Base estimator

PoniardBaseEstimator sets up 95% of the funcionality for PoniardClassifier and PoniardRegressor.

PoniardBaseEstimator

 PoniardBaseEstimator (estimators:Optional[Union[Sequence[ClassifierMixin]
                       ,Dict[str,ClassifierMixin],Sequence[RegressorMixin]
                       ,Dict[str,RegressorMixin]]]=None, metrics:Optional[
                       Union[str,Dict[str,Callable],Sequence[str]]]=None,
                       preprocess:bool=True, custom_preprocessor:Union[Non
                       e,Pipeline,TransformerMixin,PoniardPreprocessor]=No
                       ne, cv:Union[int,BaseCrossValidator,BaseShuffleSpli
                       t,Sequence]=None, verbose:int=0,
                       random_state:Optional[int]=None,
                       n_jobs:Optional[int]=None,
                       plugins:Optional[Sequence[Any]]=None,
                       plot_options:Optional[PoniardPlotFactory]=None)

Base estimator that sets up all the functionality for the classifier and regressor.

	Type	Default	Details
estimators	Optional[Union[Sequence[ClassifierMixin], Dict[str, ClassifierMixin], Sequence[RegressorMixin], Dict[str, RegressorMixin]]]	None	Estimators to evaluate.
metrics	Optional[Union[str, Dict[str, Callable], Sequence[str]]]	None	Metrics to compute for each estimator. This is more restrictive than sklearn’s scoring parameter, as it does not allow callable scorers. Single strings are cast to lists automatically.
preprocess	bool	True	If True, impute missing values, standard scale numeric data and one-hot or ordinal encode categorical data.
custom_preprocessor	Union[None, Pipeline, TransformerMixin, PoniardPreprocessor]	None	Preprocessor used instead of the default preprocessing pipeline. It must be able to be included directly in a scikit-learn Pipeline.
cv	Union[int, BaseCrossValidator, BaseShuffleSplit, Sequence]	None	Cross validation strategy. Either an integer, a scikit-learn cross validation object, or an iterable.
verbose	int	0	Verbosity level. Propagated to every scikit-learn function and estimator.
random_state	Optional[int]	None	RNG. Propagated to every scikit-learn function and estimator. The default None sets random_state to 0 so that cross_validate results are comparable.
n_jobs	Optional[int]	None	Controls parallel processing. -1 uses all cores. Propagated to every scikit-learn function.
plugins	Optional[Sequence[Any]]	None	Plugin instances that run in set moments of setup, fit and plotting.
plot_options	Optional[PoniardPlotFactory]	None	:class:poniard.plot.plot_factory.PoniardPlotFactory instance specifying Plotly format options or None, which sets the default factory.

See the guides on Getting started, Main parameters and Preprocessing for examples on how the constructor parameters work.

Main methods

source

PoniardBaseEstimator.setup

 PoniardBaseEstimator.setup
                             (X:Union[pandas.core.frame.DataFrame,numpy.nd
                             array,List], y:Union[pandas.core.frame.DataFr
                             ame,numpy.ndarray,List], show_info:bool=True)

Acts as an orchestrator for Poniard estimators by setting up everything neeeded for PoniardBaseEstimator.fit.

Converts inputs to arrays if necessary, sets metrics, preprocessor, cv and pipelines.

After running PoniardBaseEstimator.setup, both X and y will be held as attributes.

	Type	Default	Details
X	Union[pd.DataFrame, np.ndarray, List]		Features
y	Union[pd.DataFrame, np.ndarray, List]		Target.
show_info	bool	True	Whether to print information about the target, metrics and type inference.
Returns	PoniardBaseEstimator

PoniardBaseEstimator.setup takes features and target as parameters, while PoniardBaseEstimator.fit does not accept any. This runs contrary to the established convention defined by scikit-learn where there is no setting up to do and fit takes the data as params.

This is because Poniard does not only fit the models, but also infer features types and create the preprocessor based on these types. While this could all be stuffed inside PoniardBaseEstimator.fit (that was the case initially), having it separated allows the user to check whether Poniard’s assumptions are correct and adjust if needed before running fit, which can take long depending on how many models were passed to estimators, the cross validation strategy and the size of the dataset.

PoniardBaseEstimator by default includes a PoniardPreprocessor that handles building the preprocessor that will go into final estimation pipelines. However, a PoniardPreprocessor with custom parameters can be used as a custom_preprocessor.

An example

Let’s load some random data and setup a PoniardClassifier, which inherits from PoniardBaseEstimator.

from poniard import PoniardClassifier

random.seed(0)
rng = np.random.default_rng(0)

data = pd.DataFrame(
    {
        "type": random.choices(["house", "apartment"], k=500),
        "age": rng.uniform(1, 200, 500).astype(int),
        "date": pd.date_range("2022-01-01", freq="M", periods=500),
        "rating": random.choices(range(50), k=500),
        "target": random.choices([0, 1], k=500),
    }
)
data.head()

	type	age	date	rating	target
0	apartment	127	2022-01-31	1	1
1	apartment	54	2022-02-28	17	1
2	house	9	2022-03-31	0	1
3	house	4	2022-04-30	48	1
4	apartment	162	2022-05-31	40	0

Information about the data will be shown so it can be reviewed and changes can be made.

X, y = data.drop("target", axis=1), data["target"]
pnd = PoniardClassifier()
pnd.setup(X, y)

Setup info

Target

Type: binary

Shape: (500,)

Unique values: 2

Metrics

Main metric: roc_auc

Feature type inference

Minimum unique values to consider a number-like feature numeric: 50

Minimum unique values to consider a categorical feature high cardinality: 20

Inferred feature types:

	numeric	categorical_high	categorical_low	datetime
0	age	rating	type	date

PoniardClassifier()

After passing data to Poniard estimators through setup, multiple attributes become available.

feature_types is a dict that sorts features in 4 categories (numeric, categorical_high, categorical_low and datetime) using some basic heuristics. This attribute is computed in PoniardPreprocessor.build, and will not be available if a non-PoniardPreprocessor transformer is passed to custom_preprocessor.

Feature types depend on the feature dtypes, and numeric_threshold and cardinality_threshold which are used in PoniardPreprocessor’s construction.

pnd.feature_types

{'numeric': ['age'],
 'categorical_high': ['rating'],
 'categorical_low': ['type'],
 'datetime': ['date']}

The preprocessor can be the transformer produced by a PoniardPreprocessor, which in turn depends on feature_types, and the scaler, numeric_imputer and high_cardinality_encoder parameters, or a user-supplied scikit-learn compatible transformer.

As will be seen further on, the PoniardPreprocessor can be modified significantly to fit multiple use cases and datasets.

pnd.preprocessor

Each estimator has a set of default metrics, but others can be passed during construction.

pnd.metrics

['roc_auc', 'accuracy', 'precision', 'recall', 'f1']

Likewise, cv has sane defaults but can be modified accordingly.

pnd.cv

StratifiedKFold(n_splits=5, random_state=0, shuffle=True)

target_info lists information about y.

pnd.target_info

{'type_': 'binary', 'ndim': 1, 'shape': (500,), 'nunique': 2}

pipelines is a dict containing each pipeline which will be trained during fit. Each Poniard estimator has a limited set of default estimators that are used if none are specified during initialization.

pnd.pipelines["SVC"]

source

PoniardBaseEstimator.fit

 PoniardBaseEstimator.fit ()

This is the main Poniard method. It uses scikit-learn’s cross_validate function to score all metrics for every pipelines, using cv for cross validation.

pnd.fit()

PoniardClassifier()

Because features and target are passed to the Poniard estimator, fit does not take any parameters.

After fitting pipelines, cross validated results can be accessed by running get_results

source

PoniardBaseEstimator.get_results

 PoniardBaseEstimator.get_results (return_train_scores:bool=False,
                                   std:bool=False, wrt_dummy:bool=False)

Return dataframe containing scoring results. By default returns the mean score and fit and score times. Optionally returns standard deviations as well.

	Type	Default	Details
return_train_scores	bool	False	If False, only return test scores.
std	bool	False	Whether to return standard deviation of the scores. Default False.
wrt_dummy	bool	False	Whether to compute each score/time with respect to the dummy estimator results. Default False.
Returns	Union[Tuple[pd.DataFrame, pd.DataFrame], pd.DataFrame]		Results

pnd.get_results()

	test_roc_auc	test_accuracy	test_precision	test_recall	test_f1	fit_time	score_time
DecisionTreeClassifier	0.510256	0.510	0.531145	0.503846	0.516707	0.010714	0.007243
DummyClassifier	0.500000	0.520	0.520000	1.000000	0.684211	0.009618	0.007332
KNeighborsClassifier	0.496675	0.492	0.509150	0.534615	0.519465	0.009883	0.008536
SVC	0.472356	0.476	0.499007	0.688462	0.575907	0.715862	0.008426
LogisticRegression	0.468990	0.488	0.509234	0.573077	0.536862	0.019850	0.007661
XGBClassifier	0.460417	0.486	0.502401	0.500000	0.499330	0.046362	0.009421
HistGradientBoostingClassifier	0.456571	0.488	0.505975	0.484615	0.494283	0.405131	0.019346
RandomForestClassifier	0.435056	0.462	0.479861	0.476923	0.477449	0.070931	0.014314
GaussianNB	0.423317	0.468	0.492473	0.565385	0.525371	0.010134	0.007401

means, stds = pnd.get_results(std=True, return_train_scores=True)
stds

	test_roc_auc	train_roc_auc	test_accuracy	train_accuracy	test_precision	train_precision	test_recall	train_recall	test_f1	train_f1	fit_time	score_time
DecisionTreeClassifier	0.060706	0.000000e+00	0.060332	0.000000	0.059942	0.000000	0.058835	0.000000	0.057785	0.000000	0.000303	0.000047
DummyClassifier	0.000000	0.000000e+00	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000404	0.000100
KNeighborsClassifier	0.021105	8.429609e-03	0.019391	0.010840	0.019140	0.008157	0.081043	0.022053	0.049760	0.012869	0.000341	0.000070
SVC	0.038609	3.600720e-02	0.042708	0.032496	0.031965	0.028405	0.085485	0.073140	0.036968	0.026864	0.110736	0.000229
LogisticRegression	0.068079	2.545484e-02	0.041183	0.027946	0.037992	0.024759	0.065948	0.021371	0.036585	0.022583	0.004623	0.000269
XGBClassifier	0.065278	0.000000e+00	0.035553	0.000000	0.033315	0.000000	0.091826	0.000000	0.061108	0.000000	0.001688	0.000196
HistGradientBoostingClassifier	0.059681	7.749323e-04	0.041183	0.007483	0.039938	0.011912	0.070291	0.005607	0.054859	0.007046	0.049279	0.005965
RandomForestClassifier	0.060809	7.021667e-17	0.039192	0.000000	0.038392	0.000000	0.077307	0.000000	0.056132	0.000000	0.000342	0.000267
GaussianNB	0.045845	2.494438e-02	0.042143	0.018303	0.037330	0.015830	0.031246	0.038051	0.025456	0.018727	0.000729	0.000126

get_estimator is a convenience method that gets a pipeline from pipelines by name, and optionally trains it on X and y.

source

PoniardBaseEstimator.get_estimator

 PoniardBaseEstimator.get_estimator (estimator_name:str,
                                     include_preprocessor:bool=True,
                                     retrain:bool=False)

Obtain an estimator in pipelines by name. This is useful for extracting default estimators or hyperparmeter-optimized estimators (after using PoniardBaseEstimator.tune_estimator).

	Type	Default	Details
estimator_name	str		Estimator name.
include_preprocessor	bool	True	Whether to return a pipeline with a preprocessor or just the estimator. Default True.
retrain	bool	False	Whether to retrain with full data. Default False.
Returns	Union[Pipeline, ClassifierMixin, RegressorMixin]		Estimator.

source

PoniardBaseEstimator.analyze_estimator

 PoniardBaseEstimator.analyze_estimator (estimator_name:str,
                                         height:int=800, width:int=800)

Print a selection of metrics and plots for a given estimator.

By default, orders estimators according to the first metric.

	Type	Default	Details
estimator_name	str		Name of estimator to analyze.
height	int	800	Height of output `Figure`.
width	int	800	Width of output `Figure`.
Returns	Figure		Figure

PoniardBaseEstimator.analyze_estimator provides a quick overview of an estimator’s performance.

from sklearn.datasets import load_breast_cancer
from poniard import PoniardClassifier

X, y = load_breast_cancer(return_X_y=True, as_frame=True)
pnd = PoniardClassifier().setup(X, y, show_info=False)
pnd.fit()

PoniardClassifier()

pnd.analyze_estimator("SVC", height=1000, width=1000)

Modifying estimators after initialization

Estimators can be added and removed directly. Note that if other estimators have been already fit, only the added ones will be processed during PoniardBaseEstimator.fit.

source

PoniardBaseEstimator.add_estimators

 PoniardBaseEstimator.add_estimators (estimators:Union[Dict[str,sklearn.ba
                                      se.ClassifierMixin],Sequence[sklearn
                                      .base.ClassifierMixin]])

Include new estimator. This is the recommended way of adding an estimator (as opposed to modifying pipelines directly), since it also injects random state, n_jobs and verbosity.

	Type	Details
estimators	Union[Dict[str, ClassifierMixin], Sequence[ClassifierMixin]]	Estimators to add.
Returns	PoniardBaseEstimator	Self.

source

PoniardBaseEstimator.remove_estimators

 PoniardBaseEstimator.remove_estimators (estimator_names:Sequence[str],
                                         drop_results:bool=True)

Remove estimators. This is the recommended way of removing an estimator (as opposed to modifying pipelines directly), since it also removes the associated rows from the results tables.

	Type	Default	Details
estimator_names	Sequence[str]		Estimators to remove.
drop_results	bool	True	Whether to remove the results associated with the estimators. Default True.
Returns	PoniardBaseEstimator		Self.

pnd.add_estimators(ExtraTreesClassifier())
pnd.remove_estimators("RandomForestClassifier")
pnd.fit()
pnd.get_results()

	test_roc_auc	test_accuracy	test_precision	test_recall	test_f1	fit_time	score_time
LogisticRegression	0.995456	0.978916	0.975411	0.991549	0.983351	0.007645	0.002424
SVC	0.994139	0.975408	0.975111	0.985955	0.980477	0.008037	0.003919
HistGradientBoostingClassifier	0.994128	0.970129	0.967263	0.985955	0.976433	0.539054	0.016192
XGBClassifier	0.994123	0.970129	0.967554	0.985915	0.976469	0.049444	0.004278
ExtraTreesClassifier	0.991055	0.968359	0.969925	0.980321	0.974955	0.042767	0.008918
GaussianNB	0.988730	0.929700	0.940993	0.949413	0.944300	0.003169	0.004466
KNeighborsClassifier	0.980610	0.964881	0.955018	0.991628	0.972746	0.002539	0.016843
DecisionTreeClassifier	0.920983	0.926223	0.941672	0.941080	0.941054	0.005269	0.002359
DummyClassifier	0.500000	0.627418	0.627418	1.000000	0.771052	0.001970	0.002919

pnd.pipelines.keys()

dict_keys(['LogisticRegression', 'GaussianNB', 'SVC', 'KNeighborsClassifier', 'DecisionTreeClassifier', 'HistGradientBoostingClassifier', 'XGBClassifier', 'DummyClassifier', 'ExtraTreesClassifier'])

Modifying the preprocessor after initializaiton

The preprocessor can be modified from within PoniardBaseEstimator in two ways after PoniardBaseEstimator.setup:

reassign_types so that features are processed by other transformers, i.e., a numeric feature could be cast to a high cardinality categorical (for example, a store ID).
add_preprocessing_step adds a transformer or pipeline to the existing preprocessor.

See the Preprocessing guide for examples.

source

PoniardBaseEstimator.reassign_types

 PoniardBaseEstimator.reassign_types
                                      (numeric:Optional[List[Union[str,int
                                      ]]]=None, categorical_high:Optional[
                                      List[Union[str,int]]]=None, categori
                                      cal_low:Optional[List[Union[str,int]
                                      ]]=None, datetime:Optional[List[Unio
                                      n[str,int]]]=None,
                                      keep_remainder:bool=True)

Reassign feature types. By default, leaves ommitted features as they were.

	Type	Default	Details
numeric	Optional[List[Union[str, int]]]	None	List of column names or indices. Default None.
categorical_high	Optional[List[Union[str, int]]]	None	List of column names or indices. Default None.
categorical_low	Optional[List[Union[str, int]]]	None	List of column names or indices. Default None.
datetime	Optional[List[Union[str, int]]]	None	List of column names or indices. Default None.
keep_remainder	bool	True	Whether to keep features not specified in the method parameters as is or drop them
Returns	PoniardBaseEstimator		self.

source

PoniardBaseEstimator.add_preprocessing_step

 PoniardBaseEstimator.add_preprocessing_step (step:Union[sklearn.pipeline.
                                              Pipeline,sklearn.base.Transf
                                              ormerMixin,sklearn.compose._
                                              column_transformer.ColumnTra
                                              nsformer,Tuple[str,Union[skl
                                              earn.pipeline.Pipeline,sklea
                                              rn.base.TransformerMixin,skl
                                              earn.compose._column_transfo
                                              rmer.ColumnTransformer]]], p
                                              osition:Union[str,int]='end'
                                              )

Add a preprocessing step.

	Type	Default	Details
step	Union[Union[Pipeline, TransformerMixin, ColumnTransformer], Tuple[str, Union[Pipeline, TransformerMixin, ColumnTransformer]]]		A tuple of (str, transformer) or a scikit-learn transformer. Note that the transformer can also be a Pipeline or ColumnTransformer.
position	Union[str, int]	end	Either an integer denoting before which step in the existing preprocessing pipeline the new step should be added, or ‘start’ or ‘end’.
Returns	Pipeline		self

Prediction methods

Cross validated predictions (using scikit-learn’s cross_val_predict) can be obtained by calling the predict, predict_proba, decision_function or predict_all methods. Each of them takes an estimator_names parameter that specifies which models should be used.

Estimators without specific prediction methods

Not every scikit-learn model includes the predict_proba method. Poniard will set them to numpy.nan instead of throwing an error.

source

PoniardBaseEstimator.predict

 PoniardBaseEstimator.predict
                               (estimator_names:Optional[Sequence[str]]=No
                               ne)

Get cross validated target predictions where each sample belongs to a single test set.

	Type	Default	Details
estimator_names	Optional[Sequence[str]]	None	Estimators to include. If None, predict all estimators.
Returns	Dict[str, np.ndarray]		Dict where keys are estimator names and values are numpy arrays of predictions.

source

PoniardBaseEstimator.predict_proba

 PoniardBaseEstimator.predict_proba
                                     (estimator_names:Optional[Sequence[st
                                     r]]=None)

Get cross validated target probability predictions where each sample belongs to a single test set.

	Type	Default	Details
estimator_names	Optional[Sequence[str]]	None
Returns	Dict[str, np.ndarray]		Dict where keys are estimator names and values are numpy arrays of prediction probabilities.

source

PoniardBaseEstimator.decision_function

 PoniardBaseEstimator.decision_function
                                         (estimator_names:Optional[Sequenc
                                         e[str]]=None)

Get cross validated decision function predictions where each sample belongs to a single test set.

	Type	Default	Details
estimator_names	Optional[Sequence[str]]	None	Estimators to include. If None, predict all estimators.
Returns	Dict[str, np.ndarray]		Dict where keys are estimator names and values are numpy arrays of decision functions.

source

PoniardBaseEstimator.predict_all

 PoniardBaseEstimator.predict_all
                                   (estimator_names:Optional[Sequence[str]
                                   ]=None)

Get cross validated target predictions, probabilities and decision functions where each sample belongs to a test set.

	Type	Default	Details
estimator_names	Optional[Sequence[str]]	None	Estimators to include. If None, predict all estimators.
Returns	Tuple[Dict[str, np.ndarray]]		Tuple of dicts where keys are estimator names and values are numpy arrays of predictions.

Ensembles and hyperparameter tuning

source

PoniardBaseEstimator.build_ensemble

 PoniardBaseEstimator.build_ensemble (method:str='stacking',
                                      estimator_names:Optional[Sequence[st
                                      r]]=None, top_n:Optional[int]=3,
                                      sort_by:Optional[str]=None,
                                      ensemble_name:Optional[str]=None,
                                      **kwargs)

Combine estimators into an ensemble.

By default, orders estimators according to the first metric.

	Type	Default	Details
method	str	stacking	Ensemble method. Either “stacking” or “voting”. Default “stacking”.
estimator_names	Optional[Sequence[str]]	None	Names of estimators to include. Default None, which uses `top_n`
top_n	Optional[int]	3	How many of the best estimators to include.
sort_by	Optional[str]	None	Which metric to consider for ordering results. Default None, which uses the first metric.
ensemble_name	Optional[str]	None	Ensemble name when adding to `pipelines`. Default None.
kwargs			Passed to the ensemble class constructor.
Returns	PoniardBaseEstimator		Self.

pnd.build_ensemble(
    method="stacking",
    estimator_names=["DecisionTreeClassifier", "KNeighborsClassifier", "SVC"],
)
pnd.get_estimator("StackingClassifier")

pnd.fit()
pnd.get_results()

	test_roc_auc	test_accuracy	test_precision	test_recall	test_f1	fit_time	score_time
LogisticRegression	0.995456	0.978916	0.975411	0.991549	0.983351	0.007645	0.002424
SVC	0.994139	0.975408	0.975111	0.985955	0.980477	0.008037	0.003919
HistGradientBoostingClassifier	0.994128	0.970129	0.967263	0.985955	0.976433	0.539054	0.016192
XGBClassifier	0.994123	0.970129	0.967554	0.985915	0.976469	0.049444	0.004278
StackingClassifier	0.993999	0.973653	0.967485	0.991588	0.979308	0.053218	0.005176
ExtraTreesClassifier	0.991055	0.968359	0.969925	0.980321	0.974955	0.042767	0.008918
GaussianNB	0.988730	0.929700	0.940993	0.949413	0.944300	0.003169	0.004466
KNeighborsClassifier	0.980610	0.964881	0.955018	0.991628	0.972746	0.002539	0.016843
DecisionTreeClassifier	0.920983	0.926223	0.941672	0.941080	0.941054	0.005269	0.002359
DummyClassifier	0.500000	0.627418	0.627418	1.000000	0.771052	0.001970	0.002919

Use get_predictions_similarity to compute how correlated the estimators’ predictions are. This can be useful for building ensembles with PoniardBaseEstimator.build_ensemble.

source

PoniardBaseEstimator.get_predictions_similarity

 PoniardBaseEstimator.get_predictions_similarity (on_errors:bool=True)

Compute correlation/association between cross validated predictions for each estimator.

This can be useful for ensembling.

	Type	Default	Details
on_errors	bool	True	Whether to compute similarity on prediction errors instead of predictions. Default True.
Returns	pd.DataFrame		Similarity.

pnd.get_predictions_similarity()

	LogisticRegression	GaussianNB	SVC	KNeighborsClassifier	DecisionTreeClassifier	HistGradientBoostingClassifier	XGBClassifier	ExtraTreesClassifier	StackingClassifier
LogisticRegression	1.000000	0.315978	0.726194	0.401876	0.211925	0.367325	0.294833	0.426033	0.547327
GaussianNB	0.315978	1.000000	0.331160	0.524911	0.354022	0.454955	0.495528	0.518582	0.489758
SVC	0.726194	0.331160	1.000000	0.368042	0.277664	0.403438	0.336311	0.390700	0.574735
KNeighborsClassifier	0.401876	0.524911	0.368042	1.000000	0.363762	0.497702	0.497702	0.482094	0.712582
DecisionTreeClassifier	0.211925	0.354022	0.277664	0.363762	1.000000	0.362908	0.521706	0.427338	0.392178
HistGradientBoostingClassifier	0.367325	0.454955	0.403438	0.497702	0.362908	1.000000	0.726570	0.645759	0.582252
XGBClassifier	0.294833	0.495528	0.336311	0.497702	0.521706	0.726570	1.000000	0.704906	0.517572
ExtraTreesClassifier	0.426033	0.518582	0.390700	0.482094	0.427338	0.645759	0.704906	1.000000	0.564618
StackingClassifier	0.547327	0.489758	0.574735	0.712582	0.392178	0.582252	0.517572	0.564618	1.000000

Poniard offers light hyperparameter tuning through tune_estimator, as well as hyperparameter grids for its default estimators. You are however free to specify whichever grid you want.

source

PoniardBaseEstimator.tune_estimator

 PoniardBaseEstimator.tune_estimator (estimator_name:str,
                                      grid:Optional[Dict]=None,
                                      mode:str='grid', tuned_estimator_nam
                                      e:Optional[str]=None, **kwargs)

Hyperparameter tuning for a single estimator.

	Type	Default	Details
estimator_name	str		Estimator to tune.
grid	Optional[Dict]	None	Hyperparameter grid. Default None, which uses the grids available for default estimators.
mode	str	grid	Type of search. Eitherr “grid”, “halving” or “random”. Default “grid”.
tuned_estimator_name	Optional[str]	None	Estimator name when adding to `pipelines`. Default None.
kwargs			Passed to the search class constructor.
Returns	Union[GridSearchCV, RandomizedSearchCV]		Self.