Struct StreamingKAN 

pub struct StreamingKAN {
    pub rolling_loss: f64,
    /* private fields */
}

Streaming KAN model with online B-spline learning.

Processes one sample at a time. Each edge in the network learns a univariate B-spline activation function via per-sample SGD. Online Welford normalization ensures stable input distribution.

Example

use irithyll::kan::{StreamingKAN, KANConfig};
use irithyll::StreamingLearner;

let config = KANConfig::builder()
    .layer_sizes(vec![3, 10, 1])
    .learning_rate(0.1)
    .build()
    .unwrap();
let mut model = StreamingKAN::new(config);
model.train(&[1.0, 2.0, 3.0], 4.0);
let pred = model.predict(&[1.0, 2.0, 3.0]);

Fields

rolling_loss: f64

EWMA of squared prediction error for uncertainty-modulated learning.

Implementations

impl StreamingKAN

pub fn new(config: KANConfig) -> Self

Create a new StreamingKAN from config.

pub fn config(&self) -> &KANConfig

Access the config.

pub fn n_layers(&self) -> usize

Number of KAN layers (edges between node layers).

pub fn n_params(&self) -> usize

Total number of learnable parameters (B-spline coefficients).

pub fn layer_sizes(&self) -> &[usize]

The layer sizes from the config.

pub fn input_importances(&self) -> Vec<f64>

Per-input feature importance from layer 0 B-spline coefficient magnitudes.

For each input i, sums |coefficient| across all output edges and all n_coeffs = grid_size + spline_order coefficients per edge. Inputs whose edges accumulate larger spline weight after training rank higher.

Returns un-normalized scores (length = layer_sizes()[0]); the caller should sum-normalize for display. Returns empty if the model has no layers (impossible in normal use — new() requires layer_sizes.len() >= 2).

Trait Implementations

impl Debug for StreamingKAN

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

Formats the value using the given formatter.

impl DiagnosticSource for StreamingKAN

fn config_diagnostics(&self) -> Option<ConfigDiagnostics>

Return config diagnostics, or None if not supported.

impl StreamingLearner for StreamingKAN

fn train_one(&mut self, features: &[f64], target: f64, weight: f64)

Train on a single observation with explicit sample weight.

fn predict(&self, features: &[f64]) -> f64

Predict the target for the given feature vector.

fn n_samples_seen(&self) -> u64

Total number of observations trained on since creation or last reset.

fn reset(&mut self)

Reset the model to its initial (untrained) state.

fn train(&mut self, features: &[f64], target: f64)

Train on a single observation with unit weight.

fn train_one_weighted(&mut self, features: &[f64], target: f64, weight: f64)

Train on a single observation with an explicit distillation weight.

fn predict_batch(&self, feature_matrix: &[&[f64]]) -> Vec<f64>

Predict for each row in a feature matrix.

impl Tunable for StreamingKAN

fn diagnostics_array(&self) -> [f64; 5]

Raw diagnostic signals for adaptive tuning.

fn adjust_config(&mut self, lr_multiplier: f64, lambda_delta: f64)

Apply smooth learning rate and regularization adjustments.