facet.selection.ParameterSpace#

class facet.selection.ParameterSpace(estimator, name=None)[source]#

A set of parameter choices or distributions spanning a parameter space for optimizing the hyperparameters of a single estimator.

Parameter spaces provide an easy approach to define and validate search spaces for hyperparameter tuning of ML pipelines using scikit-learn’s GridSearchCV and RandomizedSearchCV.

Parameter choices (as lists) or distributions (from scipy.stats) can be set using attribute access, and will be validated for correct names and values.

Example:

ps = ParameterSpace(
    RegressorPipelineDF(
        regressor=RandomForestRegressorDF(random_state=42),
        preprocessing=simple_preprocessor,
    ),
    name="random forest",
)
ps.regressor.min_weight_fraction_leaf = scipy.stats.loguniform(0.01, 0.1)
ps.regressor.max_depth = [3, 4, 5, 7, 10]

cv = RandomizedSearchCV(
    estimator=ps.estimator,
    param_distributions=ps.parameters,
    # ...
)

# The following will raise an AttributeError for the unknown attribute xyz:
ps.regressor.xyz = [3, 4, 5, 7, 10]

# the following will raise a TypeError because we do not assign a list or distribution:
ps.regressor.max_depth = 3

Bases

BaseParameterSpace [+T_Estimator_co]

Generic types

+T_Estimator_co(bound= EstimatorDF, __covariant__=True)

Metaclasses

ABCMeta

Parameters

estimator (ParameterSpace) – the estimator to which to apply the parameters
name (Optional[str]) – a name for the estimator to be used in summary reports; defaults to the name of the estimator’s class, or the name of the final estimator’s class if arg estimator is a pipeline

Method summary

`get_name`	Get the name for this parameter space's estimator.
`get_parameters`	Generate a dictionary of parameter choices and distributions, compatible with scikit-learn's CV search API (e.g., `GridSearchCV` or `RandomizedSearchCV`).
`to_expression`	Render this object as an expression.

Attribute summary

`estimator`	The estimator associated with this parameter space.
`parameters`	The parameter choices (as lists) or distributions (from `scipy.stats`) that constitute this parameter space.

Definitions

get_name()[source]#

Get the name for this parameter space’s estimator.

If no name was passed to the constructor, determine the default name recursively as follows:

for meta-estimators, this is the default name of the delegate estimator
for pipelines, this is the default name of the final estimator
for all other estimators, this is the name of the estimator’s class

Return type: str
Returns: the name for this parameter space’s estimator

get_parameters(prefix=None)[source]#

Generate a dictionary of parameter choices and distributions, compatible with scikit-learn’s CV search API (e.g., GridSearchCV or RandomizedSearchCV).

Parameters: prefix (Optional[str]) – an optional prefix to prepend to all parameter names in the resulting dictionary, separated by two underscore characters (__) as per scikit-learn’s convention for hierarchical parameter names
Return type: Dict[str, Union[List[Any], rv_continuous, rv_discrete]]
Returns: a dictionary mapping parameter names to parameter choices (as lists) or distributions (from scipy.stats)

to_expression()[source]#

Render this object as an expression.

Return type: Expression
Returns: the expression representing this object

property estimator: base.T_Estimator#

The estimator associated with this parameter space.

Return type: TypeVar(T_Estimator, bound= EstimatorDF, covariant=True)

property parameters: Union[List[Dict[str, Union[List[Any], scipy.stats.rv_continuous, scipy.stats.rv_discrete]]], Dict[str, Union[List[Any], scipy.stats.rv_continuous, scipy.stats.rv_discrete]]]#

The parameter choices (as lists) or distributions (from scipy.stats) that constitute this parameter space.

This is a shortcut for calling method get_parameters() with no arguments.

Return type: Union[List[Dict[str, Union[List[Any], rv_continuous, rv_discrete]]], Dict[str, Union[List[Any], rv_continuous, rv_discrete]]]

facet.selection.MultiEstimatorParameterSpace

facet.selection.base