nat.data_models.optimizer

Attributes

OptimizerStrategyBaseConfigT

Classes

OptimizerStrategyBaseConfig	Base for optimizer strategy configs (numeric, prompt) registered in the optimizer registry.
OptimizerMetric	Parameters used by the workflow optimizer to define a metric to optimize.
PromptOptimizerInputSchema	Input schema for prompt optimizer mutator/recombiner helper functions.
SamplerType	Enum where members are also (and must be) strings
OptunaParameterOptimizationConfig	Configuration for Optuna-based numeric/enum parameter optimization.
PromptOptimizationConfig	Base for all prompt optimization strategy configs.
GAPromptOptimizationConfig	GA-specific prompt optimization config with typed oracle feedback fields.
BaseOptimizerConfig	Shared optimizer parameters that any optimizer strategy could reuse.
OptimizerConfig	Full optimizer config used in the app Config and for parsing YAML.
OptimizerRunConfig	Parameters used for an Optimizer R=run

Module Contents

class OptimizerStrategyBaseConfig(/, **data: Any)

Bases: nat.data_models.common.TypedBaseModel, nat.data_models.common.BaseModelRegistryTag

Base for optimizer strategy configs (numeric, prompt) registered in the optimizer registry.

Create a new model by parsing and validating input data from keyword arguments.

Raises [ValidationError][pydantic_core.ValidationError] if the input data cannot be validated to form a valid model.

self is explicitly positional-only to allow self as a field name.

enabled: bool = None

class OptimizerMetric(/, **data: Any)

Bases: pydantic.BaseModel

Parameters used by the workflow optimizer to define a metric to optimize.

Create a new model by parsing and validating input data from keyword arguments.

Raises [ValidationError][pydantic_core.ValidationError] if the input data cannot be validated to form a valid model.

self is explicitly positional-only to allow self as a field name.

evaluator_name: str = None

direction: str = None

weight: float = None

class PromptOptimizerInputSchema(/, **data: Any)

Bases: pydantic.BaseModel

Input schema for prompt optimizer mutator/recombiner helper functions.

Create a new model by parsing and validating input data from keyword arguments.

Raises [ValidationError][pydantic_core.ValidationError] if the input data cannot be validated to form a valid model.

self is explicitly positional-only to allow self as a field name.

original_prompt: str

objective: str

oracle_feedback: str | None = None

class SamplerType

Bases: enum.StrEnum

Enum where members are also (and must be) strings

Initialize self. See help(type(self)) for accurate signature.

BAYESIAN = 'bayesian'

GRID = 'grid'

class OptunaParameterOptimizationConfig(/, **data: Any)

Bases: OptimizerStrategyBaseConfig

Configuration for Optuna-based numeric/enum parameter optimization.

Create a new model by parsing and validating input data from keyword arguments.

Raises [ValidationError][pydantic_core.ValidationError] if the input data cannot be validated to form a valid model.

self is explicitly positional-only to allow self as a field name.

enabled: bool = None

n_trials: int = None

sampler: SamplerType | None = None

class PromptOptimizationConfig(/, **data: Any)

Bases: OptimizerStrategyBaseConfig

Base for all prompt optimization strategy configs.

Create a new model by parsing and validating input data from keyword arguments.

Raises [ValidationError][pydantic_core.ValidationError] if the input data cannot be validated to form a valid model.

self is explicitly positional-only to allow self as a field name.

enabled: bool = None

prompt_population_init_function: str | None = None

prompt_recombination_function: str | None = None

class GAPromptOptimizationConfig(/, **data: Any)

Bases: PromptOptimizationConfig

GA-specific prompt optimization config with typed oracle feedback fields.

Create a new model by parsing and validating input data from keyword arguments.

Raises [ValidationError][pydantic_core.ValidationError] if the input data cannot be validated to form a valid model.

self is explicitly positional-only to allow self as a field name.

ga_population_size: int = None

ga_generations: int = None

ga_crossover_rate: float = None

ga_mutation_rate: float = None

ga_elitism: int = None

ga_selection_method: Literal['tournament', 'roulette'] = None

ga_tournament_size: int = None

ga_parallel_evaluations: int = None

ga_diversity_lambda: float = None

oracle_feedback_mode: Literal['never', 'always', 'failing_only', 'adaptive'] = None

oracle_feedback_worst_n: int = None

oracle_feedback_max_chars: int = None

oracle_feedback_fitness_threshold: float = None

oracle_feedback_stagnation_generations: int = None

oracle_feedback_fitness_variance_threshold: float = None

oracle_feedback_diversity_threshold: float = None

class BaseOptimizerConfig(/, **data: Any)

Bases: pydantic.BaseModel

Shared optimizer parameters that any optimizer strategy could reuse. Strategy-specific config lives on subtypes via the registry.

Create a new model by parsing and validating input data from keyword arguments.

Raises [ValidationError][pydantic_core.ValidationError] if the input data cannot be validated to form a valid model.

self is explicitly positional-only to allow self as a field name.

output_path: pathlib.Path | None = None

eval_metrics: dict[str, OptimizerMetric] | None = None

reps_per_param_set: int = None

target: float | None = None

multi_objective_combination_mode: str = None

class OptimizerConfig(/, **data: Any)

Bases: BaseOptimizerConfig

Full optimizer config used in the app Config and for parsing YAML. Extends the shared base with strategy-specific nests: .numeric for parameter/Optuna optimization, .prompt for prompt optimization.

Create a new model by parsing and validating input data from keyword arguments.

Raises [ValidationError][pydantic_core.ValidationError] if the input data cannot be validated to form a valid model.

self is explicitly positional-only to allow self as a field name.

numeric: OptunaParameterOptimizationConfig

prompt: GAPromptOptimizationConfig

OptimizerStrategyBaseConfigT

class OptimizerRunConfig(/, **data: Any)

Bases: pydantic.BaseModel

Parameters used for an Optimizer R=run

Create a new model by parsing and validating input data from keyword arguments.

Raises [ValidationError][pydantic_core.ValidationError] if the input data cannot be validated to form a valid model.

self is explicitly positional-only to allow self as a field name.

config_file: pathlib.Path | pydantic.BaseModel

dataset: str | pathlib.Path | None

result_json_path: str = '$'

endpoint: str | None = None

endpoint_timeout: int = 300

override: tuple[tuple[str, str], Ellipsis] = ()