ML之RF&XGBoost:基于RF/XGBoost(均+5f-CrVa)算法对Titanic(泰坦尼克号)数据集进行二分类预测(乘客是否生还)

输出结

ML之RF&XGBoost:基于RF/XGBoost(均+5f-CrVa)算法对Titanic(泰坦尼克号)数据集进行二分类预测(乘客是否生还)

ML之RF&XGBoost:基于RF/XGBoost(均+5f-CrVa)算法对Titanic(泰坦尼克号)数据集进行二分类预测(乘客是否生还)

 

设计思

ML之RF&XGBoost:基于RF/XGBoost(均+5f-CrVa)算法对Titanic(泰坦尼克号)数据集进行二分类预测(乘客是否生还)

 

核心代

    rfc = RandomForestClassifier()          

   rfc_cross_val_score=cross_val_score(rfc, X_train, y_train, cv=5).mean()

   print('RF:',rfc_cross_val_score)      

   rfc.fit(X_train,y_train)

   rfc_y_predict = rfc.predict(X_test)

   rfc_submission = pd.DataFrame({'PassengerId': test['PassengerId'], 'Survived': rfc_y_predict})

   rfc_submission.to_csv('data_input/Titanic Data/Titanic_rfc_submission.csv', index=False)

   

   xgbc = XGBClassifier()                  

   xgbc_cross_val_score=cross_val_score(xgbc, X_train, y_train, cv=5).mean()

   print('XGBoost:',xgbc_cross_val_score)  

   xgbc.fit(X_train, y_train)

   xgbc_y_predict = xgbc.predict(X_test)

   xgbc_submission = pd.DataFrame({'PassengerId': test['PassengerId'], 'Survived': xgbc_y_predict})

   xgbc_submission.to_csv('data_input/Titanic Data/Titanic_xgbc_submission.csv', index=False)


class RandomForestClassifier(ForestClassifier):

   """A random forest classifier.

   

   A random forest is a meta estimator that fits a number of decision tree

   classifiers on various sub-samples of the dataset and use averaging to

   improve the predictive accuracy and control over-fitting.

   The sub-sample size is always the same as the original

   input sample size but the samples are drawn with replacement if

   `bootstrap=True` (default).

   

   Read more in the :ref:`User Guide <forest>`.

   

   Parameters

   ----------

   n_estimators : integer, optional (default=10)

   The number of trees in the forest.

   

   criterion : string, optional (default="gini")

   The function to measure the quality of a split. Supported criteria are

   "gini" for the Gini impurity and "entropy" for the information gain.

   Note: this parameter is tree-specific.

   

   max_features : int, float, string or None, optional (default="auto")

   The number of features to consider when looking for the best split:

   

   - If int, then consider `max_features` features at each split.

   - If float, then `max_features` is a percentage and

   `int(max_features * n_features)` features are considered at each

   split.

   - If "auto", then `max_features=sqrt(n_features)`.

   - If "sqrt", then `max_features=sqrt(n_features)` (same as "auto").

   - If "log2", then `max_features=log2(n_features)`.

   - If None, then `max_features=n_features`.

   

   Note: the search for a split does not stop until at least one

   valid partition of the node samples is found, even if it requires to

   effectively inspect more than ``max_features`` features.

   

   max_depth : integer or None, optional (default=None)

   The maximum depth of the tree. If None, then nodes are expanded

    until

   all leaves are pure or until all leaves contain less than

   min_samples_split samples.

   

   min_samples_split : int, float, optional (default=2)

   The minimum number of samples required to split an internal node:

   

   - If int, then consider `min_samples_split` as the minimum number.

   - If float, then `min_samples_split` is a percentage and

   `ceil(min_samples_split * n_samples)` are the minimum

   number of samples for each split.

   

   .. versionchanged:: 0.18

   Added float values for percentages.

   

   min_samples_leaf : int, float, optional (default=1)

   The minimum number of samples required to be at a leaf node:

   

   - If int, then consider `min_samples_leaf` as the minimum number.

   - If float, then `min_samples_leaf` is a percentage and

   `ceil(min_samples_leaf * n_samples)` are the minimum

   number of samples for each node.

   

   .. versionchanged:: 0.18

   Added float values for percentages.

   

   min_weight_fraction_leaf : float, optional (default=0.)

   The minimum weighted fraction of the sum total of weights (of all

   the input samples) required to be at a leaf node. Samples have

   equal weight when sample_weight is not provided.

   

   max_leaf_nodes : int or None, optional (default=None)

   Grow trees with ``max_leaf_nodes`` in best-first fashion.

   Best nodes are defined as relative reduction in impurity.

   If None then unlimited number of leaf nodes.

   

   min_impurity_split : float,

   Threshold for early stopping in tree growth. A node will split

   if its impurity is above the threshold, otherwise it is a leaf.

   

   .. deprecated:: 0.19

   ``min_impurity_split`` has been deprecated in favor of

   ``min_impurity_decrease`` in 0.19 and will be removed in 0.21.

   Use ``min_impurity_decrease`` instead.

   

   min_impurity_decrease : float, optional (default=0.)

   A node will be split if this split induces a decrease of the impurity

   greater than or equal to this value.

   

   The weighted impurity decrease equation is the following::

   

   N_t / N * (impurity - N_t_R / N_t * right_impurity

   - N_t_L / N_t * left_impurity)

   

   where ``N`` is the total number of samples, ``N_t`` is the number of

   samples at the current node, ``N_t_L`` is the number of samples in the

   left child, and ``N_t_R`` is the number of samples in the right child.

   

   ``N``, ``N_t``, ``N_t_R`` and ``N_t_L`` all refer to the weighted sum,

   if ``sample_weight`` is passed.

   

   .. versionadded:: 0.19

   

   bootstrap : boolean, optional (default=True)

   Whether bootstrap samples are used when building trees.

   

   oob_score : bool (default=False)

   Whether to use out-of-bag samples to estimate

   the generalization accuracy.

   

   n_jobs : integer, optional (default=1)

   The number of jobs to run in parallel for both `fit` and `predict`.

   If -1, then the number of jobs is set to the number of cores.

   

   random_state : int, RandomState instance or None, optional

    (default=None)

   If int, random_state is the seed used by the random number

    generator;

   If RandomState instance, random_state is the random number

    generator;

   If None, the random number generator is the RandomState instance

    used

   by `np.random`.

   

   verbose : int, optional (default=0)

   Controls the verbosity of the tree building process.

   

   warm_start : bool, optional (default=False)

   When set to ``True``, reuse the solution of the previous call to fit

   and add more estimators to the ensemble, otherwise, just fit a whole

   new forest.

   

   class_weight : dict, list of dicts, "balanced",

   "balanced_subsample" or None, optional (default=None)

   Weights associated with classes in the form ``{class_label: weight}``.

   If not given, all classes are supposed to have weight one. For

   multi-output problems, a list of dicts can be provided in the same

   order as the columns of y.

   

   Note that for multioutput (including multilabel) weights should be

   defined for each class of every column in its own dict. For example,

   for four-class multilabel classification weights should be

   [{0: 1, 1: 1}, {0: 1, 1: 5}, {0: 1, 1: 1}, {0: 1, 1: 1}] instead of

   [{1:1}, {2:5}, {3:1}, {4:1}].

   

   The "balanced" mode uses the values of y to automatically adjust

   weights inversely proportional to class frequencies in the input data

   as ``n_samples / (n_classes * np.bincount(y))``

   

   The "balanced_subsample" mode is the same as "balanced" except

    that

   weights are computed based on the bootstrap sample for every tree

   grown.

   

   For multi-output, the weights of each column of y will be multiplied.

   

   Note that these weights will be multiplied with sample_weight

    (passed

   through the fit method) if sample_weight is specified.

   

   Attributes

   ----------

   estimators_ : list of DecisionTreeClassifier

   The collection of fitted sub-estimators.

   

   classes_ : array of shape = [n_classes] or a list of such arrays

   The classes labels (single output problem), or a list of arrays of

   class labels (multi-output problem).

   

   n_classes_ : int or list

   The number of classes (single output problem), or a list containing the

   number of classes for each output (multi-output problem).

   

   n_features_ : int

   The number of features when ``fit`` is performed.

   

   n_outputs_ : int

   The number of outputs when ``fit`` is performed.

   

   feature_importances_ : array of shape = [n_features]

   The feature importances (the higher, the more important the feature).

   

   oob_score_ : float

   Score of the training dataset obtained using an out-of-bag estimate.

   

   oob_decision_function_ : array of shape = [n_samples, n_classes]

   Decision function computed with out-of-bag estimate on the training

   set. If n_estimators is small it might be possible that a data point

   was never left out during the bootstrap. In this case,

   `oob_decision_function_` might contain NaN.

   

   Examples

   --------

   >>> from sklearn.ensemble import RandomForestClassifier

   >>> from sklearn.datasets import make_classification

   >>>

   >>> X, y = make_classification(n_samples=1000, n_features=4,

   ...                            n_informative=2, n_redundant=0,

   ...                            random_state=0, shuffle=False)

   >>> clf = RandomForestClassifier(max_depth=2, random_state=0)

   >>> clf.fit(X, y)

   RandomForestClassifier(bootstrap=True, class_weight=None,

    criterion='gini',

   max_depth=2, max_features='auto', max_leaf_nodes=None,

   min_impurity_decrease=0.0, min_impurity_split=None,

   min_samples_leaf=1, min_samples_split=2,

   min_weight_fraction_leaf=0.0, n_estimators=10, n_jobs=1,

   oob_score=False, random_state=0, verbose=0, warm_start=False)

   >>> print(clf.feature_importances_)

   [ 0.17287856  0.80608704  0.01884792  0.00218648]

   >>> print(clf.predict([[0, 0, 0, 0]]))

   [1]

   

   Notes

   -----

   The default values for the parameters controlling the size of the trees

   (e.g. ``max_depth``, ``min_samples_leaf``, etc.) lead to fully grown and

   unpruned trees which can potentially be very large on some data

    sets. To

   reduce memory consumption, the complexity and size of the trees

    should be

   controlled by setting those parameter values.

   

   The features are always randomly permuted at each split. Therefore,

   the best found split may vary, even with the same training data,

   ``max_features=n_features`` and ``bootstrap=False``, if the

    improvement

   of the criterion is identical for several splits enumerated during the

   search of the best split. To obtain a deterministic behaviour during

   fitting, ``random_state`` has to be fixed.

   

   References

   ----------

   

   .. [1] L. Breiman, "Random Forests", Machine Learning, 45(1), 5-32,

    2001.

   

   See also

   --------

   DecisionTreeClassifier, ExtraTreesClassifier

   """

   def __init__(self,

       n_estimators=10,

       criterion="gini",

       max_depth=None,

       min_samples_split=2,

       min_samples_leaf=1,

       min_weight_fraction_leaf=0.,

       max_features="auto",

       max_leaf_nodes=None,

       min_impurity_decrease=0.,

       min_impurity_split=None,

       bootstrap=True,

       oob_score=False,

       n_jobs=1,

       random_state=None,

       verbose=0,

       warm_start=False,

       class_weight=None):

       super(RandomForestClassifier, self).__init__

        (base_estimator=DecisionTreeClassifier(), n_estimators=n_estimators,

        estimator_params=("criterion", "max_depth", "min_samples_split",

        "min_samples_leaf", "min_weight_fraction_leaf", "max_features",

        "max_leaf_nodes", "min_impurity_decrease", "min_impurity_split",

        "random_state"), bootstrap=bootstrap, oob_score=oob_score,

        n_jobs=n_jobs, random_state=random_state, verbose=verbose,

        warm_start=warm_start, class_weight=class_weight)

       self.criterion = criterion

       self.max_depth = max_depth

       self.min_samples_split = min_samples_split

       self.min_samples_leaf = min_samples_leaf

       self.min_weight_fraction_leaf = min_weight_fraction_leaf

       self.max_features = max_features

       self.max_leaf_nodes = max_leaf_nodes

       self.min_impurity_decrease = min_impurity_decrease

       self.min_impurity_split = min_impurity_split


上一篇:《算法基础》——第3章 链表 3.1 基本概念


下一篇:云数据库和云存储使用