: Float + ScalarOperand> { /* private fields */ }

Ranger optimizer configuration

Ranger combines RAdam (Rectified Adam) with Lookahead mechanism. This standalone implementation integrates both algorithms efficiently.

Key Features

• Fast convergence from RAdam’s variance rectification
• Stability from Lookahead’s slow weight trajectory
• Reduced sensitivity to hyperparameter choices
• Better generalization than Adam or RAdam alone

Type Parameters

• T: Floating-point type (f32 or f64)

Implementations

impl<T: Float + ScalarOperand> Ranger<T>

pub fn new( learning_rate: T, beta1: T, beta2: T, epsilon: T, weight_decay: T, lookahead_k: usize, lookahead_alpha: T, ) -> Result<Self>

Create a new Ranger optimizer

Arguments

• learning_rate: Learning rate for RAdam (typically 0.001)
• beta1: First moment decay rate (typically 0.9)
• beta2: Second moment decay rate (typically 0.999)
• epsilon: Small constant for numerical stability (typically 1e-8)
• weight_decay: L2 regularization coefficient (typically 0.0)
• lookahead_k: Number of fast updates per slow update (typically 5-6)
• lookahead_alpha: Interpolation factor for slow weights (typically 0.5)

Example

use optirs_core::optimizers::Ranger;

let optimizer = Ranger::<f32>::new(
    0.001,  // learning_rate
    0.9,    // beta1
    0.999,  // beta2
    1e-8,   // epsilon
    0.0,    // weight_decay
    5,      // lookahead_k
    0.5     // lookahead_alpha
).unwrap();

pub fn step( &mut self, params: ArrayView1<'_, T>, grads: ArrayView1<'_, T>, ) -> Result<Array1<T>>

Perform a single optimization step

Combines RAdam (fast weights) with Lookahead (slow weights)

Example

use optirs_core::optimizers::Ranger;
use scirs2_core::ndarray_ext::array;

let mut optimizer = Ranger::<f32>::default();
let params = array![1.0, 2.0, 3.0];
let grads = array![0.1, 0.2, 0.3];

let updated_params = optimizer.step(params.view(), grads.view()).unwrap();

pub fn step_count(&self) -> usize

Get the number of optimization steps performed

pub fn slow_update_count(&self) -> usize

Get the number of slow weight updates performed

pub fn reset(&mut self)

Reset the optimizer state

pub fn slow_weights(&self) -> Option<&Array1<T>>

Get the slow weights (Lookahead trajectory)

pub fn is_rectified(&self) -> bool

Check if variance rectification is active

Trait Implementations

impl<T: Clone + Float + ScalarOperand> Clone for Ranger<T>

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

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<T: Debug + Float + ScalarOperand> Debug for Ranger<T>

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<T: Float + ScalarOperand> Default for Ranger<T>

fn default() -> Self

Returns the “default value” for a type.

impl<'de, T> Deserialize<'de> for Ranger<T>

where T: Deserialize<'de> + Float + ScalarOperand,

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl<T> Serialize for Ranger<T>

where T: Serialize + Float + ScalarOperand,

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.