Skip to main content

MetaLearner

tensorlogic_train

Trait MetaLearner 

Source 
pub trait MetaLearner {
    // Required methods
    fn meta_step(
        &self,
        tasks: &[MetaTask],
        parameters: &HashMap<String, Array1<f64>>,
    ) -> TrainResult<(HashMap<String, Array1<f64>>, f64)>;
    fn adapt(
        &self,
        task: &MetaTask,
        parameters: &HashMap<String, Array1<f64>>,
    ) -> TrainResult<HashMap<String, Array1<f64>>>;
}
Expand description

Meta-learner trait for different meta-learning algorithms.

Required Methods§

Source

fn meta_step( &self, tasks: &[MetaTask], parameters: &HashMap<String, Array1<f64>>, ) -> TrainResult<(HashMap<String, Array1<f64>>, f64)>

Perform one step of meta-training on a batch of tasks.

§Arguments
  • tasks - Batch of meta-learning tasks
  • parameters - Current meta-parameters
§Returns

Updated meta-parameters and meta-loss

Source

fn adapt( &self, task: &MetaTask, parameters: &HashMap<String, Array1<f64>>, ) -> TrainResult<HashMap<String, Array1<f64>>>

Adapt parameters to a specific task (inner loop).

§Arguments
  • task - Task to adapt to
  • parameters - Initial parameters
§Returns

Task-adapted parameters

Implementors§