Struct TensorLossFunction 

pub struct TensorLossFunction { /* private fields */ }

Common loss functions for tensor computations with gradient support.

Examples

use ipfrs_tensorlogic::loss_function::{TensorLossFunction, LossConfig, LossType, Reduction};

let config = LossConfig {
    loss_type: LossType::MSE,
    reduction: Reduction::Mean,
    ..LossConfig::default()
};
let mut loss_fn = TensorLossFunction::new(config);

let preds = vec![1.0, 2.0, 3.0];
let targets = vec![1.5, 2.5, 3.5];

let loss = loss_fn.forward(&preds, &targets).expect("example: should succeed in docs");
assert!((loss - 0.25).abs() < 1e-10); // mean of [0.25, 0.25, 0.25]

Implementations

impl TensorLossFunction

pub fn new(config: LossConfig) -> Self

Create a new loss function with the given configuration.

pub fn compute( &mut self, predictions: &[f64], targets: &[f64], ) -> Result<Vec<f64>, String>

Compute per-element loss values.

Returns an error if predictions and targets have different lengths.

pub fn reduce(&self, losses: &[f64]) -> f64

Apply the configured reduction to a slice of per-element losses.

pub fn forward( &mut self, predictions: &[f64], targets: &[f64], ) -> Result<f64, String>

Compute loss and reduce in one call.

pub fn gradient( &mut self, predictions: &[f64], targets: &[f64], ) -> Result<Vec<f64>, String>

Compute dL/d(prediction) per element.

Returns an error if predictions and targets have different lengths.

pub fn mse(a: &[f64], b: &[f64]) -> f64

Static helper: compute mean squared error between two slices.

Returns 0.0 if slices are empty or have different lengths.

pub fn mae(a: &[f64], b: &[f64]) -> f64

Static helper: compute mean absolute error between two slices.

Returns 0.0 if slices are empty or have different lengths.

pub fn stats(&self) -> LossFunctionStats

Return a snapshot of accumulated statistics.