Implementation:Ucbepic Docetl MOAR Node

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

Overview

Concrete tool for representing states in a Monte Carlo Tree Search (MCTS) tree for pipeline optimization provided by DocETL.

Description

The Node class represents a single state in the MCTS search tree used by the MOAR (Multi-Objective Automated Rewriting) optimizer. Each node holds a YAML pipeline configuration, maintains visit counts and accumulated values for UCB (Upper Confidence Bound) selection, and tracks parent-child relationships for tree traversal. Nodes can execute their pipeline plans via DSLRunner, compute UCB scores, and manage a memo of applied directives from root to the current node.

Usage

Use this class when implementing or extending the MOAR search algorithm for DocETL pipeline optimization. Nodes are created during tree expansion and evaluated during simulation phases of MCTS.

Code Reference

Source Location

• Repository: Ucbepic_Docetl
• File: docetl/moar/Node.py
• Lines: 1-735

Signature

class Node:
    _id_counter = 0

    @classmethod
    def get_next_id(cls) -> int: ...

    @classmethod
    def increment_id_counter(cls) -> int: ...

    def __init__(
        self,
        yaml_file_path: str,
        parent: Optional[Node] = None,
        c: float = 1.414,
        message_history=[],
        id: Optional[int] = None,
        is_multi_instance: bool = False,
        console=None,
    ): ...

    def execute_plan(self, max_threads: Optional[int] = None) -> tuple[float, list]: ...
    def best_child(self) -> Node: ...
    def add_child(self, child: Node): ...

Import

from docetl.moar.Node import Node

I/O Contract

Inputs

Name	Type	Required	Description
yaml_file_path	str	Yes	Path to the YAML pipeline configuration file for this node
parent	Optional[Node]	No	Parent node in the search tree (None for root)
c	float	No	Exploration constant for UCB calculation (default: 1.414 = sqrt(2))
message_history	list	No	Accumulated LLM conversation history from root to this node
id	Optional[int]	No	Explicit unique ID (auto-assigned if not provided)
is_multi_instance	bool	No	Whether this node is a multi-instance candidate
console	object	No	Console instance for logging (defaults to DOCETL_CONSOLE)

Outputs

Name	Type	Description
cost	float	Total execution cost from running the pipeline (-1 on failure)
sample_result	list	Sample output data extracted from the pipeline execution
value	float	Accumulated reward value from backpropagation
visits	int	Number of times this node has been visited in MCTS

Usage Examples

from docetl.moar.Node import Node

# Create a root node from a YAML pipeline file
root = Node(yaml_file_path="pipeline.yaml", c=1.414)

# Execute the pipeline plan
total_cost = root.execute_plan()

# Select the best child based on UCB score
best = root.best_child()

# Access node properties
print(f"Node {root.id}: visits={root.visits}, value={root.value}, cost={root.cost}")

Related Pages

• Environment:Ucbepic_Docetl_Python_Runtime