Struct MultiTaskLoss 

pub struct MultiTaskLoss {
    pub task_losses: Vec<Box<dyn Loss>>,
    pub task_weights: Vec<f64>,
    pub strategy: TaskWeightingStrategy,
    pub weight_lr: f64,
    /* private fields */
}

Multi-task loss that combines multiple losses with configurable weighting.

Fields

task_losses: Vec<Box<dyn Loss>>

Individual task losses.

task_weights: Vec<f64>

Task weights (automatically managed based on strategy).

strategy: TaskWeightingStrategy

Weighting strategy.

weight_lr: f64

Learning rate for weight updates (used in some strategies).

Implementations

impl MultiTaskLoss

pub fn new_fixed( task_losses: Vec<Box<dyn Loss>>, task_weights: Vec<f64>, ) -> TrainResult<Self>

Create a new multi-task loss with fixed weights.

Arguments

• task_losses - Individual loss functions for each task
• task_weights - Fixed weights for each task (should sum to 1.0)

pub fn new_dynamic( task_losses: Vec<Box<dyn Loss>>, strategy: TaskWeightingStrategy, weight_lr: f64, ) -> TrainResult<Self>

Create a new multi-task loss with dynamic weighting.

Arguments

• task_losses - Individual loss functions for each task
• strategy - Weighting strategy to use
• weight_lr - Learning rate for weight updates

pub fn compute_multi_task( &mut self, predictions: &ArrayView<'_, f64, Ix2>, targets: &ArrayView<'_, f64, Ix2>, task_splits: &[usize], ) -> TrainResult<f64>

Compute multi-task loss.

Arguments

• predictions - Predictions for all tasks (concatenated)
• targets - Targets for all tasks (concatenated)
• task_splits - Column indices where each task starts

Returns

Weighted sum of task losses

pub fn get_weights(&self) -> &[f64]

Get current task weights.

pub fn num_tasks(&self) -> usize

Get number of tasks.