Struct ContinualLearningTrainer 

pub struct ContinualLearningTrainer<M: Model> {
    pub model: M,
    pub config: ContinualLearningConfig,
    pub memory: MemoryBuffer,
    pub task_info: HashMap<usize, TaskInfo>,
    pub current_task: Option<usize>,
    pub fisher_matrices: HashMap<String, Tensor>,
    pub optimal_parameters: HashMap<String, Tensor>,
    pub step_counter: usize,
    pub task_detector: Option<TaskDetector>,
}

Continual learning trainer

Fields

model: M

The model being trained

config: ContinualLearningConfig

Configuration

memory: MemoryBuffer

Memory buffer for experience replay

task_info: HashMap<usize, TaskInfo>

Task-specific information

current_task: Option<usize>

Current task ID

fisher_matrices: HashMap<String, Tensor>

Fisher information matrices (for EWC)

optimal_parameters: HashMap<String, Tensor>

Optimal parameters (for EWC)

step_counter: usize

Training step counter

task_detector: Option<TaskDetector>

Task detection state

Implementations

impl<M: Model<Input = Tensor, Output = Tensor>> ContinualLearningTrainer<M>

pub fn new(model: M, config: ContinualLearningConfig) -> Result<Self>

Create a new continual learning trainer

pub fn start_task(&mut self, task_id: usize) -> Result<()>

Start learning a new task

pub fn learn_batch( &mut self, inputs: &[Tensor], targets: &[Tensor], task_id: Option<usize>, ) -> Result<ContinualLearningOutput>

Learn from a batch of data

pub fn finalize_task(&mut self, task_id: usize) -> Result<()>

Finalize learning for a task

pub fn evaluate_all_tasks(&self) -> Result<HashMap<usize, TaskEvaluation>>

Evaluate on all tasks

pub fn get_metrics(&self) -> ContinualLearningMetrics

Get continual learning metrics