Struct WarmStartExactJhAgent 

pub struct WarmStartExactJhAgent { /* private fields */ }

Warm-start exact finite-horizon objective controller.

Implementations

impl WarmStartExactJhAgent

pub fn new( config: WarmStartExactJhConfig, teacher: WarmStartExactJhTeacherDataset, ) -> Result<Self, WarmStartExactJhError>

Construct a new warm-start exact-J_H agent.

pub fn from_compiled_planner_run( compiled: &CompiledPlannerRunSpec, teacher: WarmStartExactJhTeacherDataset, ) -> Result<Self, WarmStartExactJhError>

Construct from a compiled planner-run spec and same-task teacher data.

pub fn steps_observed(&self) -> usize

Number of transitions incorporated from live interaction.

pub fn teacher_label_count(&self) -> usize

Number of warm-start labels incorporated from the teacher dataset.

pub fn same_task_live_trace(&self) -> Option<WarmStartExactJhTeacherTrace>

Extract the current live same-task trajectory as a teacher trace.

The trace is admissible under the same runtime validator used for teacher datasets because it was produced through observe_transition.

Cost: despite the &self receiver this is a full materialization, not a cheap accessor. It allocates a fresh transition vector and clones every stored observation stream, so a call is O(steps * observation_stream_len) in both time and allocated memory. It is intended for occasional refresh/export points; callers in a hot loop should cache the result rather than re-deriving it per step.

pub fn num_actions(&self) -> ActionAlphabet

Configured action alphabet cardinality.

pub fn planner_simulations_per_step(&self) -> usize

Canonical direct-evaluator budget marker.

This is always 1; warm-start exact-(J_H) has no simulation loop.

pub fn resolved_random_seed(&self) -> u64

Resolved deterministic seed.

pub fn get_planned_action(&mut self) -> Action

Select the next greedy action from the current exact-return model.

Panics

Panics if retained live history violates the validated warm-start runtime contract. Use Self::try_get_planned_action to receive that condition as WarmStartExactJhError.

pub fn try_get_planned_action( &mut self, ) -> Result<Action, WarmStartExactJhError>

Fallibly select the next greedy action from the current exact-return model.

pub fn estimate_action_values( &mut self, ) -> Result<Vec<f64>, WarmStartExactJhError>

Estimate exact finite-horizon action values at the current decision state.

pub fn get_planned_action_with_extra_exploration( &mut self, extra_exploration: f64, ) -> Action

Select the next action with optional epsilon exploration.

Warm-start exact-(J_H) has no baseline exploration parameter; this method’s argument is the entire exploration probability.

Panics

Panics if greedy planning is reached and retained live history violates the validated warm-start runtime contract. Use Self::try_get_planned_action_with_extra_exploration_flag to receive that condition as WarmStartExactJhError.

pub fn get_planned_action_with_extra_exploration_flag( &mut self, extra_exploration: f64, ) -> (Action, bool)

Select the next action with optional epsilon exploration and return whether exploration fired.

Warm-start exact-(J_H) has no baseline exploration parameter; this method’s argument is the entire exploration probability.

Panics

Panics if greedy planning is reached and retained live history violates the validated warm-start runtime contract. Use Self::try_get_planned_action_with_extra_exploration_flag to receive that condition as WarmStartExactJhError.

pub fn try_get_planned_action_with_extra_exploration_flag( &mut self, extra_exploration: f64, ) -> Result<(Action, bool), WarmStartExactJhError>

Fallibly select the next action with optional epsilon exploration and return whether exploration fired.

Warm-start exact-(J_H) has no baseline exploration parameter; this method’s argument is the entire exploration probability.

pub fn observe_transition( &mut self, action: Action, observations: &[PerceptVal], reward: Reward, ) -> Result<(), WarmStartExactJhError>

Record one live environment transition.