Principle:Ucbepic Docetl Optimizer Configuration
| Knowledge Sources | |
|---|---|
| Domains | Optimization, Configuration |
| Last Updated | 2026-02-08 01:40 GMT |
Overview
A configuration principle that parameterizes the optimization search with model choices, iteration budgets, and evaluation criteria.
Description
Optimizer Configuration defines how the automated optimization process is parameterized. For the MOAR optimizer, this includes specifying the evaluation function file, the metric key to optimize, available LLM models, iteration budget, exploration constant for UCB (Upper Confidence Bound), and the output directory for optimized pipeline variants.
The configuration is provided as an optimizer_config section in the YAML pipeline file and parsed by the CLI build command.
Usage
Set the optimizer configuration before running docetl build. The configuration determines the scope, budget, and quality criteria of the optimization search.
Theoretical Basis
Optimizer configuration follows multi-objective optimization principles:
- Objective Specification: Define what "better" means via evaluation function and metric key
- Search Space: Define available models and directives
- Budget: Limit search iterations and computational cost
- Exploration vs Exploitation: Set UCB exploration constant to balance search diversity