data_models.strategies.api
data_models.strategies.api
Classes
| Name | Description |
|---|---|
| DoEStrategy | |
| FactorialStrategy | Factorial design strategy. |
| FractionalFactorialStrategy | Fractional factorial design strategy. |
| ActiveLearningStrategy | Datamodel for an ActiveLearningStrategy that focusses on pure exploration of the input space. |
| BotorchStrategy | |
| EntingStrategy | |
| MoboStrategy | |
| MultiFidelityStrategy | |
| MultiobjectiveStrategy | |
| AdditiveSoboStrategy | |
| CustomSoboStrategy | |
| LLMStrategy | Strategy that uses a large language model to propose optimization candidates. |
DoEStrategy
data_models.strategies.api.DoEStrategy()Attributes
| Name | Description |
|---|---|
| return_fixed_candidates | Datamodel for strategy for design of experiments. This strategy is used to generate a set of |
FactorialStrategy
data_models.strategies.api.FactorialStrategy()Factorial design strategy.
This strategy is deprecated, please use FractionalFactorialStrategy instead.
FractionalFactorialStrategy
data_models.strategies.api.FractionalFactorialStrategy()Fractional factorial design strategy.
This strategy generates a fractional factorial two level design for the continuous part of the domain, which is then combined with the categorical part of the domain. For every categorical combination, the continuous part of the design is repeated.
ActiveLearningStrategy
data_models.strategies.api.ActiveLearningStrategy()Datamodel for an ActiveLearningStrategy that focusses on pure exploration of the input space. This type of strategy chooses new candidate points in order to minimize the uncertainty.
Methods
| Name | Description |
|---|---|
| is_feature_implemented | Method to check if a specific feature type is implemented for the strategy |
| is_objective_implemented | Method to check if a objective type is implemented for the strategy |
is_feature_implemented
data_models.strategies.api.ActiveLearningStrategy.is_feature_implemented(
my_type,
)Method to check if a specific feature type is implemented for the strategy
Parameters
| Name | Type | Description | Default |
|---|---|---|---|
| my_type | Type[Feature] | Feature class | required |
Returns
| Name | Type | Description |
|---|---|---|
| bool | bool | True if the feature type is valid for the strategy chosen, False otherwise |
is_objective_implemented
data_models.strategies.api.ActiveLearningStrategy.is_objective_implemented(
my_type,
)Method to check if a objective type is implemented for the strategy
Parameters
| Name | Type | Description | Default |
|---|---|---|---|
| my_type | Type[Objective] | Objective class | required |
Returns
| Name | Type | Description |
|---|---|---|
| bool | bool | True if the objective type is valid for the strategy chosen, False otherwise |
BotorchStrategy
data_models.strategies.api.BotorchStrategy()Methods
| Name | Description |
|---|---|
| is_constraint_implemented | Method to check if a specific constraint type is implemented for the strategy. For optimizer-specific |
| validate_multitask_allowed | Ensures that if a multitask model is used there is only a single allowed task category |
| validate_outlier_detection_specs_for_domain | Ensures that a outlier_detection model is specified for each output feature |
| validate_surrogate_specs | Ensures that a prediction model is specified for each output feature |
is_constraint_implemented
data_models.strategies.api.BotorchStrategy.is_constraint_implemented(my_type)Method to check if a specific constraint type is implemented for the strategy. For optimizer-specific strategies, this is passed to the optimizer check.
Parameters
| Name | Type | Description | Default |
|---|---|---|---|
| my_type | Type[Constraint] | Constraint class | required |
Returns
| Name | Type | Description |
|---|---|---|
| bool | bool | True if the constraint type is valid for the strategy chosen, False otherwise |
validate_multitask_allowed
data_models.strategies.api.BotorchStrategy.validate_multitask_allowed()Ensures that if a multitask model is used there is only a single allowed task category
validate_outlier_detection_specs_for_domain
data_models.strategies.api.BotorchStrategy.validate_outlier_detection_specs_for_domain(
)Ensures that a outlier_detection model is specified for each output feature
validate_surrogate_specs
data_models.strategies.api.BotorchStrategy.validate_surrogate_specs()Ensures that a prediction model is specified for each output feature
EntingStrategy
data_models.strategies.api.EntingStrategy()MoboStrategy
data_models.strategies.api.MoboStrategy()Methods
| Name | Description |
|---|---|
| is_feature_implemented | Method to check if a specific feature type is implemented for the strategy |
| is_objective_implemented | Method to check if a objective type is implemented for the strategy |
| validate_ref_point | Validate that the provided refpoint matches the provided domain. |
is_feature_implemented
data_models.strategies.api.MoboStrategy.is_feature_implemented(my_type)Method to check if a specific feature type is implemented for the strategy
Parameters
| Name | Type | Description | Default |
|---|---|---|---|
| my_type | Type[Feature] | Feature class | required |
Returns
| Name | Type | Description |
|---|---|---|
| bool | bool | True if the feature type is valid for the strategy chosen, False otherwise |
is_objective_implemented
data_models.strategies.api.MoboStrategy.is_objective_implemented(my_type)Method to check if a objective type is implemented for the strategy
Parameters
| Name | Type | Description | Default |
|---|---|---|---|
| my_type | Type[Objective] | Objective class | required |
Returns
| Name | Type | Description |
|---|---|---|
| bool | bool | True if the objective type is valid for the strategy chosen, False otherwise |
validate_ref_point
data_models.strategies.api.MoboStrategy.validate_ref_point()Validate that the provided refpoint matches the provided domain.
MultiFidelityStrategy
data_models.strategies.api.MultiFidelityStrategy()Methods
| Name | Description |
|---|---|
| validate_multitask_allowed | Overwrites BotorchSurrogate.validate_multitask_allowed, as multiple tasks are allowed. |
| validate_only_one_target_fidelity | Ensures that there is only one target fidelity (task where fidelity==0). |
| validate_surrogate_specs | Ensures that a multi-task model is specified for each output feature |
| validate_tasks_and_fidelity_thresholds | Ensures that there is one threshold per fidelity |
validate_multitask_allowed
data_models.strategies.api.MultiFidelityStrategy.validate_multitask_allowed()Overwrites BotorchSurrogate.validate_multitask_allowed, as multiple tasks are allowed.
validate_only_one_target_fidelity
data_models.strategies.api.MultiFidelityStrategy.validate_only_one_target_fidelity(
)Ensures that there is only one target fidelity (task where fidelity==0).
validate_surrogate_specs
data_models.strategies.api.MultiFidelityStrategy.validate_surrogate_specs()Ensures that a multi-task model is specified for each output feature
validate_tasks_and_fidelity_thresholds
data_models.strategies.api.MultiFidelityStrategy.validate_tasks_and_fidelity_thresholds(
)Ensures that there is one threshold per fidelity
MultiobjectiveStrategy
data_models.strategies.api.MultiobjectiveStrategy()Methods
| Name | Description |
|---|---|
| validate_domain_is_multiobjective | Validate that the domain is multiobjective. |
validate_domain_is_multiobjective
data_models.strategies.api.MultiobjectiveStrategy.validate_domain_is_multiobjective(
v,
)Validate that the domain is multiobjective.
AdditiveSoboStrategy
data_models.strategies.api.AdditiveSoboStrategy()CustomSoboStrategy
data_models.strategies.api.CustomSoboStrategy()LLMStrategy
data_models.strategies.api.LLMStrategy()Strategy that uses a large language model to propose optimization candidates.
Instead of fitting a surrogate and optimizing an acquisition function, this strategy lets an LLM read the optimization problem — feature bounds, constraints, objectives, contextual descriptions, and prior experiments — and directly propose candidate points. It is useful for cold-start designs, mixed / categorical / molecular spaces where domain knowledge helps, and exploration informed by written context (Feature.context and Domain.context).
It is not a replacement for a Bayesian optimizer on well-understood numerical problems: there is no calibrated uncertainty model and no acquisition function. Treat candidates as informed heuristics, not optima.
On each ask(), a pydantic output schema is generated from the domain’s input features and the LLM is prompted with a textual problem description plus, optionally, a selection of prior experiments. Returned candidates are validated against the domain; bound or constraint violations are sent back to the LLM as retry messages via pydantic-ai’s output_retries.
Currently supports single-objective optimization with MaximizeObjective or MinimizeObjective, and LinearEquality, LinearInequality, and NChooseK constraints. All feature types are supported.
Example
Basic usage::
strategy = LLMStrategy.make(
domain=domain,
llm=AnthropicLLMProvider(model="claude-sonnet-4-5"),
)Enable extended reasoning for harder problems (many constraints, rich context). thinking is pydantic-ai’s cross-provider capability key — it maps to Anthropic’s extended thinking, OpenAI’s reasoning_effort, and similar mechanisms on other providers. Reasoning increases cost and latency considerably, so it is not enabled by default::
strategy = LLMStrategy.make(
domain=domain,
llm=AnthropicLLMProvider(model="claude-sonnet-4-5"),
model_settings={"thinking": "high"},
)Attributes
| Name | Type | Description |
|---|---|---|
| llm | AnyLLMProvider | LLM provider configuration. |
| model_settings | Optional[Dict[str, Any]] | Optional dict forwarded directly to pydantic-ai’s model_settings. Useful keys include temperature, max_tokens, top_p, seed, timeout, and the cross-provider capability thinking ("low" / "medium" / "high"). Provider-prefixed keys such as anthropic_thinking or openai_reasoning_effort are also accepted as escape hatches for finer control. Keys are not validated by BoFire — pydantic-ai and the underlying provider SDK are the source of truth. |
| output_retries | PositiveInt | Number of retries when output validation fails (constraint or bound violations). Each retry sends the LLM the invalid candidates and the error so it can correct. |
| n_recent_experiments | Optional[Annotated[int, Field(gt=0)]] | If set, only the most recent N experiments are shown to the LLM. Keeps prompt size bounded on long campaigns. |
| n_top_experiments | Optional[Annotated[int, Field(gt=0)]] | If set, the top N experiments by objective value are shown to the LLM. Combine with n_recent_experiments to mix recency and quality. |
| system_prompt | Optional[str] | Optional override for the default system prompt. |
Methods
| Name | Description |
|---|---|
| validate_single_objective | Validate that the domain has exactly one output with a supported objective. |
validate_single_objective
data_models.strategies.api.LLMStrategy.validate_single_objective()Validate that the domain has exactly one output with a supported objective.