Implementation:Ucbepic Docetl MOAR ParetoFrontier

Knowledge Sources	Ucbepic_Docetl
Domains	Data_Processing, Optimization, Multi_Objective_Optimization
Last Updated	2026-02-08 00:00 GMT

Overview

Concrete tool for managing a Pareto frontier of cost-accuracy tradeoffs during MCTS-based pipeline optimization provided by DocETL.

Description

The ParetoFrontier class maintains a collection of pipeline plans (represented as Node objects), estimates their accuracy through evaluation functions or pairwise comparisons, and constructs a Pareto frontier over the cost-accuracy space. It provides hypervolume-based frontier updates, distance calculations for reward signals, and integrates with the MCTS algorithm by supplying value estimates for backpropagation. The class also generates summary statistics and visualization plots of the frontier.

Usage

Use this class within the MOAR optimizer to track and manage the set of non-dominated pipeline configurations discovered during search. It serves as the reward mechanism that guides the MCTS exploration toward Pareto-optimal solutions.

Code Reference

Source Location

• Repository: Ucbepic_Docetl
• File: docetl/moar/ParetoFrontier.py
• Lines: 1-425

Signature

class ParetoFrontier:
    def __init__(
        self,
        action_rewards: Dict[str, float],
        action_cost_changes: Dict[str, float],
        action_accuracy_changes: Dict[str, float],
        dataset_name: str,
        evaluate_func: Callable[[str], Dict[str, Any]],
        console=None,
    ): ...

    def add_plan(self, node: Node) -> Dict[Node, int]: ...
    def add_plan_f1(self, node: Node, accuracy: float) -> Tuple[Dict[Node, int], bool]: ...
    def get_all_plans_summary(self) -> List[Dict[str, Any]]: ...

Import

from docetl.moar.ParetoFrontier import ParetoFrontier

I/O Contract

Inputs

Name	Type	Required	Description
action_rewards	Dict[str, float]	Yes	Reference to MCTS action rewards dictionary for tracking
action_cost_changes	Dict[str, float]	Yes	Reference to MCTS action cost changes dictionary
action_accuracy_changes	Dict[str, float]	Yes	Reference to MCTS action accuracy changes dictionary
dataset_name	str	Yes	Name of the dataset being optimized (maps to a primary metric)
evaluate_func	Callable[[str], Dict[str, Any]]	Yes	Evaluation function taking a results file path and returning metrics
console	object	No	Console instance for logging (defaults to DOCETL_CONSOLE)

Outputs

Name	Type	Description
affected_nodes	Dict[Node, int]	Mapping of nodes affected by frontier updates with their status
is_frontier_updated	bool	Whether the Pareto frontier was changed by adding a new plan
plans_summary	List[Dict[str, Any]]	Summary of all plans with cost, accuracy, and value metrics

Usage Examples

from docetl.moar.ParetoFrontier import ParetoFrontier
from docetl.moar.Node import Node

# Initialize the Pareto frontier
frontier = ParetoFrontier(
    action_rewards={},
    action_cost_changes={},
    action_accuracy_changes={},
    dataset_name="cuad",
    evaluate_func=lambda path: {"avg_f1": 0.85},
)

# Add a plan with pre-evaluated accuracy
node = Node(yaml_file_path="optimized_pipeline.yaml")
node.cost = 2.50
affected, updated = frontier.add_plan_f1(node, accuracy=0.85)

# Get summary of all plans
summary = frontier.get_all_plans_summary()
for plan in summary:
    print(f"Node {plan['node']}: cost={plan['cost']}, accuracy={plan['accuracy']}")

Related Pages

• Environment:Ucbepic_Docetl_Python_Runtime