估计器

此文件是 TPOT 库的一部分。

TPOT 当前版本由 Cedars-Sinai 开发，开发人员包括： - Pedro Henrique Ribeiro (https://github.com/perib, https://www.linkedin.com/in/pedro-ribeiro/) - Anil Saini (anil.saini@cshs.org) - Jose Hernandez (jgh9094@gmail.com) - Jay Moran (jay.moran@cshs.org) - Nicholas Matsumoto (nicholas.matsumoto@cshs.org) - Hyunjun Choi (hyunjun.choi@cshs.org) - Gabriel Ketron (gabriel.ketron@cshs.org) - Miguel E. Hernandez (miguel.e.hernandez@cshs.org) - Jason Moore (moorejh28@gmail.com)

TPOT 的原始版本主要由宾夕法尼亚大学开发，开发人员包括： - Randal S. Olson (rso@randalolson.com) - Weixuan Fu (weixuanf@upenn.edu) - Daniel Angell (dpa34@drexel.edu) - Jason Moore (moorejh28@gmail.com) - 以及许多慷慨的开源贡献者

TPOT 是自由软件：您可以根据自由软件基金会发布的 GNU 宽通用公共许可证（GNU Lesser General Public License）的条款重新分发和/或修改它，可以选择使用许可证的第 3 版，或（您选择）任何更高版本。

分发 TPOT 是希望它会有用，但没有任何担保；甚至不包括适销性或特定用途适用性的默示担保。有关更多详细信息，请参阅 GNU 宽通用公共许可证。

您应该已经随 TPOT 收到了 GNU 宽通用公共许可证 的副本。如果未收到，请访问 https://gnu.ac.cn/licenses/。

TPOTEstimator

基类：BaseEstimator

源代码位于 tpot/tpot_estimator/estimator.py

class TPOTEstimator(BaseEstimator):
    def __init__(self,  
                        search_space,
                        scorers,
                        scorers_weights,
                        classification,
                        cv = 10,
                        other_objective_functions=[],
                        other_objective_functions_weights = [],
                        objective_function_names = None,
                        bigger_is_better = True,

                        export_graphpipeline = False,
                        memory = None,

                        categorical_features = None,
                        preprocessing = False,
                        population_size = 50,
                        initial_population_size = None,
                        population_scaling = .5,
                        generations_until_end_population = 1,
                        generations = None,
                        max_time_mins=60,
                        max_eval_time_mins=10,
                        validation_strategy = "none",
                        validation_fraction = .2,
                        disable_label_encoder = False,

                        #early stopping parameters
                        early_stop = None,
                        scorers_early_stop_tol = 0.001,
                        other_objectives_early_stop_tol =None,
                        threshold_evaluation_pruning = None,
                        threshold_evaluation_scaling = .5,
                        selection_evaluation_pruning = None,
                        selection_evaluation_scaling = .5,
                        min_history_threshold = 20,

                        #evolver parameters
                        survival_percentage = 1,
                        crossover_probability=.2,
                        mutate_probability=.7,
                        mutate_then_crossover_probability=.05,
                        crossover_then_mutate_probability=.05,
                        survival_selector = survival_select_NSGA2,
                        parent_selector = tournament_selection_dominated,

                        #budget parameters
                        budget_range = None,
                        budget_scaling = .5,
                        generations_until_end_budget = 1,
                        stepwise_steps = 5,

                        #dask parameters
                        n_jobs=1,
                        memory_limit = None,
                        client = None,
                        processes = True,

                        #debugging and logging parameters
                        warm_start = False,
                        periodic_checkpoint_folder = None,
                        callback = None,

                        verbose = 0,
                        scatter = True,

                         # random seed for random number generator (rng)
                        random_state = None,

                        ):

        '''
        An sklearn baseestimator that uses genetic programming to optimize a pipeline.

        Parameters
        ----------
        search_space : (String, tpot.search_spaces.SearchSpace)
            - String : The default search space to use for the optimization.
            | String     | Description      |
            | :---        |    :----:   |
            | linear  | A linear pipeline with the structure of "Selector->(transformers+Passthrough)->(classifiers/regressors+Passthrough)->final classifier/regressor." For both the transformer and inner estimator layers, TPOT may choose one or more transformers/classifiers, or it may choose none. The inner classifier/regressor layer is optional. |
            | linear-light | Same search space as linear, but without the inner classifier/regressor layer and with a reduced set of faster running estimators. |
            | graph | TPOT will optimize a pipeline in the shape of a directed acyclic graph. The nodes of the graph can include selectors, scalers, transformers, or classifiers/regressors (inner classifiers/regressors can optionally be not included). This will return a custom GraphPipeline rather than an sklearn Pipeline. More details in Tutorial 6. |
            | graph-light | Same as graph search space, but without the inner classifier/regressors and with a reduced set of faster running estimators. |
            | mdr |TPOT will search over a series of feature selectors and Multifactor Dimensionality Reduction models to find a series of operators that maximize prediction accuracy. The TPOT MDR configuration is specialized for genome-wide association studies (GWAS), and is described in detail online here.

            Note that TPOT MDR may be slow to run because the feature selection routines are computationally expensive, especially on large datasets. |


            - SearchSpace : The search space to use for the optimization. This should be an instance of a SearchSpace.
                The search space to use for the optimization. This should be an instance of a SearchSpace.
                TPOT has groups of search spaces found in the following folders, tpot.search_spaces.nodes for the nodes in the pipeline and tpot.search_spaces.pipelines for the pipeline structure.

        scorers : (list, scorer)
            A scorer or list of scorers to be used in the cross-validation process.
            see https://scikit-learn.cn/stable/modules/model_evaluation.html

        scorers_weights : list
            A list of weights to be applied to the scorers during the optimization process.

        classification : bool
            If True, the problem is treated as a classification problem. If False, the problem is treated as a regression problem.
            Used to determine the CV strategy.

        cv : int, cross-validator
            - (int): Number of folds to use in the cross-validation process. By uses the sklearn.model_selection.KFold cross-validator for regression and StratifiedKFold for classification. In both cases, shuffled is set to True.
            - (sklearn.model_selection.BaseCrossValidator): A cross-validator to use in the cross-validation process.
                - max_depth (int): The maximum depth from any node to the root of the pipelines to be generated.

        other_objective_functions : list, default=[]
            A list of other objective functions to apply to the pipeline. The function takes a single parameter for the graphpipeline estimator and returns either a single score or a list of scores.

        other_objective_functions_weights : list, default=[]
            A list of weights to be applied to the other objective functions.

        objective_function_names : list, default=None
            A list of names to be applied to the objective functions. If None, will use the names of the objective functions.

        bigger_is_better : bool, default=True
            If True, the objective function is maximized. If False, the objective function is minimized. Use negative weights to reverse the direction.

        memory: Memory object or string, default=None
            If supplied, pipeline will cache each transformer after calling fit with joblib.Memory. This feature
            is used to avoid computing the fit transformers within a pipeline if the parameters
            and input data are identical with another fitted pipeline during optimization process.
            - String 'auto':
                TPOT uses memory caching with a temporary directory and cleans it up upon shutdown.
            - String path of a caching directory
                TPOT uses memory caching with the provided directory and TPOT does NOT clean
                the caching directory up upon shutdown. If the directory does not exist, TPOT will
                create it.
            - Memory object:
                TPOT uses the instance of joblib.Memory for memory caching,
                and TPOT does NOT clean the caching directory up upon shutdown.
            - None:
                TPOT does not use memory caching.              

        categorical_features: list or None
            Categorical columns to inpute and/or one hot encode during the preprocessing step. Used only if preprocessing is not False.
            - None : If None, TPOT will automatically use object columns in pandas dataframes as objects for one hot encoding in preprocessing.
            - List of categorical features. If X is a dataframe, this should be a list of column names. If X is a numpy array, this should be a list of column indices

        preprocessing : bool or BaseEstimator/Pipeline,
            EXPERIMENTAL - will be changed in future versions
            A pipeline that will be used to preprocess the data before CV. Note that the parameters for these steps are not optimized. Add them to the search space to be optimized.
            - bool : If True, will use a default preprocessing pipeline which includes imputation followed by one hot encoding.
            - Pipeline : If an instance of a pipeline is given, will use that pipeline as the preprocessing pipeline.

        population_size : int, default=50
            Size of the population

        initial_population_size : int, default=None
            Size of the initial population. If None, population_size will be used.

        population_scaling : int, default=0.5
            Scaling factor to use when determining how fast we move the threshold moves from the start to end percentile.

        generations_until_end_population : int, default=1
            Number of generations until the population size reaches population_size

        generations : int, default=50
            Number of generations to run

        max_time_mins : float, default=float("inf")
            Maximum time to run the optimization. If none or inf, will run until the end of the generations.

        max_eval_time_mins : float, default=5
            Maximum time to evaluate a single individual. If none or inf, there will be no time limit per evaluation.

        validation_strategy : str, default='none'
            EXPERIMENTAL The validation strategy to use for selecting the final pipeline from the population. TPOT may overfit the cross validation score. A second validation set can be used to select the final pipeline.
            - 'auto' : Automatically determine the validation strategy based on the dataset shape.
            - 'reshuffled' : Use the same data for cross validation and final validation, but with different splits for the folds. This is the default for small datasets.
            - 'split' : Use a separate validation set for final validation. Data will be split according to validation_fraction. This is the default for medium datasets.
            - 'none' : Do not use a separate validation set for final validation. Select based on the original cross-validation score. This is the default for large datasets.

        validation_fraction : float, default=0.2
          EXPERIMENTAL The fraction of the dataset to use for the validation set when validation_strategy is 'split'. Must be between 0 and 1.

        disable_label_encoder : bool, default=False
            If True, TPOT will check if the target needs to be relabeled to be sequential ints from 0 to N. This is necessary for XGBoost compatibility. If the labels need to be encoded, TPOT will use sklearn.preprocessing.LabelEncoder to encode the labels. The encoder can be accessed via the self.label_encoder_ attribute.
            If False, no additional label encoders will be used.

        early_stop : int, default=None
            Number of generations without improvement before early stopping. All objectives must have converged within the tolerance for this to be triggered. In general a value of around 5-20 is good.

        scorers_early_stop_tol :
            -list of floats
                list of tolerances for each scorer. If the difference between the best score and the current score is less than the tolerance, the individual is considered to have converged
                If an index of the list is None, that item will not be used for early stopping
            -int
                If an int is given, it will be used as the tolerance for all objectives

        other_objectives_early_stop_tol :
            -list of floats
                list of tolerances for each of the other objective function. If the difference between the best score and the current score is less than the tolerance, the individual is considered to have converged
                If an index of the list is None, that item will not be used for early stopping
            -int
                If an int is given, it will be used as the tolerance for all objectives

        threshold_evaluation_pruning : list [start, end], default=None
            starting and ending percentile to use as a threshold for the evaluation early stopping.
            Values between 0 and 100.

        threshold_evaluation_scaling : float [0,inf), default=0.5
            A scaling factor to use when determining how fast we move the threshold moves from the start to end percentile.
            Must be greater than zero. Higher numbers will move the threshold to the end faster.

        selection_evaluation_pruning : list, default=None
            A lower and upper percent of the population size to select each round of CV.
            Values between 0 and 1.

        selection_evaluation_scaling : float, default=0.5
            A scaling factor to use when determining how fast we move the threshold moves from the start to end percentile.
            Must be greater than zero. Higher numbers will move the threshold to the end faster.

        min_history_threshold : int, default=0
            The minimum number of previous scores needed before using threshold early stopping.

        survival_percentage : float, default=1
            Percentage of the population size to utilize for mutation and crossover at the beginning of the generation. The rest are discarded. Individuals are selected with the selector passed into survival_selector. The value of this parameter must be between 0 and 1, inclusive.
            For example, if the population size is 100 and the survival percentage is .5, 50 individuals will be selected with NSGA2 from the existing population. These will be used for mutation and crossover to generate the next 100 individuals for the next generation. The remainder are discarded from the live population. In the next generation, there will now be the 50 parents + the 100 individuals for a total of 150. Surivival percentage is based of the population size parameter and not the existing population size (current population size when using successive halving). Therefore, in the next generation we will still select 50 individuals from the currently existing 150.

        crossover_probability : float, default=.2
            Probability of generating a new individual by crossover between two individuals.

        mutate_probability : float, default=.7
            Probability of generating a new individual by crossover between one individuals.

        mutate_then_crossover_probability : float, default=.05
            Probability of generating a new individual by mutating two individuals followed by crossover.

        crossover_then_mutate_probability : float, default=.05
            Probability of generating a new individual by crossover between two individuals followed by a mutation of the resulting individual.

        survival_selector : function, default=survival_select_NSGA2
            Function to use to select individuals for survival. Must take a matrix of scores and return selected indexes.
            Used to selected population_size * survival_percentage individuals at the start of each generation to use for mutation and crossover.

        parent_selector : function, default=parent_select_NSGA2
            Function to use to select pairs parents for crossover and individuals for mutation. Must take a matrix of scores and return selected indexes.

        budget_range : list [start, end], default=None
            A starting and ending budget to use for the budget scaling.

        budget_scaling float : [0,1], default=0.5
            A scaling factor to use when determining how fast we move the budget from the start to end budget.

        generations_until_end_budget : int, default=1
            The number of generations to run before reaching the max budget.

        stepwise_steps : int, default=1
            The number of staircase steps to take when scaling the budget and population size.

        n_jobs : int, default=1
            Number of processes to run in parallel.

        memory_limit : str, default=None
            Memory limit for each job. See Dask [LocalCluster documentation](https://distributed.dask.org.cn/en/stable/api.html#distributed.Client) for more information.

        client : dask.distributed.Client, default=None
            A dask client to use for parallelization. If not None, this will override the n_jobs and memory_limit parameters. If None, will create a new client with num_workers=n_jobs and memory_limit=memory_limit.

        processes : bool, default=True
            If True, will use multiprocessing to parallelize the optimization process. If False, will use threading.
            True seems to perform better. However, False is required for interactive debugging.

        warm_start : bool, default=False
            If True, will use the continue the evolutionary algorithm from the last generation of the previous run.

        periodic_checkpoint_folder : str, default=None
            Folder to save the population to periodically. If None, no periodic saving will be done.
            If provided, training will resume from this checkpoint.

        callback : tpot.CallBackInterface, default=None
            Callback object. Not implemented

        verbose : int, default=1
            How much information to print during the optimization process. Higher values include the information from lower values.
            0. nothing
            1. progress bar

            3. best individual
            4. warnings
            >=5. full warnings trace
            6. evaluations progress bar. (Temporary: This used to be 2. Currently, using evaluation progress bar may prevent some instances were we terminate a generation early due to it reaching max_time_mins in the middle of a generation OR a pipeline failed to be terminated normally and we need to manually terminate it.)

        scatter : bool, default=True
            If True, will scatter the data to the dask workers. If False, will not scatter the data. This can be useful for debugging.

        random_state : int, None, default=None
            A seed for reproducability of experiments. This value will be passed to numpy.random.default_rng() to create an instnce of the genrator to pass to other classes

            - int
                Will be used to create and lock in Generator instance with 'numpy.random.default_rng()'
            - None
                Will be used to create Generator for 'numpy.random.default_rng()' where a fresh, unpredictable entropy will be pulled from the OS

        Attributes
        ----------

        fitted_pipeline_ : GraphPipeline
            A fitted instance of the GraphPipeline that inherits from sklearn BaseEstimator. This is fitted on the full X, y passed to fit.

        evaluated_individuals : A pandas data frame containing data for all evaluated individuals in the run.
            Columns:
            - *objective functions : The first few columns correspond to the passed in scorers and objective functions
            - Parents : A tuple containing the indexes of the pipelines used to generate the pipeline of that row. If NaN, this pipeline was generated randomly in the initial population.
            - Variation_Function : Which variation function was used to mutate or crossover the parents. If NaN, this pipeline was generated randomly in the initial population.
            - Individual : The internal representation of the individual that is used during the evolutionary algorithm. This is not an sklearn BaseEstimator.
            - Generation : The generation the pipeline first appeared.
            - Pareto_Front	: The nondominated front that this pipeline belongs to. 0 means that its scores is not strictly dominated by any other individual.
                            To save on computational time, the best frontier is updated iteratively each generation.
                            The pipelines with the 0th pareto front do represent the exact best frontier. However, the pipelines with pareto front >= 1 are only in reference to the other pipelines in the final population.
                            All other pipelines are set to NaN.
            - Instance	: The unfitted GraphPipeline BaseEstimator.
            - *validation objective functions : Objective function scores evaluated on the validation set.
            - Validation_Pareto_Front : The full pareto front calculated on the validation set. This is calculated for all pipelines with Pareto_Front equal to 0. Unlike the Pareto_Front which only calculates the frontier and the final population, the Validation Pareto Front is calculated for all pipelines tested on the validation set.

        pareto_front : The same pandas dataframe as evaluated individuals, but containing only the frontier pareto front pipelines.
        '''

        # sklearn BaseEstimator must have a corresponding attribute for each parameter.
        # These should not be modified once set.

        self.scorers = scorers
        self.scorers_weights = scorers_weights
        self.classification = classification
        self.cv = cv
        self.other_objective_functions = other_objective_functions
        self.other_objective_functions_weights = other_objective_functions_weights
        self.objective_function_names = objective_function_names
        self.bigger_is_better = bigger_is_better

        self.search_space = search_space

        self.export_graphpipeline = export_graphpipeline
        self.memory = memory

        self.categorical_features = categorical_features

        self.preprocessing = preprocessing
        self.validation_strategy = validation_strategy
        self.validation_fraction = validation_fraction
        self.disable_label_encoder = disable_label_encoder
        self.population_size = population_size
        self.initial_population_size = initial_population_size
        self.population_scaling = population_scaling
        self.generations_until_end_population = generations_until_end_population
        self.generations = generations
        self.early_stop = early_stop
        self.scorers_early_stop_tol = scorers_early_stop_tol
        self.other_objectives_early_stop_tol = other_objectives_early_stop_tol
        self.max_time_mins = max_time_mins
        self.max_eval_time_mins = max_eval_time_mins
        self.n_jobs= n_jobs
        self.memory_limit = memory_limit
        self.client = client
        self.survival_percentage = survival_percentage
        self.crossover_probability = crossover_probability
        self.mutate_probability = mutate_probability
        self.mutate_then_crossover_probability= mutate_then_crossover_probability
        self.crossover_then_mutate_probability= crossover_then_mutate_probability
        self.survival_selector=survival_selector
        self.parent_selector=parent_selector
        self.budget_range = budget_range
        self.budget_scaling = budget_scaling
        self.generations_until_end_budget = generations_until_end_budget
        self.stepwise_steps = stepwise_steps
        self.threshold_evaluation_pruning =threshold_evaluation_pruning
        self.threshold_evaluation_scaling =  threshold_evaluation_scaling
        self.min_history_threshold = min_history_threshold
        self.selection_evaluation_pruning = selection_evaluation_pruning
        self.selection_evaluation_scaling =  selection_evaluation_scaling
        self.warm_start = warm_start
        self.verbose = verbose
        self.periodic_checkpoint_folder = periodic_checkpoint_folder
        self.callback = callback
        self.processes = processes


        self.scatter = scatter


        timer_set = self.max_time_mins != float("inf") and self.max_time_mins is not None
        if self.generations is not None and timer_set:
            warnings.warn("Both generations and max_time_mins are set. TPOT will terminate when the first condition is met.")

        # create random number generator based on rngseed
        self.rng = np.random.default_rng(random_state)
        # save random state passed to us for other functions that use random_state
        self.random_state = random_state

        #Initialize other used params


        if self.initial_population_size is None:
            self._initial_population_size = self.population_size
        else:
            self._initial_population_size = self.initial_population_size

        if isinstance(self.scorers, str):
            self._scorers = [self.scorers]

        elif callable(self.scorers):
            self._scorers = [self.scorers]
        else:
            self._scorers = self.scorers

        self._scorers = [sklearn.metrics.get_scorer(scoring) for scoring in self._scorers]
        self._scorers_early_stop_tol = self.scorers_early_stop_tol

        self._evolver = tpot.evolvers.BaseEvolver

        self.objective_function_weights = [*scorers_weights, *other_objective_functions_weights]


        if self.objective_function_names is None:
            obj_names = [f.__name__ for f in other_objective_functions]
        else:
            obj_names = self.objective_function_names
        self.objective_names = [f._score_func.__name__ if hasattr(f,"_score_func") else f.__name__ for f in self._scorers] + obj_names


        if not isinstance(self.other_objectives_early_stop_tol, list):
            self._other_objectives_early_stop_tol = [self.other_objectives_early_stop_tol for _ in range(len(self.other_objective_functions))]
        else:
            self._other_objectives_early_stop_tol = self.other_objectives_early_stop_tol

        if not isinstance(self._scorers_early_stop_tol, list):
            self._scorers_early_stop_tol = [self._scorers_early_stop_tol for _ in range(len(self._scorers))]
        else:
            self._scorers_early_stop_tol = self._scorers_early_stop_tol

        self.early_stop_tol = [*self._scorers_early_stop_tol, *self._other_objectives_early_stop_tol]

        self._evolver_instance = None
        self.evaluated_individuals = None


        self.label_encoder_ = None


        set_dask_settings()


    def fit(self, X, y):
        if self.client is not None: #If user passed in a client manually
           _client = self.client
        else:

            if self.verbose >= 4:
                silence_logs = 30
            elif self.verbose >=5:
                silence_logs = 40
            else:
                silence_logs = 50
            cluster = LocalCluster(n_workers=self.n_jobs, #if no client is passed in and no global client exists, create our own
                    threads_per_worker=1,
                    processes=self.processes,
                    silence_logs=silence_logs,
                    memory_limit=self.memory_limit)
            _client = Client(cluster)

        if self.classification and not self.disable_label_encoder and not check_if_y_is_encoded(y):
            warnings.warn("Labels are not encoded as ints from 0 to N. For compatibility with some classifiers such as sklearn, TPOT has encoded y with the sklearn LabelEncoder. When using pipelines outside the main TPOT estimator class, you can encode the labels with est.label_encoder_")
            self.label_encoder_ = LabelEncoder()
            y = self.label_encoder_.fit_transform(y)

        self.evaluated_individuals = None
        #determine validation strategy
        if self.validation_strategy == 'auto':
            nrows = X.shape[0]
            ncols = X.shape[1]

            if nrows/ncols < 20:
                validation_strategy = 'reshuffled'
            elif nrows/ncols < 100:
                validation_strategy = 'split'
            else:
                validation_strategy = 'none'
        else:
            validation_strategy = self.validation_strategy

        if validation_strategy == 'split':
            if self.classification:
                X, X_val, y, y_val = train_test_split(X, y, test_size=self.validation_fraction, stratify=y, random_state=self.random_state)
            else:
                X, X_val, y, y_val = train_test_split(X, y, test_size=self.validation_fraction, random_state=self.random_state)


        X_original = X
        y_original = y
        if isinstance(self.cv, int) or isinstance(self.cv, float):
            n_folds = self.cv
        else:
            n_folds = self.cv.get_n_splits(X, y)

        if self.classification:
            X, y = remove_underrepresented_classes(X, y, n_folds)

        if self.preprocessing:
            #X = pd.DataFrame(X)

            if not isinstance(self.preprocessing, bool) and isinstance(self.preprocessing, sklearn.base.BaseEstimator):
                self._preprocessing_pipeline = sklearn.base.clone(self.preprocessing)

            #TODO: check if there are missing values in X before imputation. If not, don't include imputation in pipeline. Check if there are categorical columns. If not, don't include one hot encoding in pipeline
            else: #if self.preprocessing is True or not a sklearn estimator

                pipeline_steps = []

                if self.categorical_features is not None: #if categorical features are specified, use those
                    pipeline_steps.append(("impute_categorical", tpot.builtin_modules.ColumnSimpleImputer(self.categorical_features, strategy='most_frequent')))
                    pipeline_steps.append(("impute_numeric", tpot.builtin_modules.ColumnSimpleImputer("numeric", strategy='mean')))
                    pipeline_steps.append(("ColumnOneHotEncoder", tpot.builtin_modules.ColumnOneHotEncoder(self.categorical_features, min_frequency=0.0001))) # retain wrong param fix

                else:
                    if isinstance(X, pd.DataFrame):
                        categorical_columns = X.select_dtypes(include=['object']).columns
                        if len(categorical_columns) > 0:
                            pipeline_steps.append(("impute_categorical", tpot.builtin_modules.ColumnSimpleImputer("categorical", strategy='most_frequent')))
                            pipeline_steps.append(("impute_numeric", tpot.builtin_modules.ColumnSimpleImputer("numeric", strategy='mean')))
                            pipeline_steps.append(("ColumnOneHotEncoder", tpot.builtin_modules.ColumnOneHotEncoder("categorical", min_frequency=0.0001))) # retain wrong param fix
                        else:
                            pipeline_steps.append(("impute_numeric", tpot.builtin_modules.ColumnSimpleImputer("all", strategy='mean')))
                    else:
                        pipeline_steps.append(("impute_numeric", tpot.builtin_modules.ColumnSimpleImputer("all", strategy='mean')))

                self._preprocessing_pipeline = sklearn.pipeline.Pipeline(pipeline_steps)

            X = self._preprocessing_pipeline.fit_transform(X, y)

        else:
            self._preprocessing_pipeline = None

        #_, y = sklearn.utils.check_X_y(X, y, y_numeric=True)

        #Set up the configuation dictionaries and the search spaces

        #check if self.cv is a number
        if isinstance(self.cv, int) or isinstance(self.cv, float):
            if self.classification:
                self.cv_gen = sklearn.model_selection.StratifiedKFold(n_splits=self.cv, shuffle=True, random_state=self.random_state)
            else:
                self.cv_gen = sklearn.model_selection.KFold(n_splits=self.cv, shuffle=True, random_state=self.random_state)

        else:
            self.cv_gen = sklearn.model_selection.check_cv(self.cv, y, classifier=self.classification)



        n_samples= int(math.floor(X.shape[0]/n_folds))
        n_features=X.shape[1]

        if isinstance(X, pd.DataFrame):
            self.feature_names = X.columns
        else:
            self.feature_names = None



        def objective_function(pipeline_individual,
                                            X,
                                            y,
                                            is_classification=self.classification,
                                            scorers= self._scorers,
                                            cv=self.cv_gen,
                                            other_objective_functions=self.other_objective_functions,
                                            export_graphpipeline=self.export_graphpipeline,
                                            memory=self.memory,
                                            **kwargs):
            return objective_function_generator(
                pipeline_individual,
                X,
                y,
                is_classification=is_classification,
                scorers= scorers,
                cv=cv,
                other_objective_functions=other_objective_functions,
                export_graphpipeline=export_graphpipeline,
                memory=memory,
                **kwargs,
            )



        if self.threshold_evaluation_pruning is not None or self.selection_evaluation_pruning is not None:
            evaluation_early_stop_steps = self.cv
        else:
            evaluation_early_stop_steps = None

        if self.scatter:
            X_future = _client.scatter(X)
            y_future = _client.scatter(y)
        else:
            X_future = X
            y_future = y

        if self.classification:
            n_classes = len(np.unique(y))
        else:
            n_classes = None

        get_search_space_params = {"n_classes": n_classes, 
                        "n_samples":len(y), 
                        "n_features":X.shape[1], 
                        "random_state":self.random_state}

        self._search_space = get_template_search_spaces(self.search_space, classification=self.classification, inner_predictors=True, **get_search_space_params)


        # TODO : Add check for empty values in X and if so, add imputation to the search space
        # make this depend on self.preprocessing
        # if check_empty_values(X):
        #     from sklearn.experimental import enable_iterative_imputer

        #     from ConfigSpace import ConfigurationSpace
        #     from ConfigSpace import ConfigurationSpace, Integer, Float, Categorical, Normal
        #     iterative_imputer_cs = ConfigurationSpace(
        #         space = {
        #             'n_nearest_features' : Categorical('n_nearest_features', [100]),
        #             'initial_strategy' : Categorical('initial_strategy', ['mean','median', 'most_frequent', ]),
        #             'add_indicator' : Categorical('add_indicator', [True, False]),
        #         }
        #     )

        #     imputation_search = tpot.search_spaces.pipelines.ChoicePipeline([
        #         tpot.config.get_search_space("SimpleImputer"),
        #         tpot.search_spaces.nodes.EstimatorNode(sklearn.impute.IterativeImputer, iterative_imputer_cs)
        #     ])




        #     self.search_space_final = tpot.search_spaces.pipelines.SequentialPipeline(search_spaces=[ imputation_search, self._search_space], memory="sklearn_pipeline_memory")
        # else:
        #     self.search_space_final = self._search_space

        self.search_space_final = self._search_space

        def ind_generator(rng):
            rng = np.random.default_rng(rng)
            while True:
                yield self.search_space_final.generate(rng)

        #If warm start and we have an evolver instance, use the existing one
        if not(self.warm_start and self._evolver_instance is not None):
            self._evolver_instance = self._evolver(   individual_generator=ind_generator(self.rng),
                                            objective_functions= [objective_function],
                                            objective_function_weights = self.objective_function_weights,
                                            objective_names=self.objective_names,
                                            bigger_is_better = self.bigger_is_better,
                                            population_size= self.population_size,
                                            generations=self.generations,
                                            initial_population_size = self._initial_population_size,
                                            n_jobs=self.n_jobs,
                                            verbose = self.verbose,
                                            max_time_mins =      self.max_time_mins ,
                                            max_eval_time_mins = self.max_eval_time_mins,

                                            periodic_checkpoint_folder = self.periodic_checkpoint_folder,
                                            threshold_evaluation_pruning = self.threshold_evaluation_pruning,
                                            threshold_evaluation_scaling =  self.threshold_evaluation_scaling,
                                            min_history_threshold = self.min_history_threshold,

                                            selection_evaluation_pruning = self.selection_evaluation_pruning,
                                            selection_evaluation_scaling =  self.selection_evaluation_scaling,
                                            evaluation_early_stop_steps = evaluation_early_stop_steps,

                                            early_stop_tol = self.early_stop_tol,
                                            early_stop= self.early_stop,

                                            budget_range = self.budget_range,
                                            budget_scaling = self.budget_scaling,
                                            generations_until_end_budget = self.generations_until_end_budget,

                                            population_scaling = self.population_scaling,
                                            generations_until_end_population = self.generations_until_end_population,
                                            stepwise_steps = self.stepwise_steps,
                                            client = _client,
                                            objective_kwargs = {"X": X_future, "y": y_future},
                                            survival_selector=self.survival_selector,
                                            parent_selector=self.parent_selector,
                                            survival_percentage = self.survival_percentage,
                                            crossover_probability = self.crossover_probability,
                                            mutate_probability = self.mutate_probability,
                                            mutate_then_crossover_probability= self.mutate_then_crossover_probability,
                                            crossover_then_mutate_probability= self.crossover_then_mutate_probability,

                                            rng=self.rng,
                                            )


        self._evolver_instance.optimize()
        #self._evolver_instance.population.update_pareto_fronts(self.objective_names, self.objective_function_weights)
        self.make_evaluated_individuals()




        tpot.utils.get_pareto_frontier(self.evaluated_individuals, column_names=self.objective_names, weights=self.objective_function_weights)

        if validation_strategy == 'reshuffled':
            best_pareto_front_idx = list(self.pareto_front.index)
            best_pareto_front = list(self.pareto_front.loc[best_pareto_front_idx]['Individual'])

            #reshuffle rows
            X, y = sklearn.utils.shuffle(X, y, random_state=self.random_state)

            if self.scatter:
                X_future = _client.scatter(X)
                y_future = _client.scatter(y)
            else:
                X_future = X
                y_future = y

            val_objective_function_list = [lambda   ind,
                                                    X,
                                                    y,
                                                    is_classification=self.classification,
                                                    scorers= self._scorers,
                                                    cv=self.cv_gen,
                                                    other_objective_functions=self.other_objective_functions,
                                                    export_graphpipeline=self.export_graphpipeline,
                                                    memory=self.memory,
                                                    **kwargs: objective_function_generator(
                                                                                                ind,
                                                                                                X,
                                                                                                y,
                                                                                                is_classification=is_classification,
                                                                                                scorers= scorers,
                                                                                                cv=cv,
                                                                                                other_objective_functions=other_objective_functions,
                                                                                                export_graphpipeline=export_graphpipeline,
                                                                                                memory=memory,
                                                                                                **kwargs,
                                                                                                )]

            objective_kwargs = {"X": X_future, "y": y_future}
            val_scores, start_times, end_times, eval_errors = tpot.utils.eval_utils.parallel_eval_objective_list(best_pareto_front, val_objective_function_list, verbose=self.verbose, max_eval_time_mins=self.max_eval_time_mins, n_expected_columns=len(self.objective_names), client=_client, **objective_kwargs)



            val_objective_names = ['validation_'+name for name in self.objective_names]
            self.objective_names_for_selection = val_objective_names
            self.evaluated_individuals.loc[best_pareto_front_idx,val_objective_names] = val_scores
            self.evaluated_individuals.loc[best_pareto_front_idx,'validation_start_times'] = start_times
            self.evaluated_individuals.loc[best_pareto_front_idx,'validation_end_times'] = end_times
            self.evaluated_individuals.loc[best_pareto_front_idx,'validation_eval_errors'] = eval_errors

            self.evaluated_individuals["Validation_Pareto_Front"] = tpot.utils.get_pareto_frontier(self.evaluated_individuals, column_names=val_objective_names, weights=self.objective_function_weights)


        elif validation_strategy == 'split':


            if self.scatter:
                X_future = _client.scatter(X)
                y_future = _client.scatter(y)
                X_val_future = _client.scatter(X_val)
                y_val_future = _client.scatter(y_val)
            else:
                X_future = X
                y_future = y
                X_val_future = X_val
                y_val_future = y_val

            objective_kwargs = {"X": X_future, "y": y_future, "X_val" : X_val_future, "y_val":y_val_future }

            best_pareto_front_idx = list(self.pareto_front.index)
            best_pareto_front = list(self.pareto_front.loc[best_pareto_front_idx]['Individual'])
            val_objective_function_list = [lambda   ind,
                                                    X,
                                                    y,
                                                    X_val,
                                                    y_val,
                                                    scorers= self._scorers,
                                                    other_objective_functions=self.other_objective_functions,
                                                    export_graphpipeline=self.export_graphpipeline,
                                                    memory=self.memory,
                                                    **kwargs: val_objective_function_generator(
                                                        ind,
                                                        X,
                                                        y,
                                                        X_val,
                                                        y_val,
                                                        scorers= scorers,
                                                        other_objective_functions=other_objective_functions,
                                                        export_graphpipeline=export_graphpipeline,
                                                        memory=memory,
                                                        **kwargs,
                                                        )]

            val_scores, start_times, end_times, eval_errors = tpot.utils.eval_utils.parallel_eval_objective_list(best_pareto_front, val_objective_function_list, verbose=self.verbose, max_eval_time_mins=self.max_eval_time_mins, n_expected_columns=len(self.objective_names), client=_client, **objective_kwargs)



            val_objective_names = ['validation_'+name for name in self.objective_names]
            self.objective_names_for_selection = val_objective_names
            self.evaluated_individuals.loc[best_pareto_front_idx,val_objective_names] = val_scores
            self.evaluated_individuals.loc[best_pareto_front_idx,'validation_start_times'] = start_times
            self.evaluated_individuals.loc[best_pareto_front_idx,'validation_end_times'] = end_times
            self.evaluated_individuals.loc[best_pareto_front_idx,'validation_eval_errors'] = eval_errors

            self.evaluated_individuals["Validation_Pareto_Front"] = tpot.utils.get_pareto_frontier(self.evaluated_individuals, column_names=val_objective_names, weights=self.objective_function_weights)

        else:
            self.objective_names_for_selection = self.objective_names

        val_scores = self.evaluated_individuals[self.evaluated_individuals[self.objective_names_for_selection].isna().all(1).ne(True)][self.objective_names_for_selection]
        weighted_scores = val_scores*self.objective_function_weights

        if self.bigger_is_better:
            best_indices = list(weighted_scores.sort_values(by=self.objective_names_for_selection, ascending=False).index)
        else:
            best_indices = list(weighted_scores.sort_values(by=self.objective_names_for_selection, ascending=True).index)

        for best_idx in best_indices:

            best_individual = self.evaluated_individuals.loc[best_idx]['Individual']
            self.selected_best_score =  self.evaluated_individuals.loc[best_idx]


            #TODO
            #best_individual_pipeline = best_individual.export_pipeline(memory=self.memory, cross_val_predict_cv=self.cross_val_predict_cv)
            if self.export_graphpipeline:
                best_individual_pipeline = best_individual.export_flattened_graphpipeline(memory=self.memory)
            else:
                best_individual_pipeline = best_individual.export_pipeline(memory=self.memory)

            if self.preprocessing:
                self.fitted_pipeline_ = sklearn.pipeline.make_pipeline(sklearn.base.clone(self._preprocessing_pipeline), best_individual_pipeline )
            else:
                self.fitted_pipeline_ = best_individual_pipeline

            try:
                self.fitted_pipeline_.fit(X_original,y_original) #TODO use y_original as well?
                break
            except Exception as e:
                if self.verbose >= 4:
                    warnings.warn("Final pipeline failed to fit. Rarely, the pipeline might work on the objective function but fail on the full dataset. Generally due to interactions with different features being selected or transformations having different properties. Trying next pipeline")
                    print(e)
                continue


        if self.client is None: #no client was passed in
            #close cluster and client
            # _client.close()
            # cluster.close()
            try:
                _client.shutdown()
                cluster.close()
            #catch exception
            except Exception as e:
                print("Error shutting down client and cluster")
                Warning(e)

        return self

    def _estimator_has(attr):
        '''Check if we can delegate a method to the underlying estimator.
        First, we check the first fitted final estimator if available, otherwise we
        check the unfitted final estimator.
        '''
        return  lambda self: (self.fitted_pipeline_ is not None and
            hasattr(self.fitted_pipeline_, attr)
        )






    @available_if(_estimator_has('predict'))
    def predict(self, X, **predict_params):
        check_is_fitted(self)
        #X = check_array(X)

        preds = self.fitted_pipeline_.predict(X,**predict_params)
        if self.classification and self.label_encoder_:
            preds = self.label_encoder_.inverse_transform(preds)

        return preds

    @available_if(_estimator_has('predict_proba'))
    def predict_proba(self, X, **predict_params):
        check_is_fitted(self)
        #X = check_array(X)
        return self.fitted_pipeline_.predict_proba(X,**predict_params)

    @available_if(_estimator_has('decision_function'))
    def decision_function(self, X, **predict_params):
        check_is_fitted(self)
        #X = check_array(X)
        return self.fitted_pipeline_.decision_function(X,**predict_params)

    @available_if(_estimator_has('transform'))
    def transform(self, X, **predict_params):
        check_is_fitted(self)
        #X = check_array(X)
        return self.fitted_pipeline_.transform(X,**predict_params)

    @property
    def classes_(self):
        """The classes labels. Only exist if the last step is a classifier."""
        if self.label_encoder_:
            return self.label_encoder_.classes_
        else:
            return self.fitted_pipeline_.classes_


    @property
    def _estimator_type(self):
        return self.fitted_pipeline_._estimator_type


    def make_evaluated_individuals(self):
        #check if _evolver_instance exists
        if self.evaluated_individuals is None:
            self.evaluated_individuals  =  self._evolver_instance.population.evaluated_individuals.copy()
            objects = list(self.evaluated_individuals.index)
            object_to_int = dict(zip(objects, range(len(objects))))
            self.evaluated_individuals = self.evaluated_individuals.set_index(self.evaluated_individuals.index.map(object_to_int))
            self.evaluated_individuals['Parents'] = self.evaluated_individuals['Parents'].apply(lambda row: convert_parents_tuples_to_integers(row, object_to_int))

            self.evaluated_individuals["Instance"] = self.evaluated_individuals["Individual"].apply(lambda ind: apply_make_pipeline(ind, preprocessing_pipeline=self._preprocessing_pipeline, export_graphpipeline=self.export_graphpipeline, memory=self.memory))

        return self.evaluated_individuals

    @property
    def pareto_front(self):
        #check if _evolver_instance exists
        if self.evaluated_individuals is None:
            return None
        else:
            if "Pareto_Front" not in self.evaluated_individuals:
                return self.evaluated_individuals
            else:
                return self.evaluated_individuals[self.evaluated_individuals["Pareto_Front"]==1]

classes_ 属性

类别标签。仅当最后一步是分类器时存在。

__init__(search_space, scorers, scorers_weights, classification, cv=10, other_objective_functions=[], other_objective_functions_weights=[], objective_function_names=None, bigger_is_better=True, export_graphpipeline=False, memory=None, categorical_features=None, preprocessing=False, population_size=50, initial_population_size=None, population_scaling=0.5, generations_until_end_population=1, generations=None, max_time_mins=60, max_eval_time_mins=10, validation_strategy='none', validation_fraction=0.2, disable_label_encoder=False, early_stop=None, scorers_early_stop_tol=0.001, other_objectives_early_stop_tol=None, threshold_evaluation_pruning=None, threshold_evaluation_scaling=0.5, selection_evaluation_pruning=None, selection_evaluation_scaling=0.5, min_history_threshold=20, survival_percentage=1, crossover_probability=0.2, mutate_probability=0.7, mutate_then_crossover_probability=0.05, crossover_then_mutate_probability=0.05, survival_selector=survival_select_NSGA2, parent_selector=tournament_selection_dominated, budget_range=None, budget_scaling=0.5, generations_until_end_budget=1, stepwise_steps=5, n_jobs=1, memory_limit=None, client=None, processes=True, warm_start=False, periodic_checkpoint_folder=None, callback=None, verbose=0, scatter=True, random_state=None)

一个使用遗传编程优化流水线的 sklearn 基本估计器。

参数

名称	类型	描述	默认值
search_space	(String, SearchSpace)	字符串：用于优化的默认搜索空间。| 字符串 | 描述 | | :--- | :----: | | linear | 一个线性流水线，结构为“选择器->(转换器+直通)->(分类器/回归器+直通)->最终分类器/回归器”。对于转换器层和内部估计器层，TPOT 可以选择一个或多个转换器/分类器，也可以不选择任何一个。内部分类器/回归器层是可选的。| | linear-light | 与 linear 搜索空间相同，但没有内部分类器/回归器层，并使用一组更快的估计器。| | graph | TPOT 将优化一个有向无环图形状的流水线。图的节点可以包括选择器、缩放器、转换器或分类器/回归器（内部分类器/回归器可选不包含）。这将返回一个自定义的 GraphPipeline 而不是 sklearn Pipeline。更多详细信息请参见教程 6。| | graph-light | 与 graph 搜索空间相同，但没有内部分类器/回归器，并使用一组更快的估计器。| | mdr | TPOT 将搜索一系列特征选择器和多因素维度约减模型（Multifactor Dimensionality Reduction models）以找到最大化预测准确性的操作序列。TPOT MDR 配置专门用于全基因组关联研究 (GWAS)，并在此处在线详细描述。 请注意，TPOT MDR 运行可能较慢，因为特征选择例程计算成本高昂，尤其是在大型数据集上。| SearchSpace：用于优化的搜索空间。这应该是 SearchSpace 的一个实例。用于优化的搜索空间。这应该是 SearchSpace 的一个实例。TPOT 在以下文件夹中分组了搜索空间：tpot.search_spaces.nodes 用于流水线中的节点，tpot.search_spaces.pipelines 用于流水线结构。	必需
scorers	(list, scorer)	交叉验证过程中使用的评分器或评分器列表。参见 https://scikit-learn.cn/stable/modules/model_evaluation.html	必需
scorers_weights	列表	在优化过程中应用于评分器的权重列表。	必需
classification	布尔值	如果为 True，则问题被视为分类问题。如果为 False，则问题被视为回归问题。用于确定 CV 策略。	必需
cv	(int, cross - validator)	(int)：交叉验证过程中使用的折叠数。回归使用 sklearn.model_selection.KFold 交叉验证器，分类使用 StratifiedKFold。在这两种情况下，shuffled 都设置为 True。 (sklearn.model_selection.BaseCrossValidator)：交叉验证过程中使用的交叉验证器。 max_depth (int)：从任何节点到要生成的流水线根部的最大深度。	10
other_objective_functions	列表	应用于流水线的其他目标函数列表。函数接受 graphpipeline 估计器的单个参数，并返回单个分数或分数列表。	[]
other_objective_functions_weights	列表	应用于其他目标函数的权重列表。	[]
objective_function_names	列表	应用于目标函数的名称列表。如果为 None，将使用目标函数的名称。	无
bigger_is_better	布尔值	如果为 True，则最大化目标函数。如果为 False，则最小化目标函数。使用负权重来反转方向。	True
memory		如果提供，流水线在调用 fit 时会使用 joblib.Memory 缓存每个转换器。此功能用于避免在优化过程中，当参数和输入数据与另一个已拟合的流水线相同时，重新计算流水线内的已拟合转换器。- 字符串 'auto'：TPOT 使用临时目录进行内存缓存，并在关闭时清除它。- 字符串缓存目录路径：TPOT 使用提供的目录进行内存缓存，TPOT 在关闭时不会清除缓存目录。如果目录不存在，TPOT 将创建它。- Memory 对象：TPOT 使用 joblib.Memory 的实例进行内存缓存，TPOT 在关闭时不会清除缓存目录。- None：TPOT 不使用内存缓存。	无
categorical_features		预处理步骤中用于填充和/或独热编码的分类列。仅当 preprocessing 不为 False 时使用。- None：如果为 None，TPOT 将自动使用 pandas dataframes 中的对象列作为预处理中独热编码的对象。- 分类特征列表。如果 X 是 dataframe，这应该是一个列名列表。如果 X 是 numpy 数组，这应该是一个列索引列表。	无
preprocessing	(bool 或 BaseEstimator / Pipeline)	实验性 - 未来版本可能会更改 在 CV 之前用于预处理数据的流水线。请注意，这些步骤的参数不会被优化。将它们添加到搜索空间中进行优化。- bool：如果为 True，将使用默认预处理流水线，包括填充和独热编码。- Pipeline：如果给出了流水线实例，将使用该流水线作为预处理流水线。	False
population_size	整数	种群大小	50
initial_population_size	整数	初始种群大小。如果为 None，将使用 population_size。	无
population_scaling	整数	确定阈值从起始百分位移到结束百分位的速度时使用的缩放因子。	0.5
generations_until_end_population	整数	种群大小达到 population_size 的代数。	1
generations	整数	运行的代数	50
max_time_mins	浮点数	运行优化的最大时间（分钟）。如果为 None 或 inf，将运行直到达到代数上限。	float("inf")
max_eval_time_mins	浮点数	评估单个个体的最大时间（分钟）。如果为 None 或 inf，则每次评估没有时间限制。	5
validation_strategy	字符串	实验性 用于从种群中选择最终流水线的验证策略。TPOT 可能会过拟合交叉验证分数。可以使用第二个验证集来选择最终流水线。- 'auto'：根据数据集形状自动确定验证策略。- 'reshuffled'：对交叉验证和最终验证使用相同的数据，但折叠的划分不同。这是小型数据集的默认值。- 'split'：使用单独的验证集进行最终验证。数据将根据 validation_fraction 进行划分。这是中型数据集的默认值。- 'none'：不使用单独的验证集进行最终验证。根据原始交叉验证分数选择。这是大型数据集的默认值。	'none'
validation_fraction	浮点数	实验性 当 validation_strategy 为 'split' 时，用于验证集的数据集比例。必须介于 0 和 1 之间。	0.2
disable_label_encoder	布尔值	如果为 True，TPOT 将检查目标是否需要重新标记为从 0 到 N 的连续整数。这对于 XGBoost 兼容性是必需的。如果标签需要编码，TPOT 将使用 sklearn.preprocessing.LabelEncoder 对标签进行编码。编码器可以通过 self.label_encoder_ 属性访问。如果为 False，将不使用额外的标签编码器。	False
early_stop	整数	在提前停止之前没有改进的代数。所有目标必须在容差范围内收敛才能触发此机制。通常，值在 5-20 之间比较好。	无
scorers_early_stop_tol		-浮点数列表：每个评分器的容差列表。如果最佳分数与当前分数之间的差小于容差，则认为该个体已收敛 如果列表中的某个索引为 None，则该项不会用于提前停止 -整数：如果给定整数，将作为所有目标的容差。	0.001
other_objectives_early_stop_tol		-浮点数列表：每个其他目标函数的容差列表。如果最佳分数与当前分数之间的差小于容差，则认为该个体已收敛 如果列表中的某个索引为 None，则该项不会用于提前停止 -整数：如果给定整数，将作为所有目标的容差。	无
threshold_evaluation_pruning	list[start, end]	用于评估提前停止阈值的起始和结束百分位。值介于 0 和 100 之间。	无
threshold_evaluation_scaling	浮点数 [0,inf)	确定阈值从起始百分位移到结束百分位的速度时使用的缩放因子。必须大于零。数值越大，阈值移到结束百分位的速度越快。	0.5
selection_evaluation_pruning	列表	每轮 CV 选择的种群大小的较低和较高百分比。值介于 0 和 1 之间。	无
selection_evaluation_scaling	浮点数	确定阈值从起始百分位移到结束百分位的速度时使用的缩放因子。必须大于零。数值越大，阈值移到结束百分位的速度越快。	0.5
min_history_threshold	整数	在使用阈值提前停止之前所需的最小历史分数数量。	0
survival_percentage	浮点数	在世代开始时用于变异和交叉的种群大小百分比。其余的被丢弃。个体使用传递到 survival_selector 的选择器进行选择。此参数的值必须介于 0 和 1 之间（包含）。例如，如果种群大小为 100，生存百分比为 0.5，将从现有种群中通过 NSGA2 选择 50 个体。这些个体将用于变异和交叉，以生成下一代的 100 个个体。其余的从当前种群中丢弃。在下一代，将有 50 个父代个体 + 100 个个体，总计 150 个。生存百分比基于 population size 参数，而不是现有种群大小（使用逐次减半时的当前种群大小）。因此，在下一代，我们仍然将从现有的 150 个体中选择 50 个。	1
crossover_probability	浮点数	通过两个个体交叉生成新个体的概率。	.2
mutate_probability	浮点数	通过一个个体变异生成新个体的概率。	.7
mutate_then_crossover_probability	浮点数	对两个个体进行变异后再进行交叉生成新个体的概率。	.05
crossover_then_mutate_probability	浮点数	通过两个个体交叉生成新个体，然后对结果个体进行变异的概率。	.05
survival_selector	函数	用于选择生存个体的函数。必须接受一个分数矩阵并返回选定的索引。用于在每个世代开始时选择 population_size * survival_percentage 个体进行变异和交叉。	survival_select_NSGA2
parent_selector	函数	用于选择进行交叉的父代对和进行变异的个体的函数。必须接受一个分数矩阵并返回选定的索引。	parent_select_NSGA2
budget_range	list[start, end]	用于预算缩放的起始和结束预算。	无
budget_scaling		确定预算从起始预算移到结束预算的速度时使用的缩放因子。	0.5
generations_until_end_budget	整数	在达到最大预算之前运行的代数。	1
stepwise_steps	整数	缩放预算和种群大小时采取的阶梯步数。	1
n_jobs	整数	并行运行的进程数。	1
memory_limit	字符串	每个作业的内存限制。有关更多信息，请参阅 Dask LocalCluster 文档。	无
client	客户端	用于并行化的 dask 客户端。如果不是 None，这将覆盖 n_jobs 和 memory_limit 参数。如果为 None，将创建一个新的客户端，num_workers=n_jobs，memory_limit=memory_limit。	无
processes	布尔值	如果为 True，将使用多进程并行化优化过程。如果为 False，将使用多线程。True 似乎性能更好。然而，交互式调试需要 False。	True
warm_start	布尔值	如果为 True，将从上次运行的最后一代继续演化算法。	False
periodic_checkpoint_folder	字符串	定期保存种群的文件夹。如果为 None，将不进行定期保存。如果提供，训练将从该检查点恢复。	无
callback	CallBackInterface	回调对象。未实现	无
verbose	整数	优化过程中打印的信息量。值越高，包含的信息越多。0. 无 1. 进度条 最佳个体 警告 =5. 完整的警告跟踪 评估进度条。(临时：这以前是 2。目前，使用评估进度条可能会阻止某些情况下我们提前终止一代（因为它在世代中间达到 max_time_mins）或者流水线未能正常终止而我们需要手动终止它。)	1
scatter	布尔值	如果为 True，将数据分散到 dask workers。如果为 False，将不分散数据。这对于调试很有用。	True
random_state	(int, None)	实验重现性的种子。此值将传递给 numpy.random.default_rng() 以创建生成器实例，然后传递给其他类 整数：将用于通过 'numpy.random.default_rng()' 创建和锁定 Generator 实例 None：将用于为 'numpy.random.default_rng()' 创建 Generator，其中将从操作系统获取全新的、不可预测的熵	无

属性

名称	类型	描述
fitted_pipeline_	GraphPipeline	GraphPipeline 的已拟合实例，继承自 sklearn BaseEstimator。这是在传递给 fit 的完整 X, y 上拟合的。
evaluated_individuals	一个 pandas 数据框，包含运行中所有已评估个体的数据。	列：- 目标函数：前几列对应于传入的评分器和目标函数 - Parents：一个元组，包含用于生成该行流水线的流水线索引。如果为 NaN，则此流水线是在初始种群中随机生成的。- Variation_Function：用于变异或交叉父代的变异函数。如果为 NaN，则此流水线是在初始种群中随机生成的。- Individual：在演化算法中使用的个体内部表示。这不是一个 sklearn BaseEstimator。- Generation：流水线首次出现的世代。- Pareto_Front：此流水线所属的非支配前沿。0 表示其分数不被任何其他个体严格支配。为了节省计算时间，最佳前沿在每个世代迭代更新。帕累托前沿为 0 的流水线确实代表了精确的最佳前沿。然而，帕累托前沿 >= 1 的流水线仅参考最终种群中的其他流水线。所有其他流水线设置为 NaN。- Instance：未拟合的 GraphPipeline BaseEstimator。- 验证目标函数：在验证集上评估的目标函数分数。- Validation_Pareto_Front：在验证集上计算的完整帕累托前沿。这是针对所有 Pareto_Front 等于 0 的流水线计算的。与仅计算前沿和最终种群的 Pareto_Front 不同，Validation Pareto Front 是针对在验证集上测试的所有流水线计算的。
pareto_front	与 evaluated_individuals 相同的 pandas 数据框，但仅包含前沿帕累托前沿流水线。

源代码位于 tpot/tpot_estimator/estimator.py

def __init__(self,  
                    search_space,
                    scorers,
                    scorers_weights,
                    classification,
                    cv = 10,
                    other_objective_functions=[],
                    other_objective_functions_weights = [],
                    objective_function_names = None,
                    bigger_is_better = True,

                    export_graphpipeline = False,
                    memory = None,

                    categorical_features = None,
                    preprocessing = False,
                    population_size = 50,
                    initial_population_size = None,
                    population_scaling = .5,
                    generations_until_end_population = 1,
                    generations = None,
                    max_time_mins=60,
                    max_eval_time_mins=10,
                    validation_strategy = "none",
                    validation_fraction = .2,
                    disable_label_encoder = False,

                    #early stopping parameters
                    early_stop = None,
                    scorers_early_stop_tol = 0.001,
                    other_objectives_early_stop_tol =None,
                    threshold_evaluation_pruning = None,
                    threshold_evaluation_scaling = .5,
                    selection_evaluation_pruning = None,
                    selection_evaluation_scaling = .5,
                    min_history_threshold = 20,

                    #evolver parameters
                    survival_percentage = 1,
                    crossover_probability=.2,
                    mutate_probability=.7,
                    mutate_then_crossover_probability=.05,
                    crossover_then_mutate_probability=.05,
                    survival_selector = survival_select_NSGA2,
                    parent_selector = tournament_selection_dominated,

                    #budget parameters
                    budget_range = None,
                    budget_scaling = .5,
                    generations_until_end_budget = 1,
                    stepwise_steps = 5,

                    #dask parameters
                    n_jobs=1,
                    memory_limit = None,
                    client = None,
                    processes = True,

                    #debugging and logging parameters
                    warm_start = False,
                    periodic_checkpoint_folder = None,
                    callback = None,

                    verbose = 0,
                    scatter = True,

                     # random seed for random number generator (rng)
                    random_state = None,

                    ):

    '''
    An sklearn baseestimator that uses genetic programming to optimize a pipeline.

    Parameters
    ----------
    search_space : (String, tpot.search_spaces.SearchSpace)
        - String : The default search space to use for the optimization.
        | String     | Description      |
        | :---        |    :----:   |
        | linear  | A linear pipeline with the structure of "Selector->(transformers+Passthrough)->(classifiers/regressors+Passthrough)->final classifier/regressor." For both the transformer and inner estimator layers, TPOT may choose one or more transformers/classifiers, or it may choose none. The inner classifier/regressor layer is optional. |
        | linear-light | Same search space as linear, but without the inner classifier/regressor layer and with a reduced set of faster running estimators. |
        | graph | TPOT will optimize a pipeline in the shape of a directed acyclic graph. The nodes of the graph can include selectors, scalers, transformers, or classifiers/regressors (inner classifiers/regressors can optionally be not included). This will return a custom GraphPipeline rather than an sklearn Pipeline. More details in Tutorial 6. |
        | graph-light | Same as graph search space, but without the inner classifier/regressors and with a reduced set of faster running estimators. |
        | mdr |TPOT will search over a series of feature selectors and Multifactor Dimensionality Reduction models to find a series of operators that maximize prediction accuracy. The TPOT MDR configuration is specialized for genome-wide association studies (GWAS), and is described in detail online here.

        Note that TPOT MDR may be slow to run because the feature selection routines are computationally expensive, especially on large datasets. |


        - SearchSpace : The search space to use for the optimization. This should be an instance of a SearchSpace.
            The search space to use for the optimization. This should be an instance of a SearchSpace.
            TPOT has groups of search spaces found in the following folders, tpot.search_spaces.nodes for the nodes in the pipeline and tpot.search_spaces.pipelines for the pipeline structure.

    scorers : (list, scorer)
        A scorer or list of scorers to be used in the cross-validation process.
        see https://scikit-learn.cn/stable/modules/model_evaluation.html

    scorers_weights : list
        A list of weights to be applied to the scorers during the optimization process.

    classification : bool
        If True, the problem is treated as a classification problem. If False, the problem is treated as a regression problem.
        Used to determine the CV strategy.

    cv : int, cross-validator
        - (int): Number of folds to use in the cross-validation process. By uses the sklearn.model_selection.KFold cross-validator for regression and StratifiedKFold for classification. In both cases, shuffled is set to True.
        - (sklearn.model_selection.BaseCrossValidator): A cross-validator to use in the cross-validation process.
            - max_depth (int): The maximum depth from any node to the root of the pipelines to be generated.

    other_objective_functions : list, default=[]
        A list of other objective functions to apply to the pipeline. The function takes a single parameter for the graphpipeline estimator and returns either a single score or a list of scores.

    other_objective_functions_weights : list, default=[]
        A list of weights to be applied to the other objective functions.

    objective_function_names : list, default=None
        A list of names to be applied to the objective functions. If None, will use the names of the objective functions.

    bigger_is_better : bool, default=True
        If True, the objective function is maximized. If False, the objective function is minimized. Use negative weights to reverse the direction.

    memory: Memory object or string, default=None
        If supplied, pipeline will cache each transformer after calling fit with joblib.Memory. This feature
        is used to avoid computing the fit transformers within a pipeline if the parameters
        and input data are identical with another fitted pipeline during optimization process.
        - String 'auto':
            TPOT uses memory caching with a temporary directory and cleans it up upon shutdown.
        - String path of a caching directory
            TPOT uses memory caching with the provided directory and TPOT does NOT clean
            the caching directory up upon shutdown. If the directory does not exist, TPOT will
            create it.
        - Memory object:
            TPOT uses the instance of joblib.Memory for memory caching,
            and TPOT does NOT clean the caching directory up upon shutdown.
        - None:
            TPOT does not use memory caching.              

    categorical_features: list or None
        Categorical columns to inpute and/or one hot encode during the preprocessing step. Used only if preprocessing is not False.
        - None : If None, TPOT will automatically use object columns in pandas dataframes as objects for one hot encoding in preprocessing.
        - List of categorical features. If X is a dataframe, this should be a list of column names. If X is a numpy array, this should be a list of column indices

    preprocessing : bool or BaseEstimator/Pipeline,
        EXPERIMENTAL - will be changed in future versions
        A pipeline that will be used to preprocess the data before CV. Note that the parameters for these steps are not optimized. Add them to the search space to be optimized.
        - bool : If True, will use a default preprocessing pipeline which includes imputation followed by one hot encoding.
        - Pipeline : If an instance of a pipeline is given, will use that pipeline as the preprocessing pipeline.

    population_size : int, default=50
        Size of the population

    initial_population_size : int, default=None
        Size of the initial population. If None, population_size will be used.

    population_scaling : int, default=0.5
        Scaling factor to use when determining how fast we move the threshold moves from the start to end percentile.

    generations_until_end_population : int, default=1
        Number of generations until the population size reaches population_size

    generations : int, default=50
        Number of generations to run

    max_time_mins : float, default=float("inf")
        Maximum time to run the optimization. If none or inf, will run until the end of the generations.

    max_eval_time_mins : float, default=5
        Maximum time to evaluate a single individual. If none or inf, there will be no time limit per evaluation.

    validation_strategy : str, default='none'
        EXPERIMENTAL The validation strategy to use for selecting the final pipeline from the population. TPOT may overfit the cross validation score. A second validation set can be used to select the final pipeline.
        - 'auto' : Automatically determine the validation strategy based on the dataset shape.
        - 'reshuffled' : Use the same data for cross validation and final validation, but with different splits for the folds. This is the default for small datasets.
        - 'split' : Use a separate validation set for final validation. Data will be split according to validation_fraction. This is the default for medium datasets.
        - 'none' : Do not use a separate validation set for final validation. Select based on the original cross-validation score. This is the default for large datasets.

    validation_fraction : float, default=0.2
      EXPERIMENTAL The fraction of the dataset to use for the validation set when validation_strategy is 'split'. Must be between 0 and 1.

    disable_label_encoder : bool, default=False
        If True, TPOT will check if the target needs to be relabeled to be sequential ints from 0 to N. This is necessary for XGBoost compatibility. If the labels need to be encoded, TPOT will use sklearn.preprocessing.LabelEncoder to encode the labels. The encoder can be accessed via the self.label_encoder_ attribute.
        If False, no additional label encoders will be used.

    early_stop : int, default=None
        Number of generations without improvement before early stopping. All objectives must have converged within the tolerance for this to be triggered. In general a value of around 5-20 is good.

    scorers_early_stop_tol :
        -list of floats
            list of tolerances for each scorer. If the difference between the best score and the current score is less than the tolerance, the individual is considered to have converged
            If an index of the list is None, that item will not be used for early stopping
        -int
            If an int is given, it will be used as the tolerance for all objectives

    other_objectives_early_stop_tol :
        -list of floats
            list of tolerances for each of the other objective function. If the difference between the best score and the current score is less than the tolerance, the individual is considered to have converged
            If an index of the list is None, that item will not be used for early stopping
        -int
            If an int is given, it will be used as the tolerance for all objectives

    threshold_evaluation_pruning : list [start, end], default=None
        starting and ending percentile to use as a threshold for the evaluation early stopping.
        Values between 0 and 100.

    threshold_evaluation_scaling : float [0,inf), default=0.5
        A scaling factor to use when determining how fast we move the threshold moves from the start to end percentile.
        Must be greater than zero. Higher numbers will move the threshold to the end faster.

    selection_evaluation_pruning : list, default=None
        A lower and upper percent of the population size to select each round of CV.
        Values between 0 and 1.

    selection_evaluation_scaling : float, default=0.5
        A scaling factor to use when determining how fast we move the threshold moves from the start to end percentile.
        Must be greater than zero. Higher numbers will move the threshold to the end faster.

    min_history_threshold : int, default=0
        The minimum number of previous scores needed before using threshold early stopping.

    survival_percentage : float, default=1
        Percentage of the population size to utilize for mutation and crossover at the beginning of the generation. The rest are discarded. Individuals are selected with the selector passed into survival_selector. The value of this parameter must be between 0 and 1, inclusive.
        For example, if the population size is 100 and the survival percentage is .5, 50 individuals will be selected with NSGA2 from the existing population. These will be used for mutation and crossover to generate the next 100 individuals for the next generation. The remainder are discarded from the live population. In the next generation, there will now be the 50 parents + the 100 individuals for a total of 150. Surivival percentage is based of the population size parameter and not the existing population size (current population size when using successive halving). Therefore, in the next generation we will still select 50 individuals from the currently existing 150.

    crossover_probability : float, default=.2
        Probability of generating a new individual by crossover between two individuals.

    mutate_probability : float, default=.7
        Probability of generating a new individual by crossover between one individuals.

    mutate_then_crossover_probability : float, default=.05
        Probability of generating a new individual by mutating two individuals followed by crossover.

    crossover_then_mutate_probability : float, default=.05
        Probability of generating a new individual by crossover between two individuals followed by a mutation of the resulting individual.

    survival_selector : function, default=survival_select_NSGA2
        Function to use to select individuals for survival. Must take a matrix of scores and return selected indexes.
        Used to selected population_size * survival_percentage individuals at the start of each generation to use for mutation and crossover.

    parent_selector : function, default=parent_select_NSGA2
        Function to use to select pairs parents for crossover and individuals for mutation. Must take a matrix of scores and return selected indexes.

    budget_range : list [start, end], default=None
        A starting and ending budget to use for the budget scaling.

    budget_scaling float : [0,1], default=0.5
        A scaling factor to use when determining how fast we move the budget from the start to end budget.

    generations_until_end_budget : int, default=1
        The number of generations to run before reaching the max budget.

    stepwise_steps : int, default=1
        The number of staircase steps to take when scaling the budget and population size.

    n_jobs : int, default=1
        Number of processes to run in parallel.

    memory_limit : str, default=None
        Memory limit for each job. See Dask [LocalCluster documentation](https://distributed.dask.org.cn/en/stable/api.html#distributed.Client) for more information.

    client : dask.distributed.Client, default=None
        A dask client to use for parallelization. If not None, this will override the n_jobs and memory_limit parameters. If None, will create a new client with num_workers=n_jobs and memory_limit=memory_limit.

    processes : bool, default=True
        If True, will use multiprocessing to parallelize the optimization process. If False, will use threading.
        True seems to perform better. However, False is required for interactive debugging.

    warm_start : bool, default=False
        If True, will use the continue the evolutionary algorithm from the last generation of the previous run.

    periodic_checkpoint_folder : str, default=None
        Folder to save the population to periodically. If None, no periodic saving will be done.
        If provided, training will resume from this checkpoint.

    callback : tpot.CallBackInterface, default=None
        Callback object. Not implemented

    verbose : int, default=1
        How much information to print during the optimization process. Higher values include the information from lower values.
        0. nothing
        1. progress bar

        3. best individual
        4. warnings
        >=5. full warnings trace
        6. evaluations progress bar. (Temporary: This used to be 2. Currently, using evaluation progress bar may prevent some instances were we terminate a generation early due to it reaching max_time_mins in the middle of a generation OR a pipeline failed to be terminated normally and we need to manually terminate it.)

    scatter : bool, default=True
        If True, will scatter the data to the dask workers. If False, will not scatter the data. This can be useful for debugging.

    random_state : int, None, default=None
        A seed for reproducability of experiments. This value will be passed to numpy.random.default_rng() to create an instnce of the genrator to pass to other classes

        - int
            Will be used to create and lock in Generator instance with 'numpy.random.default_rng()'
        - None
            Will be used to create Generator for 'numpy.random.default_rng()' where a fresh, unpredictable entropy will be pulled from the OS

    Attributes
    ----------

    fitted_pipeline_ : GraphPipeline
        A fitted instance of the GraphPipeline that inherits from sklearn BaseEstimator. This is fitted on the full X, y passed to fit.

    evaluated_individuals : A pandas data frame containing data for all evaluated individuals in the run.
        Columns:
        - *objective functions : The first few columns correspond to the passed in scorers and objective functions
        - Parents : A tuple containing the indexes of the pipelines used to generate the pipeline of that row. If NaN, this pipeline was generated randomly in the initial population.
        - Variation_Function : Which variation function was used to mutate or crossover the parents. If NaN, this pipeline was generated randomly in the initial population.
        - Individual : The internal representation of the individual that is used during the evolutionary algorithm. This is not an sklearn BaseEstimator.
        - Generation : The generation the pipeline first appeared.
        - Pareto_Front	: The nondominated front that this pipeline belongs to. 0 means that its scores is not strictly dominated by any other individual.
                        To save on computational time, the best frontier is updated iteratively each generation.
                        The pipelines with the 0th pareto front do represent the exact best frontier. However, the pipelines with pareto front >= 1 are only in reference to the other pipelines in the final population.
                        All other pipelines are set to NaN.
        - Instance	: The unfitted GraphPipeline BaseEstimator.
        - *validation objective functions : Objective function scores evaluated on the validation set.
        - Validation_Pareto_Front : The full pareto front calculated on the validation set. This is calculated for all pipelines with Pareto_Front equal to 0. Unlike the Pareto_Front which only calculates the frontier and the final population, the Validation Pareto Front is calculated for all pipelines tested on the validation set.

    pareto_front : The same pandas dataframe as evaluated individuals, but containing only the frontier pareto front pipelines.
    '''

    # sklearn BaseEstimator must have a corresponding attribute for each parameter.
    # These should not be modified once set.

    self.scorers = scorers
    self.scorers_weights = scorers_weights
    self.classification = classification
    self.cv = cv
    self.other_objective_functions = other_objective_functions
    self.other_objective_functions_weights = other_objective_functions_weights
    self.objective_function_names = objective_function_names
    self.bigger_is_better = bigger_is_better

    self.search_space = search_space

    self.export_graphpipeline = export_graphpipeline
    self.memory = memory

    self.categorical_features = categorical_features

    self.preprocessing = preprocessing
    self.validation_strategy = validation_strategy
    self.validation_fraction = validation_fraction
    self.disable_label_encoder = disable_label_encoder
    self.population_size = population_size
    self.initial_population_size = initial_population_size
    self.population_scaling = population_scaling
    self.generations_until_end_population = generations_until_end_population
    self.generations = generations
    self.early_stop = early_stop
    self.scorers_early_stop_tol = scorers_early_stop_tol
    self.other_objectives_early_stop_tol = other_objectives_early_stop_tol
    self.max_time_mins = max_time_mins
    self.max_eval_time_mins = max_eval_time_mins
    self.n_jobs= n_jobs
    self.memory_limit = memory_limit
    self.client = client
    self.survival_percentage = survival_percentage
    self.crossover_probability = crossover_probability
    self.mutate_probability = mutate_probability
    self.mutate_then_crossover_probability= mutate_then_crossover_probability
    self.crossover_then_mutate_probability= crossover_then_mutate_probability
    self.survival_selector=survival_selector
    self.parent_selector=parent_selector
    self.budget_range = budget_range
    self.budget_scaling = budget_scaling
    self.generations_until_end_budget = generations_until_end_budget
    self.stepwise_steps = stepwise_steps
    self.threshold_evaluation_pruning =threshold_evaluation_pruning
    self.threshold_evaluation_scaling =  threshold_evaluation_scaling
    self.min_history_threshold = min_history_threshold
    self.selection_evaluation_pruning = selection_evaluation_pruning
    self.selection_evaluation_scaling =  selection_evaluation_scaling
    self.warm_start = warm_start
    self.verbose = verbose
    self.periodic_checkpoint_folder = periodic_checkpoint_folder
    self.callback = callback
    self.processes = processes


    self.scatter = scatter


    timer_set = self.max_time_mins != float("inf") and self.max_time_mins is not None
    if self.generations is not None and timer_set:
        warnings.warn("Both generations and max_time_mins are set. TPOT will terminate when the first condition is met.")

    # create random number generator based on rngseed
    self.rng = np.random.default_rng(random_state)
    # save random state passed to us for other functions that use random_state
    self.random_state = random_state

    #Initialize other used params


    if self.initial_population_size is None:
        self._initial_population_size = self.population_size
    else:
        self._initial_population_size = self.initial_population_size

    if isinstance(self.scorers, str):
        self._scorers = [self.scorers]

    elif callable(self.scorers):
        self._scorers = [self.scorers]
    else:
        self._scorers = self.scorers

    self._scorers = [sklearn.metrics.get_scorer(scoring) for scoring in self._scorers]
    self._scorers_early_stop_tol = self.scorers_early_stop_tol

    self._evolver = tpot.evolvers.BaseEvolver

    self.objective_function_weights = [*scorers_weights, *other_objective_functions_weights]


    if self.objective_function_names is None:
        obj_names = [f.__name__ for f in other_objective_functions]
    else:
        obj_names = self.objective_function_names
    self.objective_names = [f._score_func.__name__ if hasattr(f,"_score_func") else f.__name__ for f in self._scorers] + obj_names


    if not isinstance(self.other_objectives_early_stop_tol, list):
        self._other_objectives_early_stop_tol = [self.other_objectives_early_stop_tol for _ in range(len(self.other_objective_functions))]
    else:
        self._other_objectives_early_stop_tol = self.other_objectives_early_stop_tol

    if not isinstance(self._scorers_early_stop_tol, list):
        self._scorers_early_stop_tol = [self._scorers_early_stop_tol for _ in range(len(self._scorers))]
    else:
        self._scorers_early_stop_tol = self._scorers_early_stop_tol

    self.early_stop_tol = [*self._scorers_early_stop_tol, *self._other_objectives_early_stop_tol]

    self._evolver_instance = None
    self.evaluated_individuals = None


    self.label_encoder_ = None


    set_dask_settings()

apply_make_pipeline(ind, preprocessing_pipeline=None, export_graphpipeline=False, **pipeline_kwargs)

辅助函数，用于从 tpot individual 类创建一列 sklearn 流水线。

参数

名称	类型	描述	默认值
ind		要转换为流水线的个体。	必需
preprocessing_pipeline		要包含在个体流水线之前的预处理流水线。	无
export_graphpipeline		强制将流水线导出为图流水线。将所有嵌套的流水线、FeatureUnions 和 GraphPipelines 平展到一个单一的 GraphPipeline 中。	False
pipeline_kwargs		传递给 export_pipeline 或 export_flattened_graphpipeline 方法的关键字参数。	{}

返回

类型	描述
sklearn 估计器

源代码位于 tpot/tpot_estimator/estimator_utils.py

def apply_make_pipeline(ind, preprocessing_pipeline=None, export_graphpipeline=False, **pipeline_kwargs):
    """
    Helper function to create a column of sklearn pipelines from the tpot individual class.

    Parameters
    ----------
    ind: tpot.SklearnIndividual
        The individual to convert to a pipeline.
    preprocessing_pipeline: sklearn.pipeline.Pipeline, optional
        The preprocessing pipeline to include before the individual's pipeline.
    export_graphpipeline: bool, default=False
        Force the pipeline to be exported as a graph pipeline. Flattens all nested pipelines, FeatureUnions, and GraphPipelines into a single GraphPipeline.
    pipeline_kwargs: dict
        Keyword arguments to pass to the export_pipeline or export_flattened_graphpipeline method.

    Returns
    -------
    sklearn estimator
    """

    try:

        if export_graphpipeline:
            est = ind.export_flattened_graphpipeline(**pipeline_kwargs)
        else:
            est = ind.export_pipeline(**pipeline_kwargs)


        if preprocessing_pipeline is None:
            return est
        else:
            return sklearn.pipeline.make_pipeline(sklearn.base.clone(preprocessing_pipeline), est)
    except:
        return None

check_empty_values(data)

检查数据集中的空值。

参数：data (numpy.ndarray 或 pandas.DataFrame)：要检查的数据集。

返回：bool：如果数据集包含空值，则为 True，否则为 False。

源代码位于 tpot/tpot_estimator/estimator.py

def check_empty_values(data):
    """
    Checks for empty values in a dataset.

    Args:
        data (numpy.ndarray or pandas.DataFrame): The dataset to check.

    Returns:
        bool: True if the dataset contains empty values, False otherwise.
    """
    if isinstance(data, pd.DataFrame):
        return data.isnull().values.any()
    elif isinstance(data, np.ndarray):
        return np.isnan(data).any()
    else:
        raise ValueError("Unsupported data type")

check_if_y_is_encoded(y)

检查目标 y 是否由从 0 到 N 的连续整数组成。XGBoost 要求目标以这种方式编码。

参数

名称	类型	描述	默认值
y		目标向量。	必需

返回

类型	描述
布尔值	如果目标编码为从 0 到 N 的连续整数，则为 True，否则为 False

源代码位于 tpot/tpot_estimator/estimator_utils.py

def check_if_y_is_encoded(y):
    '''
    Checks if the target y is composed of sequential ints from 0 to N.
    XGBoost requires the target to be encoded in this way.

    Parameters
    ----------
    y: np.ndarray
        The target vector.

    Returns
    -------
    bool
        True if the target is encoded as sequential ints from 0 to N, False otherwise
    '''
    y = sorted(set(y))
    return all(i == j for i, j in enumerate(y))

convert_parents_tuples_to_integers(row, object_to_int)

辅助函数，用于将父代行转换为表示父代在种群中索引的整数。

使用自定义索引作为父代的原始 pandas 数据框。此函数将自定义索引转换为整数索引，方便最终用户操作。

参数

名称	类型	描述	默认值
row		要转换的行。	必需
object_to_int		一个将对象映射到整数索引的字典。	必需

返回

元组：自定义索引已转换为整数索引的行。

源代码位于 tpot/tpot_estimator/estimator_utils.py

def convert_parents_tuples_to_integers(row, object_to_int):
    """
    Helper function to convert the parent rows into integers representing the index of the parent in the population.

    Original pandas dataframe using a custom index for the parents. This function converts the custom index to an integer index for easier manipulation by end users.

    Parameters
    ----------
    row: list, np.ndarray, tuple
        The row to convert.
    object_to_int: dict
        A dictionary mapping the object to an integer index.

    Returns 
    -------
    tuple
        The row with the custom index converted to an integer index.
    """
    if type(row) == list or type(row) == np.ndarray or type(row) == tuple:
        return tuple(object_to_int[obj] for obj in row)
    else:
        return np.nan

cross_val_score_objective(estimator, X, y, scorers, cv, fold=None)

计算估计器的交叉验证分数。每个折叠仅拟合估计器一次，并遍历评分器来评估估计器。

参数

名称	类型	描述	默认值
估计器		要拟合和评分的估计器。	必需
X		特征矩阵。	必需
y		目标向量。	必需
scorers		要使用的评分器。如果是一个列表，将遍历评分器并返回一个评分器列表。如果是一个评分器，将返回一个分数。	必需
cv		要使用的交叉验证器。例如，sklearn.model_selection.KFold 或 sklearn.model_selection.StratifiedKFold。	必需
fold		返回分数的折叠。如果为 None，将返回所有分数的平均值（每个评分器）。默认值为 None。	无

返回

名称	类型	描述
scores	ndarray 或 float	每个评分器对估计器的分数。如果 fold 为 None，将返回所有分数的平均值（每个评分器）。如果使用多个评分器，则返回列表，否则返回单个评分器的浮点数。

源代码位于 tpot/tpot_estimator/cross_val_utils.py

def cross_val_score_objective(estimator, X, y, scorers, cv, fold=None):
    """
    Compute the cross validated scores for a estimator. Only fits the estimator once per fold, and loops over the scorers to evaluate the estimator.

    Parameters
    ----------
    estimator: sklearn.base.BaseEstimator
        The estimator to fit and score.
    X: np.ndarray or pd.DataFrame
        The feature matrix.
    y: np.ndarray or pd.Series
        The target vector.
    scorers: list or scorer
        The scorers to use. 
        If a list, will loop over the scorers and return a list of scorers.
        If a single scorer, will return a single score.
    cv: sklearn cross-validator
        The cross-validator to use. For example, sklearn.model_selection.KFold or sklearn.model_selection.StratifiedKFold.
    fold: int, optional
        The fold to return the scores for. If None, will return the mean of all the scores (per scorer). Default is None.

    Returns
    -------
    scores: np.ndarray or float
        The scores for the estimator per scorer. If fold is None, will return the mean of all the scores (per scorer).
        Returns a list if multiple scorers are used, otherwise returns a float for the single scorer.

    """

    #check if scores is not iterable
    if not isinstance(scorers, Iterable): 
        scorers = [scorers]
    scores = []
    if fold is None:
        for train_index, test_index in cv.split(X, y):
            this_fold_estimator = sklearn.base.clone(estimator)
            if isinstance(X, pd.DataFrame) or isinstance(X, pd.Series):
                X_train, X_test = X.iloc[train_index], X.iloc[test_index]
            else:
                X_train, X_test = X[train_index], X[test_index]

            if isinstance(y, pd.DataFrame) or isinstance(y, pd.Series):
                y_train, y_test = y.iloc[train_index], y.iloc[test_index]
            else:
                y_train, y_test = y[train_index], y[test_index]


            start = time.time()
            this_fold_estimator.fit(X_train,y_train)
            duration = time.time() - start

            this_fold_scores = [sklearn.metrics.get_scorer(scorer)(this_fold_estimator, X_test, y_test) for scorer in scorers] 
            scores.append(this_fold_scores)
            del this_fold_estimator
            del X_train
            del X_test
            del y_train
            del y_test


        return np.mean(scores,0)
    else:
        this_fold_estimator = sklearn.base.clone(estimator)
        train_index, test_index = list(cv.split(X, y))[fold]
        if isinstance(X, pd.DataFrame) or isinstance(X, pd.Series):
            X_train, X_test = X.iloc[train_index], X.iloc[test_index]
        else:
            X_train, X_test = X[train_index], X[test_index]

        if isinstance(y, pd.DataFrame) or isinstance(y, pd.Series):
            y_train, y_test = y.iloc[train_index], y.iloc[test_index]
        else:
            y_train, y_test = y[train_index], y[test_index]

        start = time.time()
        this_fold_estimator.fit(X_train,y_train)
        duration = time.time() - start
        this_fold_scores = [sklearn.metrics.get_scorer(scorer)(this_fold_estimator, X_test, y_test) for scorer in scorers] 
        return this_fold_scores

objective_function_generator(pipeline, x, y, scorers, cv, other_objective_functions, step=None, budget=None, is_classification=True, export_graphpipeline=False, **pipeline_kwargs)

使用交叉验证评估流水线，并将其结果与独立的其他目标函数分数连接起来。

参数

名称	类型	描述	默认值
pipeline		要评估的个体。	必需
x		特征矩阵。	必需
y		目标向量。	必需
scorers		交叉验证中使用的评分器。	必需
cv		要使用的交叉验证器。例如，sklearn.model_selection.KFold 或 sklearn.model_selection.StratifiedKFold。如果是一个整数，将使用 sklearn.model_selection.KFold，其中 n_splits=cv。	必需
other_objective_functions		一个独立的目标函数列表，用于评估流水线。签名应为 obj(pipeline) -> float 或 obj(pipeline) -> np.ndarray。这些函数接受未拟合的估计器。	必需
step		返回分数的折叠。如果为 None，将返回所有分数的平均值（每个评分器）。默认值为 None。	无
budget		用于对数据进行二次抽样的预算。如果为 None，将使用完整数据集。默认值为 None。将对 budget*len(x) 个样本进行二次抽样。	无
is_classification		如果为 True，将对二次抽样进行分层。默认值为 True。	True
export_graphpipeline		强制将流水线导出为图流水线。将所有嵌套的 sklearn 流水线、FeatureUnions 和 GraphPipelines 平展到一个单一的 GraphPipeline 中。	False
pipeline_kwargs		传递给 export_pipeline 或 export_flattened_graphpipeline 方法的关键字参数。	{}

返回

类型	描述
ndarray	流水线的连接分数。前 len(scorers) 个元素是交叉验证分数，其余元素是独立的客观函数。

源代码位于 tpot/tpot_estimator/estimator_utils.py

def objective_function_generator(pipeline, x,y, scorers, cv, other_objective_functions, step=None, budget=None, is_classification=True, export_graphpipeline=False, **pipeline_kwargs):
    """
    Uses cross validation to evaluate the pipeline using the scorers, and concatenates results with scores from standalone other objective functions.

    Parameters
    ----------
    pipeline: tpot.SklearnIndividual
        The individual to evaluate.
    x: np.ndarray
        The feature matrix.
    y: np.ndarray
        The target vector.
    scorers: list
        The scorers to use for cross validation. 
    cv: int, float, or sklearn cross-validator
        The cross-validator to use. For example, sklearn.model_selection.KFold or sklearn.model_selection.StratifiedKFold.
        If an int, will use sklearn.model_selection.KFold with n_splits=cv.
    other_objective_functions: list
        A list of standalone objective functions to evaluate the pipeline. With signature obj(pipeline) -> float. or obj(pipeline) -> np.ndarray
        These functions take in the unfitted estimator.
    step: int, optional
        The fold to return the scores for. If None, will return the mean of all the scores (per scorer). Default is None.
    budget: float, optional
        The budget to subsample the data. If None, will use the full dataset. Default is None.
        Will subsample budget*len(x) samples.
    is_classification: bool, default=True
        If True, will stratify the subsampling. Default is True.
    export_graphpipeline: bool, default=False
        Force the pipeline to be exported as a graph pipeline. Flattens all nested sklearn pipelines, FeatureUnions, and GraphPipelines into a single GraphPipeline.
    pipeline_kwargs: dict
        Keyword arguments to pass to the export_pipeline or export_flattened_graphpipeline method.

    Returns
    -------
    np.ndarray
        The concatenated scores for the pipeline. The first len(scorers) elements are the cross validation scores, and the remaining elements are the standalone objective functions.

    """

    if export_graphpipeline:
        pipeline = pipeline.export_flattened_graphpipeline(**pipeline_kwargs)
    else:
        pipeline = pipeline.export_pipeline(**pipeline_kwargs)

    if budget is not None and budget < 1:
        if is_classification:
            x,y = sklearn.utils.resample(x,y, stratify=y, n_samples=int(budget*len(x)), replace=False, random_state=1)
        else:
            x,y = sklearn.utils.resample(x,y, n_samples=int(budget*len(x)), replace=False, random_state=1)

        if isinstance(cv, int) or isinstance(cv, float):
            n_splits = cv
        else:
            n_splits = cv.n_splits

    if len(scorers) > 0:
        cv_obj_scores = cross_val_score_objective(sklearn.base.clone(pipeline),x,y,scorers=scorers, cv=cv , fold=step)
    else:
        cv_obj_scores = []

    if other_objective_functions is not None and len(other_objective_functions) >0:
        other_scores = [obj(sklearn.base.clone(pipeline)) for obj in other_objective_functions]
        #flatten
        other_scores = np.array(other_scores).flatten().tolist()
    else:
        other_scores = []

    return np.concatenate([cv_obj_scores,other_scores])

remove_underrepresented_classes(x, y, min_count)

辅助函数，用于从数据集中移除样本数少于 min_count 的类别。

参数

名称	类型	描述	默认值
x		特征矩阵。	必需
y		目标向量。	必需
min_count		保留类别的最小样本数。	必需

返回

类型	描述
(ndarray, ndarray)	已移除样本数少于 min_count 的类别的特征矩阵和目标向量。

源代码位于 tpot/tpot_estimator/estimator_utils.py

def remove_underrepresented_classes(x, y, min_count):
    """
    Helper function to remove classes with less than min_count samples from the dataset.

    Parameters
    ----------
    x: np.ndarray or pd.DataFrame
        The feature matrix.
    y: np.ndarray or pd.Series
        The target vector.
    min_count: int
        The minimum number of samples to keep a class.

    Returns
    -------
    np.ndarray, np.ndarray
        The feature matrix and target vector with rows from classes with less than min_count samples removed.
    """
    if isinstance(y, (np.ndarray, pd.Series)):
        unique, counts = np.unique(y, return_counts=True)
        if min(counts) >= min_count:
            return x, y
        keep_classes = unique[counts >= min_count]
        mask = np.isin(y, keep_classes)
        x = x[mask]
        y = y[mask]
    elif isinstance(y, pd.DataFrame):
        counts = y.apply(pd.Series.value_counts)
        if min(counts) >= min_count:
            return x, y
        keep_classes = counts.index[counts >= min_count].tolist()
        mask = y.isin(keep_classes).all(axis=1)
        x = x[mask]
        y = y[mask]
    else:
        raise TypeError("y must be a numpy array or a pandas Series/DataFrame")
    return x, y

val_objective_function_generator(pipeline, X_train, y_train, X_test, y_test, scorers, other_objective_functions, export_graphpipeline=False, **pipeline_kwargs)

在训练集上训练流水线，并使用评分器和其他目标函数在测试集上评估它。

参数

名称	类型	描述	默认值
pipeline		要评估的个体。	必需
X_train		训练集的特征矩阵。	必需
y_train		训练集的目标向量。	必需
X_test		测试集的特征矩阵。	必需
y_test		测试集的目标向量。	必需
scorers		交叉验证中使用的评分器。	必需
other_objective_functions		一个独立的目标函数列表，用于评估流水线。签名应为 obj(pipeline) -> float 或 obj(pipeline) -> np.ndarray。这些函数接受未拟合的估计器。	必需
export_graphpipeline		强制将流水线导出为图流水线。将所有嵌套的 sklearn 流水线、FeatureUnions 和 GraphPipelines 平展到一个单一的 GraphPipeline 中。	False
pipeline_kwargs		传递给 export_pipeline 或 export_flattened_graphpipeline 方法的关键字参数。	{}

返回

类型	描述
ndarray	流水线的连接分数。前 len(scorers) 个元素是交叉验证分数，其余元素是独立的客观函数。

源代码位于 tpot/tpot_estimator/estimator_utils.py

def val_objective_function_generator(pipeline, X_train, y_train, X_test, y_test, scorers, other_objective_functions, export_graphpipeline=False, **pipeline_kwargs):
    """
    Trains a pipeline on a training set and evaluates it on a test set using the scorers and other objective functions.

    Parameters
    ----------

    pipeline: tpot.SklearnIndividual
        The individual to evaluate.
    X_train: np.ndarray
        The feature matrix of the training set.
    y_train: np.ndarray
        The target vector of the training set.
    X_test: np.ndarray
        The feature matrix of the test set.
    y_test: np.ndarray
        The target vector of the test set.
    scorers: list
        The scorers to use for cross validation.
    other_objective_functions: list
        A list of standalone objective functions to evaluate the pipeline. With signature obj(pipeline) -> float. or obj(pipeline) -> np.ndarray
        These functions take in the unfitted estimator.
    export_graphpipeline: bool, default=False
        Force the pipeline to be exported as a graph pipeline. Flattens all nested sklearn pipelines, FeatureUnions, and GraphPipelines into a single GraphPipeline.
    pipeline_kwargs: dict
        Keyword arguments to pass to the export_pipeline or export_flattened_graphpipeline method.

    Returns
    -------
    np.ndarray
        The concatenated scores for the pipeline. The first len(scorers) elements are the cross validation scores, and the remaining elements are the standalone objective functions.


    """

    #subsample the data
    if export_graphpipeline:
        pipeline = pipeline.export_flattened_graphpipeline(**pipeline_kwargs)
    else:
        pipeline = pipeline.export_pipeline(**pipeline_kwargs)

    fitted_pipeline = sklearn.base.clone(pipeline)
    fitted_pipeline.fit(X_train, y_train)

    if len(scorers) > 0:
        scores =[sklearn.metrics.get_scorer(scorer)(fitted_pipeline, X_test, y_test) for scorer in scorers]

    other_scores = []
    if other_objective_functions is not None and len(other_objective_functions) >0:
        other_scores = [obj(sklearn.base.clone(pipeline)) for obj in other_objective_functions]

    return np.concatenate([scores,other_scores])