""" The :mod:`sklearn.model_selection._search` includes utilities to fine-tune the parameters of an estimator. """ from __future__ import print_function from __future__ import division # Author: Alexandre Gramfort , # Gael Varoquaux # Andreas Mueller # Olivier Grisel # Raghav RV # License: BSD 3 clause from abc import ABCMeta, abstractmethod from collections import namedtuple, defaultdict from functools import partial, reduce from itertools import product import operator import warnings import numpy as np from scipy.stats import rankdata from ..base import BaseEstimator, is_classifier, clone from ..base import MetaEstimatorMixin from ._split import check_cv from ._validation import _fit_and_score from ._validation import _aggregate_score_dicts from ..exceptions import NotFittedError from ..externals.joblib import Parallel, delayed from ..externals import six from ..utils import check_random_state from ..utils.fixes import sp_version from ..utils.fixes import MaskedArray from ..utils.fixes import _Mapping as Mapping, _Sequence as Sequence from ..utils.random import sample_without_replacement from ..utils.validation import indexable, check_is_fitted from ..utils.metaestimators import if_delegate_has_method from ..utils.deprecation import DeprecationDict from ..metrics.scorer import _check_multimetric_scoring from ..metrics.scorer import check_scoring __all__ = ['GridSearchCV', 'ParameterGrid', 'fit_grid_point', 'ParameterSampler', 'RandomizedSearchCV'] class ParameterGrid(object): """Grid of parameters with a discrete number of values for each. Can be used to iterate over parameter value combinations with the Python built-in function iter. Read more in the :ref:`User Guide `. Parameters ---------- param_grid : dict of string to sequence, or sequence of such The parameter grid to explore, as a dictionary mapping estimator parameters to sequences of allowed values. An empty dict signifies default parameters. A sequence of dicts signifies a sequence of grids to search, and is useful to avoid exploring parameter combinations that make no sense or have no effect. See the examples below. Examples -------- >>> from sklearn.model_selection import ParameterGrid >>> param_grid = {'a': [1, 2], 'b': [True, False]} >>> list(ParameterGrid(param_grid)) == ( ... [{'a': 1, 'b': True}, {'a': 1, 'b': False}, ... {'a': 2, 'b': True}, {'a': 2, 'b': False}]) True >>> grid = [{'kernel': ['linear']}, {'kernel': ['rbf'], 'gamma': [1, 10]}] >>> list(ParameterGrid(grid)) == [{'kernel': 'linear'}, ... {'kernel': 'rbf', 'gamma': 1}, ... {'kernel': 'rbf', 'gamma': 10}] True >>> ParameterGrid(grid)[1] == {'kernel': 'rbf', 'gamma': 1} True See also -------- :class:`GridSearchCV`: Uses :class:`ParameterGrid` to perform a full parallelized parameter search. """ def __init__(self, param_grid): if isinstance(param_grid, Mapping): # wrap dictionary in a singleton list to support either dict # or list of dicts param_grid = [param_grid] self.param_grid = param_grid def __iter__(self): """Iterate over the points in the grid. Returns ------- params : iterator over dict of string to any Yields dictionaries mapping each estimator parameter to one of its allowed values. """ for p in self.param_grid: # Always sort the keys of a dictionary, for reproducibility items = sorted(p.items()) if not items: yield {} else: keys, values = zip(*items) for v in product(*values): params = dict(zip(keys, v)) yield params def __len__(self): """Number of points on the grid.""" # Product function that can handle iterables (np.product can't). product = partial(reduce, operator.mul) return sum(product(len(v) for v in p.values()) if p else 1 for p in self.param_grid) def __getitem__(self, ind): """Get the parameters that would be ``ind``th in iteration Parameters ---------- ind : int The iteration index Returns ------- params : dict of string to any Equal to list(self)[ind] """ # This is used to make discrete sampling without replacement memory # efficient. for sub_grid in self.param_grid: # XXX: could memoize information used here if not sub_grid: if ind == 0: return {} else: ind -= 1 continue # Reverse so most frequent cycling parameter comes first keys, values_lists = zip(*sorted(sub_grid.items())[::-1]) sizes = [len(v_list) for v_list in values_lists] total = np.product(sizes) if ind >= total: # Try the next grid ind -= total else: out = {} for key, v_list, n in zip(keys, values_lists, sizes): ind, offset = divmod(ind, n) out[key] = v_list[offset] return out raise IndexError('ParameterGrid index out of range') class ParameterSampler(object): """Generator on parameters sampled from given distributions. Non-deterministic iterable over random candidate combinations for hyper- parameter search. If all parameters are presented as a list, sampling without replacement is performed. If at least one parameter is given as a distribution, sampling with replacement is used. It is highly recommended to use continuous distributions for continuous parameters. Note that before SciPy 0.16, the ``scipy.stats.distributions`` do not accept a custom RNG instance and always use the singleton RNG from ``numpy.random``. Hence setting ``random_state`` will not guarantee a deterministic iteration whenever ``scipy.stats`` distributions are used to define the parameter search space. Deterministic behavior is however guaranteed from SciPy 0.16 onwards. Read more in the :ref:`User Guide `. Parameters ---------- param_distributions : dict Dictionary where the keys are parameters and values are distributions from which a parameter is to be sampled. Distributions either have to provide a ``rvs`` function to sample from them, or can be given as a list of values, where a uniform distribution is assumed. n_iter : integer Number of parameter settings that are produced. random_state : int, RandomState instance or None, optional (default=None) Pseudo random number generator state used for random uniform sampling from lists of possible values instead of scipy.stats distributions. If int, random_state is the seed used by the random number generator; If RandomState instance, random_state is the random number generator; If None, the random number generator is the RandomState instance used by `np.random`. Returns ------- params : dict of string to any **Yields** dictionaries mapping each estimator parameter to as sampled value. Examples -------- >>> from sklearn.model_selection import ParameterSampler >>> from scipy.stats.distributions import expon >>> import numpy as np >>> np.random.seed(0) >>> param_grid = {'a':[1, 2], 'b': expon()} >>> param_list = list(ParameterSampler(param_grid, n_iter=4)) >>> rounded_list = [dict((k, round(v, 6)) for (k, v) in d.items()) ... for d in param_list] >>> rounded_list == [{'b': 0.89856, 'a': 1}, ... {'b': 0.923223, 'a': 1}, ... {'b': 1.878964, 'a': 2}, ... {'b': 1.038159, 'a': 2}] True """ def __init__(self, param_distributions, n_iter, random_state=None): self.param_distributions = param_distributions self.n_iter = n_iter self.random_state = random_state def __iter__(self): # check if all distributions are given as lists # in this case we want to sample without replacement all_lists = np.all([not hasattr(v, "rvs") for v in self.param_distributions.values()]) rnd = check_random_state(self.random_state) if all_lists: # look up sampled parameter settings in parameter grid param_grid = ParameterGrid(self.param_distributions) grid_size = len(param_grid) if grid_size < self.n_iter: raise ValueError( "The total space of parameters %d is smaller " "than n_iter=%d. For exhaustive searches, use " "GridSearchCV." % (grid_size, self.n_iter)) for i in sample_without_replacement(grid_size, self.n_iter, random_state=rnd): yield param_grid[i] else: # Always sort the keys of a dictionary, for reproducibility items = sorted(self.param_distributions.items()) for _ in six.moves.range(self.n_iter): params = dict() for k, v in items: if hasattr(v, "rvs"): if sp_version < (0, 16): params[k] = v.rvs() else: params[k] = v.rvs(random_state=rnd) else: params[k] = v[rnd.randint(len(v))] yield params def __len__(self): """Number of points that will be sampled.""" return self.n_iter def fit_grid_point(X, y, estimator, parameters, train, test, scorer, verbose, error_score='raise', **fit_params): """Run fit on one set of parameters. Parameters ---------- X : array-like, sparse matrix or list Input data. y : array-like or None Targets for input data. estimator : estimator object A object of that type is instantiated for each grid point. This is assumed to implement the scikit-learn estimator interface. Either estimator needs to provide a ``score`` function, or ``scoring`` must be passed. parameters : dict Parameters to be set on estimator for this grid point. train : ndarray, dtype int or bool Boolean mask or indices for training set. test : ndarray, dtype int or bool Boolean mask or indices for test set. scorer : callable or None The scorer callable object / function must have its signature as ``scorer(estimator, X, y)``. If ``None`` the estimator's default scorer is used. verbose : int Verbosity level. **fit_params : kwargs Additional parameter passed to the fit function of the estimator. error_score : 'raise' (default) or numeric Value to assign to the score if an error occurs in estimator fitting. If set to 'raise', the error is raised. If a numeric value is given, FitFailedWarning is raised. This parameter does not affect the refit step, which will always raise the error. Returns ------- score : float Score of this parameter setting on given training / test split. parameters : dict The parameters that have been evaluated. n_samples_test : int Number of test samples in this split. """ # NOTE we are not using the return value as the scorer by itself should be # validated before. We use check_scoring only to reject multimetric scorer check_scoring(estimator, scorer) scores, n_samples_test = _fit_and_score(estimator, X, y, scorer, train, test, verbose, parameters, fit_params=fit_params, return_n_test_samples=True, error_score=error_score) return scores, parameters, n_samples_test def _check_param_grid(param_grid): if hasattr(param_grid, 'items'): param_grid = [param_grid] for p in param_grid: for name, v in p.items(): if isinstance(v, np.ndarray) and v.ndim > 1: raise ValueError("Parameter array should be one-dimensional.") if (isinstance(v, six.string_types) or not isinstance(v, (np.ndarray, Sequence))): raise ValueError("Parameter values for parameter ({0}) need " "to be a sequence(but not a string) or" " np.ndarray.".format(name)) if len(v) == 0: raise ValueError("Parameter values for parameter ({0}) need " "to be a non-empty sequence.".format(name)) # XXX Remove in 0.20 class _CVScoreTuple (namedtuple('_CVScoreTuple', ('parameters', 'mean_validation_score', 'cv_validation_scores'))): # A raw namedtuple is very memory efficient as it packs the attributes # in a struct to get rid of the __dict__ of attributes in particular it # does not copy the string for the keys on each instance. # By deriving a namedtuple class just to introduce the __repr__ method we # would also reintroduce the __dict__ on the instance. By telling the # Python interpreter that this subclass uses static __slots__ instead of # dynamic attributes. Furthermore we don't need any additional slot in the # subclass so we set __slots__ to the empty tuple. __slots__ = () def __repr__(self): """Simple custom repr to summarize the main info""" return "mean: {0:.5f}, std: {1:.5f}, params: {2}".format( self.mean_validation_score, np.std(self.cv_validation_scores), self.parameters) class BaseSearchCV(six.with_metaclass(ABCMeta, BaseEstimator, MetaEstimatorMixin)): """Base class for hyper parameter search with cross-validation.""" @abstractmethod def __init__(self, estimator, scoring=None, fit_params=None, n_jobs=1, iid=True, refit=True, cv=None, verbose=0, pre_dispatch='2*n_jobs', error_score='raise', return_train_score=True): self.scoring = scoring self.estimator = estimator self.n_jobs = n_jobs self.fit_params = fit_params self.iid = iid self.refit = refit self.cv = cv self.verbose = verbose self.pre_dispatch = pre_dispatch self.error_score = error_score self.return_train_score = return_train_score @property def _estimator_type(self): return self.estimator._estimator_type def score(self, X, y=None): """Returns the score on the given data, if the estimator has been refit. This uses the score defined by ``scoring`` where provided, and the ``best_estimator_.score`` method otherwise. Parameters ---------- X : array-like, shape = [n_samples, n_features] Input data, where n_samples is the number of samples and n_features is the number of features. y : array-like, shape = [n_samples] or [n_samples, n_output], optional Target relative to X for classification or regression; None for unsupervised learning. Returns ------- score : float """ self._check_is_fitted('score') if self.scorer_ is None: raise ValueError("No score function explicitly defined, " "and the estimator doesn't provide one %s" % self.best_estimator_) score = self.scorer_[self.refit] if self.multimetric_ else self.scorer_ return score(self.best_estimator_, X, y) def _check_is_fitted(self, method_name): if not self.refit: raise NotFittedError('This %s instance was initialized ' 'with refit=False. %s is ' 'available only after refitting on the best ' 'parameters. You can refit an estimator ' 'manually using the ``best_parameters_`` ' 'attribute' % (type(self).__name__, method_name)) else: check_is_fitted(self, 'best_estimator_') @if_delegate_has_method(delegate=('best_estimator_', 'estimator')) def predict(self, X): """Call predict on the estimator with the best found parameters. Only available if ``refit=True`` and the underlying estimator supports ``predict``. Parameters ----------- X : indexable, length n_samples Must fulfill the input assumptions of the underlying estimator. """ self._check_is_fitted('predict') return self.best_estimator_.predict(X) @if_delegate_has_method(delegate=('best_estimator_', 'estimator')) def predict_proba(self, X): """Call predict_proba on the estimator with the best found parameters. Only available if ``refit=True`` and the underlying estimator supports ``predict_proba``. Parameters ----------- X : indexable, length n_samples Must fulfill the input assumptions of the underlying estimator. """ self._check_is_fitted('predict_proba') return self.best_estimator_.predict_proba(X) @if_delegate_has_method(delegate=('best_estimator_', 'estimator')) def predict_log_proba(self, X): """Call predict_log_proba on the estimator with the best found parameters. Only available if ``refit=True`` and the underlying estimator supports ``predict_log_proba``. Parameters ----------- X : indexable, length n_samples Must fulfill the input assumptions of the underlying estimator. """ self._check_is_fitted('predict_log_proba') return self.best_estimator_.predict_log_proba(X) @if_delegate_has_method(delegate=('best_estimator_', 'estimator')) def decision_function(self, X): """Call decision_function on the estimator with the best found parameters. Only available if ``refit=True`` and the underlying estimator supports ``decision_function``. Parameters ----------- X : indexable, length n_samples Must fulfill the input assumptions of the underlying estimator. """ self._check_is_fitted('decision_function') return self.best_estimator_.decision_function(X) @if_delegate_has_method(delegate=('best_estimator_', 'estimator')) def transform(self, X): """Call transform on the estimator with the best found parameters. Only available if the underlying estimator supports ``transform`` and ``refit=True``. Parameters ----------- X : indexable, length n_samples Must fulfill the input assumptions of the underlying estimator. """ self._check_is_fitted('transform') return self.best_estimator_.transform(X) @if_delegate_has_method(delegate=('best_estimator_', 'estimator')) def inverse_transform(self, Xt): """Call inverse_transform on the estimator with the best found params. Only available if the underlying estimator implements ``inverse_transform`` and ``refit=True``. Parameters ----------- Xt : indexable, length n_samples Must fulfill the input assumptions of the underlying estimator. """ self._check_is_fitted('inverse_transform') return self.best_estimator_.inverse_transform(Xt) @property def classes_(self): self._check_is_fitted("classes_") return self.best_estimator_.classes_ def fit(self, X, y=None, groups=None, **fit_params): """Run fit with all sets of parameters. Parameters ---------- X : array-like, shape = [n_samples, n_features] Training vector, where n_samples is the number of samples and n_features is the number of features. y : array-like, shape = [n_samples] or [n_samples, n_output], optional Target relative to X for classification or regression; None for unsupervised learning. groups : array-like, with shape (n_samples,), optional Group labels for the samples used while splitting the dataset into train/test set. **fit_params : dict of string -> object Parameters passed to the ``fit`` method of the estimator """ if self.fit_params is not None: warnings.warn('"fit_params" as a constructor argument was ' 'deprecated in version 0.19 and will be removed ' 'in version 0.21. Pass fit parameters to the ' '"fit" method instead.', DeprecationWarning) if fit_params: warnings.warn('Ignoring fit_params passed as a constructor ' 'argument in favor of keyword arguments to ' 'the "fit" method.', RuntimeWarning) else: fit_params = self.fit_params estimator = self.estimator cv = check_cv(self.cv, y, classifier=is_classifier(estimator)) scorers, self.multimetric_ = _check_multimetric_scoring( self.estimator, scoring=self.scoring) if self.multimetric_: if self.refit is not False and ( not isinstance(self.refit, six.string_types) or # This will work for both dict / list (tuple) self.refit not in scorers): raise ValueError("For multi-metric scoring, the parameter " "refit must be set to a scorer key " "to refit an estimator with the best " "parameter setting on the whole data and " "make the best_* attributes " "available for that metric. If this is not " "needed, refit should be set to False " "explicitly. %r was passed." % self.refit) else: refit_metric = self.refit else: refit_metric = 'score' X, y, groups = indexable(X, y, groups) n_splits = cv.get_n_splits(X, y, groups) # Regenerate parameter iterable for each fit candidate_params = list(self._get_param_iterator()) n_candidates = len(candidate_params) if self.verbose > 0: print("Fitting {0} folds for each of {1} candidates, totalling" " {2} fits".format(n_splits, n_candidates, n_candidates * n_splits)) base_estimator = clone(self.estimator) pre_dispatch = self.pre_dispatch out = Parallel( n_jobs=self.n_jobs, verbose=self.verbose, pre_dispatch=pre_dispatch )(delayed(_fit_and_score)(clone(base_estimator), X, y, scorers, train, test, self.verbose, parameters, fit_params=fit_params, return_train_score=self.return_train_score, return_n_test_samples=True, return_times=True, return_parameters=False, error_score=self.error_score) for parameters, (train, test) in product(candidate_params, cv.split(X, y, groups))) # if one choose to see train score, "out" will contain train score info if self.return_train_score: (train_score_dicts, test_score_dicts, test_sample_counts, fit_time, score_time) = zip(*out) else: (test_score_dicts, test_sample_counts, fit_time, score_time) = zip(*out) # test_score_dicts and train_score dicts are lists of dictionaries and # we make them into dict of lists test_scores = _aggregate_score_dicts(test_score_dicts) if self.return_train_score: train_scores = _aggregate_score_dicts(train_score_dicts) # TODO: replace by a dict in 0.21 results = (DeprecationDict() if self.return_train_score == 'warn' else {}) def _store(key_name, array, weights=None, splits=False, rank=False): """A small helper to store the scores/times to the cv_results_""" # When iterated first by splits, then by parameters # We want `array` to have `n_candidates` rows and `n_splits` cols. array = np.array(array, dtype=np.float64).reshape(n_candidates, n_splits) if splits: for split_i in range(n_splits): # Uses closure to alter the results results["split%d_%s" % (split_i, key_name)] = array[:, split_i] array_means = np.average(array, axis=1, weights=weights) results['mean_%s' % key_name] = array_means # Weighted std is not directly available in numpy array_stds = np.sqrt(np.average((array - array_means[:, np.newaxis]) ** 2, axis=1, weights=weights)) results['std_%s' % key_name] = array_stds if rank: results["rank_%s" % key_name] = np.asarray( rankdata(-array_means, method='min'), dtype=np.int32) _store('fit_time', fit_time) _store('score_time', score_time) # Use one MaskedArray and mask all the places where the param is not # applicable for that candidate. Use defaultdict as each candidate may # not contain all the params param_results = defaultdict(partial(MaskedArray, np.empty(n_candidates,), mask=True, dtype=object)) for cand_i, params in enumerate(candidate_params): for name, value in params.items(): # An all masked empty array gets created for the key # `"param_%s" % name` at the first occurence of `name`. # Setting the value at an index also unmasks that index param_results["param_%s" % name][cand_i] = value results.update(param_results) # Store a list of param dicts at the key 'params' results['params'] = candidate_params # NOTE test_sample counts (weights) remain the same for all candidates test_sample_counts = np.array(test_sample_counts[:n_splits], dtype=np.int) for scorer_name in scorers.keys(): # Computed the (weighted) mean and std for test scores alone _store('test_%s' % scorer_name, test_scores[scorer_name], splits=True, rank=True, weights=test_sample_counts if self.iid else None) if self.return_train_score: prev_keys = set(results.keys()) _store('train_%s' % scorer_name, train_scores[scorer_name], splits=True) if self.return_train_score == 'warn': for key in set(results.keys()) - prev_keys: message = ( 'You are accessing a training score ({!r}), ' 'which will not be available by default ' 'any more in 0.21. If you need training scores, ' 'please set return_train_score=True').format(key) # warn on key access results.add_warning(key, message, FutureWarning) # For multi-metric evaluation, store the best_index_, best_params_ and # best_score_ iff refit is one of the scorer names # In single metric evaluation, refit_metric is "score" if self.refit or not self.multimetric_: self.best_index_ = results["rank_test_%s" % refit_metric].argmin() self.best_params_ = candidate_params[self.best_index_] self.best_score_ = results["mean_test_%s" % refit_metric][ self.best_index_] if self.refit: self.best_estimator_ = clone(base_estimator).set_params( **self.best_params_) if y is not None: self.best_estimator_.fit(X, y, **fit_params) else: self.best_estimator_.fit(X, **fit_params) # Store the only scorer not as a dict for single metric evaluation self.scorer_ = scorers if self.multimetric_ else scorers['score'] self.cv_results_ = results self.n_splits_ = n_splits return self @property def grid_scores_(self): check_is_fitted(self, 'cv_results_') if self.multimetric_: raise AttributeError("grid_scores_ attribute is not available for" " multi-metric evaluation.") warnings.warn( "The grid_scores_ attribute was deprecated in version 0.18" " in favor of the more elaborate cv_results_ attribute." " The grid_scores_ attribute will not be available from 0.20", DeprecationWarning) grid_scores = list() for i, (params, mean, std) in enumerate(zip( self.cv_results_['params'], self.cv_results_['mean_test_score'], self.cv_results_['std_test_score'])): scores = np.array(list(self.cv_results_['split%d_test_score' % s][i] for s in range(self.n_splits_)), dtype=np.float64) grid_scores.append(_CVScoreTuple(params, mean, scores)) return grid_scores class GridSearchCV(BaseSearchCV): """Exhaustive search over specified parameter values for an estimator. Important members are fit, predict. GridSearchCV implements a "fit" and a "score" method. It also implements "predict", "predict_proba", "decision_function", "transform" and "inverse_transform" if they are implemented in the estimator used. The parameters of the estimator used to apply these methods are optimized by cross-validated grid-search over a parameter grid. Read more in the :ref:`User Guide `. Parameters ---------- estimator : estimator object. This is assumed to implement the scikit-learn estimator interface. Either estimator needs to provide a ``score`` function, or ``scoring`` must be passed. param_grid : dict or list of dictionaries Dictionary with parameters names (string) as keys and lists of parameter settings to try as values, or a list of such dictionaries, in which case the grids spanned by each dictionary in the list are explored. This enables searching over any sequence of parameter settings. scoring : string, callable, list/tuple, dict or None, default: None A single string (see :ref:`scoring_parameter`) or a callable (see :ref:`scoring`) to evaluate the predictions on the test set. For evaluating multiple metrics, either give a list of (unique) strings or a dict with names as keys and callables as values. NOTE that when using custom scorers, each scorer should return a single value. Metric functions returning a list/array of values can be wrapped into multiple scorers that return one value each. See :ref:`multimetric_grid_search` for an example. If None, the estimator's default scorer (if available) is used. fit_params : dict, optional Parameters to pass to the fit method. .. deprecated:: 0.19 ``fit_params`` as a constructor argument was deprecated in version 0.19 and will be removed in version 0.21. Pass fit parameters to the ``fit`` method instead. n_jobs : int, default=1 Number of jobs to run in parallel. pre_dispatch : int, or string, optional Controls the number of jobs that get dispatched during parallel execution. Reducing this number can be useful to avoid an explosion of memory consumption when more jobs get dispatched than CPUs can process. This parameter can be: - None, in which case all the jobs are immediately created and spawned. Use this for lightweight and fast-running jobs, to avoid delays due to on-demand spawning of the jobs - An int, giving the exact number of total jobs that are spawned - A string, giving an expression as a function of n_jobs, as in '2*n_jobs' iid : boolean, default=True If True, the data is assumed to be identically distributed across the folds, and the loss minimized is the total loss per sample, and not the mean loss across the folds. cv : int, cross-validation generator or an iterable, optional Determines the cross-validation splitting strategy. Possible inputs for cv are: - None, to use the default 3-fold cross validation, - integer, to specify the number of folds in a `(Stratified)KFold`, - An object to be used as a cross-validation generator. - An iterable yielding train, test splits. For integer/None inputs, if the estimator is a classifier and ``y`` is either binary or multiclass, :class:`StratifiedKFold` is used. In all other cases, :class:`KFold` is used. Refer :ref:`User Guide ` for the various cross-validation strategies that can be used here. refit : boolean, or string, default=True Refit an estimator using the best found parameters on the whole dataset. For multiple metric evaluation, this needs to be a string denoting the scorer is used to find the best parameters for refitting the estimator at the end. The refitted estimator is made available at the ``best_estimator_`` attribute and permits using ``predict`` directly on this ``GridSearchCV`` instance. Also for multiple metric evaluation, the attributes ``best_index_``, ``best_score_`` and ``best_parameters_`` will only be available if ``refit`` is set and all of them will be determined w.r.t this specific scorer. See ``scoring`` parameter to know more about multiple metric evaluation. verbose : integer Controls the verbosity: the higher, the more messages. error_score : 'raise' (default) or numeric Value to assign to the score if an error occurs in estimator fitting. If set to 'raise', the error is raised. If a numeric value is given, FitFailedWarning is raised. This parameter does not affect the refit step, which will always raise the error. return_train_score : boolean, optional If ``False``, the ``cv_results_`` attribute will not include training scores. Current default is ``'warn'``, which behaves as ``True`` in addition to raising a warning when a training score is looked up. That default will be changed to ``False`` in 0.21. Computing training scores is used to get insights on how different parameter settings impact the overfitting/underfitting trade-off. However computing the scores on the training set can be computationally expensive and is not strictly required to select the parameters that yield the best generalization performance. Examples -------- >>> from sklearn import svm, datasets >>> from sklearn.model_selection import GridSearchCV >>> iris = datasets.load_iris() >>> parameters = {'kernel':('linear', 'rbf'), 'C':[1, 10]} >>> svc = svm.SVC() >>> clf = GridSearchCV(svc, parameters) >>> clf.fit(iris.data, iris.target) ... # doctest: +NORMALIZE_WHITESPACE +ELLIPSIS GridSearchCV(cv=None, error_score=..., estimator=SVC(C=1.0, cache_size=..., class_weight=..., coef0=..., decision_function_shape='ovr', degree=..., gamma=..., kernel='rbf', max_iter=-1, probability=False, random_state=None, shrinking=True, tol=..., verbose=False), fit_params=None, iid=..., n_jobs=1, param_grid=..., pre_dispatch=..., refit=..., return_train_score=..., scoring=..., verbose=...) >>> sorted(clf.cv_results_.keys()) ... # doctest: +NORMALIZE_WHITESPACE +ELLIPSIS ['mean_fit_time', 'mean_score_time', 'mean_test_score',... 'mean_train_score', 'param_C', 'param_kernel', 'params',... 'rank_test_score', 'split0_test_score',... 'split0_train_score', 'split1_test_score', 'split1_train_score',... 'split2_test_score', 'split2_train_score',... 'std_fit_time', 'std_score_time', 'std_test_score', 'std_train_score'...] Attributes ---------- cv_results_ : dict of numpy (masked) ndarrays A dict with keys as column headers and values as columns, that can be imported into a pandas ``DataFrame``. For instance the below given table +------------+-----------+------------+-----------------+---+---------+ |param_kernel|param_gamma|param_degree|split0_test_score|...|rank_t...| +============+===========+============+=================+===+=========+ | 'poly' | -- | 2 | 0.8 |...| 2 | +------------+-----------+------------+-----------------+---+---------+ | 'poly' | -- | 3 | 0.7 |...| 4 | +------------+-----------+------------+-----------------+---+---------+ | 'rbf' | 0.1 | -- | 0.8 |...| 3 | +------------+-----------+------------+-----------------+---+---------+ | 'rbf' | 0.2 | -- | 0.9 |...| 1 | +------------+-----------+------------+-----------------+---+---------+ will be represented by a ``cv_results_`` dict of:: { 'param_kernel': masked_array(data = ['poly', 'poly', 'rbf', 'rbf'], mask = [False False False False]...) 'param_gamma': masked_array(data = [-- -- 0.1 0.2], mask = [ True True False False]...), 'param_degree': masked_array(data = [2.0 3.0 -- --], mask = [False False True True]...), 'split0_test_score' : [0.8, 0.7, 0.8, 0.9], 'split1_test_score' : [0.82, 0.5, 0.7, 0.78], 'mean_test_score' : [0.81, 0.60, 0.75, 0.82], 'std_test_score' : [0.02, 0.01, 0.03, 0.03], 'rank_test_score' : [2, 4, 3, 1], 'split0_train_score' : [0.8, 0.9, 0.7], 'split1_train_score' : [0.82, 0.5, 0.7], 'mean_train_score' : [0.81, 0.7, 0.7], 'std_train_score' : [0.03, 0.03, 0.04], 'mean_fit_time' : [0.73, 0.63, 0.43, 0.49], 'std_fit_time' : [0.01, 0.02, 0.01, 0.01], 'mean_score_time' : [0.007, 0.06, 0.04, 0.04], 'std_score_time' : [0.001, 0.002, 0.003, 0.005], 'params' : [{'kernel': 'poly', 'degree': 2}, ...], } NOTE The key ``'params'`` is used to store a list of parameter settings dicts for all the parameter candidates. The ``mean_fit_time``, ``std_fit_time``, ``mean_score_time`` and ``std_score_time`` are all in seconds. For multi-metric evaluation, the scores for all the scorers are available in the ``cv_results_`` dict at the keys ending with that scorer's name (``'_'``) instead of ``'_score'`` shown above. ('split0_test_precision', 'mean_train_precision' etc.) best_estimator_ : estimator or dict Estimator that was chosen by the search, i.e. estimator which gave highest score (or smallest loss if specified) on the left out data. Not available if ``refit=False``. See ``refit`` parameter for more information on allowed values. best_score_ : float Mean cross-validated score of the best_estimator For multi-metric evaluation, this is present only if ``refit`` is specified. best_params_ : dict Parameter setting that gave the best results on the hold out data. For multi-metric evaluation, this is present only if ``refit`` is specified. best_index_ : int The index (of the ``cv_results_`` arrays) which corresponds to the best candidate parameter setting. The dict at ``search.cv_results_['params'][search.best_index_]`` gives the parameter setting for the best model, that gives the highest mean score (``search.best_score_``). For multi-metric evaluation, this is present only if ``refit`` is specified. scorer_ : function or a dict Scorer function used on the held out data to choose the best parameters for the model. For multi-metric evaluation, this attribute holds the validated ``scoring`` dict which maps the scorer key to the scorer callable. n_splits_ : int The number of cross-validation splits (folds/iterations). Notes ------ The parameters selected are those that maximize the score of the left out data, unless an explicit score is passed in which case it is used instead. If `n_jobs` was set to a value higher than one, the data is copied for each point in the grid (and not `n_jobs` times). This is done for efficiency reasons if individual jobs take very little time, but may raise errors if the dataset is large and not enough memory is available. A workaround in this case is to set `pre_dispatch`. Then, the memory is copied only `pre_dispatch` many times. A reasonable value for `pre_dispatch` is `2 * n_jobs`. See Also --------- :class:`ParameterGrid`: generates all the combinations of a hyperparameter grid. :func:`sklearn.model_selection.train_test_split`: utility function to split the data into a development set usable for fitting a GridSearchCV instance and an evaluation set for its final evaluation. :func:`sklearn.metrics.make_scorer`: Make a scorer from a performance metric or loss function. """ def __init__(self, estimator, param_grid, scoring=None, fit_params=None, n_jobs=1, iid=True, refit=True, cv=None, verbose=0, pre_dispatch='2*n_jobs', error_score='raise', return_train_score="warn"): super(GridSearchCV, self).__init__( estimator=estimator, scoring=scoring, fit_params=fit_params, n_jobs=n_jobs, iid=iid, refit=refit, cv=cv, verbose=verbose, pre_dispatch=pre_dispatch, error_score=error_score, return_train_score=return_train_score) self.param_grid = param_grid _check_param_grid(param_grid) def _get_param_iterator(self): """Return ParameterGrid instance for the given param_grid""" return ParameterGrid(self.param_grid) class RandomizedSearchCV(BaseSearchCV): """Randomized search on hyper parameters. RandomizedSearchCV implements a "fit" and a "score" method. It also implements "predict", "predict_proba", "decision_function", "transform" and "inverse_transform" if they are implemented in the estimator used. The parameters of the estimator used to apply these methods are optimized by cross-validated search over parameter settings. In contrast to GridSearchCV, not all parameter values are tried out, but rather a fixed number of parameter settings is sampled from the specified distributions. The number of parameter settings that are tried is given by n_iter. If all parameters are presented as a list, sampling without replacement is performed. If at least one parameter is given as a distribution, sampling with replacement is used. It is highly recommended to use continuous distributions for continuous parameters. Read more in the :ref:`User Guide `. Parameters ---------- estimator : estimator object. A object of that type is instantiated for each grid point. This is assumed to implement the scikit-learn estimator interface. Either estimator needs to provide a ``score`` function, or ``scoring`` must be passed. param_distributions : dict Dictionary with parameters names (string) as keys and distributions or lists of parameters to try. Distributions must provide a ``rvs`` method for sampling (such as those from scipy.stats.distributions). If a list is given, it is sampled uniformly. n_iter : int, default=10 Number of parameter settings that are sampled. n_iter trades off runtime vs quality of the solution. scoring : string, callable, list/tuple, dict or None, default: None A single string (see :ref:`scoring_parameter`) or a callable (see :ref:`scoring`) to evaluate the predictions on the test set. For evaluating multiple metrics, either give a list of (unique) strings or a dict with names as keys and callables as values. NOTE that when using custom scorers, each scorer should return a single value. Metric functions returning a list/array of values can be wrapped into multiple scorers that return one value each. See :ref:`multimetric_grid_search` for an example. If None, the estimator's default scorer (if available) is used. fit_params : dict, optional Parameters to pass to the fit method. .. deprecated:: 0.19 ``fit_params`` as a constructor argument was deprecated in version 0.19 and will be removed in version 0.21. Pass fit parameters to the ``fit`` method instead. n_jobs : int, default=1 Number of jobs to run in parallel. pre_dispatch : int, or string, optional Controls the number of jobs that get dispatched during parallel execution. Reducing this number can be useful to avoid an explosion of memory consumption when more jobs get dispatched than CPUs can process. This parameter can be: - None, in which case all the jobs are immediately created and spawned. Use this for lightweight and fast-running jobs, to avoid delays due to on-demand spawning of the jobs - An int, giving the exact number of total jobs that are spawned - A string, giving an expression as a function of n_jobs, as in '2*n_jobs' iid : boolean, default=True If True, the data is assumed to be identically distributed across the folds, and the loss minimized is the total loss per sample, and not the mean loss across the folds. cv : int, cross-validation generator or an iterable, optional Determines the cross-validation splitting strategy. Possible inputs for cv are: - None, to use the default 3-fold cross validation, - integer, to specify the number of folds in a `(Stratified)KFold`, - An object to be used as a cross-validation generator. - An iterable yielding train, test splits. For integer/None inputs, if the estimator is a classifier and ``y`` is either binary or multiclass, :class:`StratifiedKFold` is used. In all other cases, :class:`KFold` is used. Refer :ref:`User Guide ` for the various cross-validation strategies that can be used here. refit : boolean, or string default=True Refit an estimator using the best found parameters on the whole dataset. For multiple metric evaluation, this needs to be a string denoting the scorer that would be used to find the best parameters for refitting the estimator at the end. The refitted estimator is made available at the ``best_estimator_`` attribute and permits using ``predict`` directly on this ``RandomizedSearchCV`` instance. Also for multiple metric evaluation, the attributes ``best_index_``, ``best_score_`` and ``best_parameters_`` will only be available if ``refit`` is set and all of them will be determined w.r.t this specific scorer. See ``scoring`` parameter to know more about multiple metric evaluation. verbose : integer Controls the verbosity: the higher, the more messages. random_state : int, RandomState instance or None, optional, default=None Pseudo random number generator state used for random uniform sampling from lists of possible values instead of scipy.stats distributions. If int, random_state is the seed used by the random number generator; If RandomState instance, random_state is the random number generator; If None, the random number generator is the RandomState instance used by `np.random`. error_score : 'raise' (default) or numeric Value to assign to the score if an error occurs in estimator fitting. If set to 'raise', the error is raised. If a numeric value is given, FitFailedWarning is raised. This parameter does not affect the refit step, which will always raise the error. return_train_score : boolean, optional If ``False``, the ``cv_results_`` attribute will not include training scores. Current default is ``'warn'``, which behaves as ``True`` in addition to raising a warning when a training score is looked up. That default will be changed to ``False`` in 0.21. Computing training scores is used to get insights on how different parameter settings impact the overfitting/underfitting trade-off. However computing the scores on the training set can be computationally expensive and is not strictly required to select the parameters that yield the best generalization performance. Attributes ---------- cv_results_ : dict of numpy (masked) ndarrays A dict with keys as column headers and values as columns, that can be imported into a pandas ``DataFrame``. For instance the below given table +--------------+-------------+-------------------+---+---------------+ | param_kernel | param_gamma | split0_test_score |...|rank_test_score| +==============+=============+===================+===+===============+ | 'rbf' | 0.1 | 0.8 |...| 2 | +--------------+-------------+-------------------+---+---------------+ | 'rbf' | 0.2 | 0.9 |...| 1 | +--------------+-------------+-------------------+---+---------------+ | 'rbf' | 0.3 | 0.7 |...| 1 | +--------------+-------------+-------------------+---+---------------+ will be represented by a ``cv_results_`` dict of:: { 'param_kernel' : masked_array(data = ['rbf', 'rbf', 'rbf'], mask = False), 'param_gamma' : masked_array(data = [0.1 0.2 0.3], mask = False), 'split0_test_score' : [0.8, 0.9, 0.7], 'split1_test_score' : [0.82, 0.5, 0.7], 'mean_test_score' : [0.81, 0.7, 0.7], 'std_test_score' : [0.02, 0.2, 0.], 'rank_test_score' : [3, 1, 1], 'split0_train_score' : [0.8, 0.9, 0.7], 'split1_train_score' : [0.82, 0.5, 0.7], 'mean_train_score' : [0.81, 0.7, 0.7], 'std_train_score' : [0.03, 0.03, 0.04], 'mean_fit_time' : [0.73, 0.63, 0.43, 0.49], 'std_fit_time' : [0.01, 0.02, 0.01, 0.01], 'mean_score_time' : [0.007, 0.06, 0.04, 0.04], 'std_score_time' : [0.001, 0.002, 0.003, 0.005], 'params' : [{'kernel' : 'rbf', 'gamma' : 0.1}, ...], } NOTE The key ``'params'`` is used to store a list of parameter settings dicts for all the parameter candidates. The ``mean_fit_time``, ``std_fit_time``, ``mean_score_time`` and ``std_score_time`` are all in seconds. For multi-metric evaluation, the scores for all the scorers are available in the ``cv_results_`` dict at the keys ending with that scorer's name (``'_'``) instead of ``'_score'`` shown above. ('split0_test_precision', 'mean_train_precision' etc.) best_estimator_ : estimator or dict Estimator that was chosen by the search, i.e. estimator which gave highest score (or smallest loss if specified) on the left out data. Not available if ``refit=False``. For multi-metric evaluation, this attribute is present only if ``refit`` is specified. See ``refit`` parameter for more information on allowed values. best_score_ : float Mean cross-validated score of the best_estimator. For multi-metric evaluation, this is not available if ``refit`` is ``False``. See ``refit`` parameter for more information. best_params_ : dict Parameter setting that gave the best results on the hold out data. For multi-metric evaluation, this is not available if ``refit`` is ``False``. See ``refit`` parameter for more information. best_index_ : int The index (of the ``cv_results_`` arrays) which corresponds to the best candidate parameter setting. The dict at ``search.cv_results_['params'][search.best_index_]`` gives the parameter setting for the best model, that gives the highest mean score (``search.best_score_``). For multi-metric evaluation, this is not available if ``refit`` is ``False``. See ``refit`` parameter for more information. scorer_ : function or a dict Scorer function used on the held out data to choose the best parameters for the model. For multi-metric evaluation, this attribute holds the validated ``scoring`` dict which maps the scorer key to the scorer callable. n_splits_ : int The number of cross-validation splits (folds/iterations). Notes ----- The parameters selected are those that maximize the score of the held-out data, according to the scoring parameter. If `n_jobs` was set to a value higher than one, the data is copied for each parameter setting(and not `n_jobs` times). This is done for efficiency reasons if individual jobs take very little time, but may raise errors if the dataset is large and not enough memory is available. A workaround in this case is to set `pre_dispatch`. Then, the memory is copied only `pre_dispatch` many times. A reasonable value for `pre_dispatch` is `2 * n_jobs`. See Also -------- :class:`GridSearchCV`: Does exhaustive search over a grid of parameters. :class:`ParameterSampler`: A generator over parameter settins, constructed from param_distributions. """ def __init__(self, estimator, param_distributions, n_iter=10, scoring=None, fit_params=None, n_jobs=1, iid=True, refit=True, cv=None, verbose=0, pre_dispatch='2*n_jobs', random_state=None, error_score='raise', return_train_score="warn"): self.param_distributions = param_distributions self.n_iter = n_iter self.random_state = random_state super(RandomizedSearchCV, self).__init__( estimator=estimator, scoring=scoring, fit_params=fit_params, n_jobs=n_jobs, iid=iid, refit=refit, cv=cv, verbose=verbose, pre_dispatch=pre_dispatch, error_score=error_score, return_train_score=return_train_score) def _get_param_iterator(self): """Return ParameterSampler instance for the given distributions""" return ParameterSampler( self.param_distributions, self.n_iter, random_state=self.random_state)