• 萌新向Python数据分析及数据挖掘 第三章 机器学习常用算法 第三节 梯度下降法 (下)实操篇


    In [1]:
     
     
     
     
     
    from sklearn import datasets
     
     
    In [2]:
     
     
     
     
     
    boston  = datasets.load_boston()
    X = boston.data 
    y = boston.target
    #去除不真实的数据 
    X = X[y < 50]
    y = y[y < 50]
     
     
    In [3]:
     
     
     
     
     
    from sklearn.model_selection import train_test_split #载入数据切分工具
     
     
    In [5]:
     
     
     
     
     
    X_train, X_test, y_train, y_test = train_test_split(X,y,test_size = 0.2,random_state=666)
    #切分数据
     
     
    In [6]:
     
     
     
     
     
    from sklearn.preprocessing import StandardScaler#数据归一化
     
     
    In [9]:
     
     
     
     
     
    standardScaler = StandardScaler()
    standardScaler.fit(X_train)
     
     
    Out[9]:
    StandardScaler(copy=True, with_mean=True, with_std=True)
    In [10]:
     
     
     
     
     
    X_train_standard = standardScaler.transform(X_train)
    X_test_standard = standardScaler.transform(X_test)
     
     
    In [16]:
     
     
     
     
     
    from sklearn.linear_model import SGDRegressor #导入包
    sgd_reg = SGDRegressor()#实例化
     
     
     
     
     
     
     
     
    Init signature: SGDRegressor(loss='squared_loss', penalty='l2', alpha=0.0001, l1_ratio=0.15, fit_intercept=True, max_iter=None, tol=None, shuffle=True, verbose=0, epsilon=0.1, random_state=None, learning_rate='invscaling', eta0=0.01, power_t=0.25, warm_start=False, average=False, n_iter=None)
    Docstring:    
    Linear model fitted by minimizing a regularized empirical loss with SGD
    SGD stands for Stochastic Gradient Descent: the gradient of the loss is
    estimated each sample at a time and the model is updated along the way with
    a decreasing strength schedule (aka learning rate).
    The regularizer is a penalty added to the loss function that shrinks model
    parameters towards the zero vector using either the squared euclidean norm
    L2 or the absolute norm L1 or a combination of both (Elastic Net). If the
    parameter update crosses the 0.0 value because of the regularizer, the
    update is truncated to 0.0 to allow for learning sparse models and achieve
    online feature selection.
    This implementation works with data represented as dense numpy arrays of
    floating point values for the features.
    Read more in the :ref:`User Guide <sgd>`.
    Parameters
    ----------
    loss : str, default: 'squared_loss'
        The loss function to be used. The possible values are 'squared_loss',
        'huber', 'epsilon_insensitive', or 'squared_epsilon_insensitive'
        The 'squared_loss' refers to the ordinary least squares fit.
        'huber' modifies 'squared_loss' to focus less on getting outliers
        correct by switching from squared to linear loss past a distance of
        epsilon. 'epsilon_insensitive' ignores errors less than epsilon and is
        linear past that; this is the loss function used in SVR.
        'squared_epsilon_insensitive' is the same but becomes squared loss past
        a tolerance of epsilon.
    penalty : str, 'none', 'l2', 'l1', or 'elasticnet'
        The penalty (aka regularization term) to be used. Defaults to 'l2'
        which is the standard regularizer for linear SVM models. 'l1' and
        'elasticnet' might bring sparsity to the model (feature selection)
        not achievable with 'l2'.
    alpha : float
        Constant that multiplies the regularization term. Defaults to 0.0001
        Also used to compute learning_rate when set to 'optimal'.
    l1_ratio : float
        The Elastic Net mixing parameter, with 0 <= l1_ratio <= 1.
        l1_ratio=0 corresponds to L2 penalty, l1_ratio=1 to L1.
        Defaults to 0.15.
    fit_intercept : bool
        Whether the intercept should be estimated or not. If False, the
        data is assumed to be already centered. Defaults to True.
    max_iter : int, optional
        The maximum number of passes over the training data (aka epochs).
        It only impacts the behavior in the ``fit`` method, and not the
        `partial_fit`.
        Defaults to 5. Defaults to 1000 from 0.21, or if tol is not None.
        .. versionadded:: 0.19
    tol : float or None, optional
        The stopping criterion. If it is not None, the iterations will stop
        when (loss > previous_loss - tol). Defaults to None.
        Defaults to 1e-3 from 0.21.
        .. versionadded:: 0.19
    shuffle : bool, optional
        Whether or not the training data should be shuffled after each epoch.
        Defaults to True.
    verbose : integer, optional
        The verbosity level.
    epsilon : float
        Epsilon in the epsilon-insensitive loss functions; only if `loss` is
        'huber', 'epsilon_insensitive', or 'squared_epsilon_insensitive'.
        For 'huber', determines the threshold at which it becomes less
        important to get the prediction exactly right.
        For epsilon-insensitive, any differences between the current prediction
        and the correct label are ignored if they are less than this threshold.
    random_state : int, RandomState instance or None, optional (default=None)
        The seed of the pseudo random number generator to use when shuffling
        the data.  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`.
    learning_rate : string, optional
        The learning rate schedule:
        - 'constant': eta = eta0
        - 'optimal': eta = 1.0 / (alpha * (t + t0)) [default]
        - 'invscaling': eta = eta0 / pow(t, power_t)
        where t0 is chosen by a heuristic proposed by Leon Bottou.
    eta0 : double, optional
        The initial learning rate [default 0.01].
    power_t : double, optional
        The exponent for inverse scaling learning rate [default 0.25].
    warm_start : bool, optional
        When set to True, reuse the solution of the previous call to fit as
        initialization, otherwise, just erase the previous solution.
    average : bool or int, optional
        When set to True, computes the averaged SGD weights and stores the
        result in the ``coef_`` attribute. If set to an int greater than 1,
        averaging will begin once the total number of samples seen reaches
        average. So ``average=10`` will begin averaging after seeing 10
        samples.
    n_iter : int, optional
        The number of passes over the training data (aka epochs).
        Defaults to None. Deprecated, will be removed in 0.21.
        .. versionchanged:: 0.19
            Deprecated
    Attributes
    ----------
    coef_ : array, shape (n_features,)
        Weights assigned to the features.
    intercept_ : array, shape (1,)
        The intercept term.
    average_coef_ : array, shape (n_features,)
        Averaged weights assigned to the features.
    average_intercept_ : array, shape (1,)
        The averaged intercept term.
    n_iter_ : int
        The actual number of iterations to reach the stopping criterion.
    Examples
    --------
    >>> import numpy as np
    >>> from sklearn import linear_model
    >>> n_samples, n_features = 10, 5
    >>> np.random.seed(0)
    >>> y = np.random.randn(n_samples)
    >>> X = np.random.randn(n_samples, n_features)
    >>> clf = linear_model.SGDRegressor()
    >>> clf.fit(X, y)
    ... #doctest: +NORMALIZE_WHITESPACE
    SGDRegressor(alpha=0.0001, average=False, epsilon=0.1, eta0=0.01,
           fit_intercept=True, l1_ratio=0.15, learning_rate='invscaling',
           loss='squared_loss', max_iter=None, n_iter=None, penalty='l2',
           power_t=0.25, random_state=None, shuffle=True, tol=None,
           verbose=0, warm_start=False)
    See also
     
    In [14]:
     
     
     
     
     
    %time sgd_reg.fit(X_train_standard,y_train)
    sgd_reg.score(X_test_standard,y_test)
     
     
     
    c:usersqq123anaconda3libsite-packagessklearnlinear_modelstochastic_gradient.py:128: FutureWarning: max_iter and tol parameters have been added in <class 'sklearn.linear_model.stochastic_gradient.SGDRegressor'> in 0.19. If both are left unset, they default to max_iter=5 and tol=None. If tol is not None, max_iter defaults to max_iter=1000. From 0.21, default max_iter will be 1000, and default tol will be 1e-3.
      "and default tol will be 1e-3." % type(self), FutureWarning)
    
     
    Wall time: 178 ms
    
    Out[14]:
    0.8046509470914386
    In [15]:
     
     
     
     
     
    sgd_reg = SGDRegressor(n_iter=100)
    %time sgd_reg.fit(X_train_standard,y_train)
    sgd_reg.score(X_test_standard,y_test)
     
     
     
    Wall time: 5 ms
    
     
    c:usersqq123anaconda3libsite-packagessklearnlinear_modelstochastic_gradient.py:117: DeprecationWarning: n_iter parameter is deprecated in 0.19 and will be removed in 0.21. Use max_iter and tol instead.
      DeprecationWarning)
    
    Out[15]:
    0.8131724938971269
  • 相关阅读:
    201871010113-刘兴瑞 实验二 个人项目—《背包问题》项目报告
    201871010113-刘兴瑞 实验一 软件工程准备—初识软件工程
    201871010113-刘兴瑞《面向对象程序设计(java)》课程学习总结
    201871010113-刘兴瑞《面向对象程序设计(java)》第十七周学习总结
    201871010113-刘兴瑞《面向对象程序设计(java)》第十六周学习总结
    201871010113-刘兴瑞《面向对象程序设计(java)》第十五周学习总结
    201871010113-刘兴瑞《面向对象程序设计(java)》第十四周学习总结
    201871010113-刘兴瑞《面向对象程序设计(java)》第十三周学习总结
    《2019面向对象程序设计(java)课程学习进度条》
    201871010114-李岩松 实验四 团队作业1:软件研发团队组建(个人博客)
  • 原文地址:https://www.cnblogs.com/romannista/p/10761715.html
Copyright © 2020-2023  润新知