Configuration Reference¶

All class attributes, parameter constraints, and configuration functions available when defining wrapper classes or using EstimatorConfig.

Wrapper Class Attributes¶

Required¶

_estimator_name : str: Parameter name for the wrapped class. Used as the key in get_params() and as the keyword argument name in the wrapper constructor. Example: "regressor", "classifier", "model".
_estimator_base_class : type: Base class the wrapped class must inherit from. Validated at instantiation. Use object for no restriction.

Optional¶

_estimator_default_class : type | None: Default class to use when the _estimator_name keyword is not passed to the constructor. Defaults to None (the class argument is required).
_parameter_constraints : dict[str, list]: Dictionary mapping parameter names to lists of allowed types or constraint objects. When defined, parameters are validated automatically by _fit_context before instantiation. Defaults to {} (no validation).

Parameter Constraint Types¶

Constraints use sklearn.utils._param_validation. Each parameter maps to a list of allowed constraints, and the value must match at least one.

Constraint	Import	Description
`Interval(type, low, high, closed)`	`sklearn.utils._param_validation`	Numeric range. `closed` is one of `"left"`, `"right"`, `"both"`, `"neither"`.
`StrOptions(options)`	`sklearn.utils._param_validation`	Set of allowed strings.
`type`	built-in	Exact type check (`int`, `str`, `float`, etc.)
`None`	built-in	Allow `None` as a value.
`callable`	built-in	Any callable object.
`{"wrapper_base_class": type}`	sklearn-wrap specific	Validates that the value is a `BaseClassWrapper` whose `_estimator_base_class` matches the given type.

Sentinel Values¶

REQUIRED_PARAM_VALUE: String sentinel ("__REQUIRED__") used internally for required parameters without defaults. If any parameter still holds this value when instantiate() is called, a ValueError is raised.

Configuration Functions¶

These functions manage the EstimatorConfig trusted modules setting.

`get_config() -> dict[str, Any]`¶

Return a copy of the current sklearn-wrap configuration. Thread-local overrides take precedence over global defaults.

`set_config(*, trusted_modules: frozenset[str] | None = None) -> None`¶

Persistently update the configuration for the current thread.

`reset_config() -> None`¶

Reset the configuration to defaults (trusted_modules = frozenset({"sklearn", "sklearn_wrap"})).

`config_context(*, trusted_modules: frozenset[str] | None = None) -> Generator`¶

Context manager that temporarily overrides the configuration. Reverts on exit:

from sklearn_wrap.config import config_context

with config_context(trusted_modules=frozenset({"sklearn", "xgboost"})):
    estimator = config.build()
# trusted_modules reverts here

EstimatorConfig¶

EstimatorConfig is a Pydantic BaseModel for declarative estimator configuration.

Fields¶

Field	Type	Description
`estimator_class`	`str`	Dotted import path (e.g. `"sklearn.linear_model.Ridge"`).
`params`	`dict[str, Any]`	Constructor keyword arguments. Default: `{}`.

Methods¶

Method	Signature	Description
`build`	`(trusted_modules=None) -> object`	Import the class, validate params, return an instance. Uses thread config if `trusted_modules` is `None`.
`from_estimator`	`(estimator) -> EstimatorConfig`	Class method. Capture config from an existing sklearn-compatible estimator.
`from_yaml`	`(path) -> EstimatorConfig`	Class method. Load from a YAML file. Supports `!include` tags.
`to_yaml`	`(path) -> None`	Save the config to a YAML file.

Security¶

build() only resolves classes from trusted modules. The default set is {"sklearn", "sklearn_wrap"}. Use set_config or config_context to add modules like "xgboost". The UntrustedModuleError exception is raised for disallowed modules.