\c@sdZddlZddlZddlmZmZddlmZm Z ddl m Z m Z ddl m Z mZdd l mZdd l mZmZdd lmZmZmZdd lmZdd lmZddlmZddlmZddlmZddl m!Z!m"Z"ddl#m$Z$m%Z%ddlm&Z&ddlm'Z'ddl#m(Z(ddl#m)Z)ddl#m*Z*ddl#m+Z+ddl#m,Z,ddl#m-Z-ddl#m.Z.ddl#m/Z/ddl0m1Z1idd6dd6d d!6d"d#6d$d%6d&d'6Z2id(d)6dd*6dd+6d d,6Z3d-Z4d.e5fd/YZ6d0ej7eeefd1YZ8d2Z9e:d3Z;d4ej7ee8e fd5YZ<d6e<fd7YZ=d8e8efd9YZ>d:e>fd;YZ?dS(<sFClassification and regression using Stochastic Gradient Descent (SGD).iN(tABCMetatabstractmethodi(tParalleltdelayed(tclonet is_classifieri(tLinearClassifierMixintSparseCoefMixin(t make_dataset(t BaseEstimatortRegressorMixin(t check_arraytcheck_random_statet check_X_y(tsafe_sparse_dot(t_check_partial_fit_first_call(tcheck_is_fitted(tConvergenceWarning(tsix(tStratifiedShuffleSplitt ShuffleSplit(t plain_sgdt average_sgd(tcompute_class_weight(t deprecated(tHinge(t SquaredHinge(tLog(t ModifiedHuber(t SquaredLoss(tHuber(tEpsilonInsensitive(tSquaredEpsilonInsensitive(t_joblib_parallel_argstconstanttoptimalit invscalingitadaptiveitpa1itpa2itnonetl2tl1t elasticnetg?t_ValidationScoreCallbackcBs#eZdZddZdZRS(s5Callback for early stopping based on validation scorecCsUt||_d|j_|dk r6||j_n||_||_||_dS(Ni(Rt estimatortt_tNonetclasses_tX_valty_valtsample_weight_val(tselfR-R1R2R3tclasses((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyt__init__3s    cCsL|j}|jdd|_tj||_|j|j|j|j S(Nii( R-treshapetcoef_tnpt atleast_1dt intercept_tscoreR1R2R3(R4tcoeft intercepttest((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyt__call__=s N(t__name__t __module__t__doc__R/R6R@(((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyR,0s tBaseSGDcBseZdZddddeddedddddd edd eedd Zd Zed Z eedZ dZ dZ dZ dZdddZdZddZRS(s1Base class for SGD classification and regression.R)g-C6?g?g333333?ig?R#gg?icCs||_||_| |_| |_||_||_||_||_| |_| |_ | |_ ||_ ||_ ||_ ||_||_||_||_||_||_||_|jdtdS(Nt set_max_iter(tlosstpenaltyt learning_ratetepsilontalphatCtl1_ratiot fit_intercepttshufflet random_statetverboseteta0tpower_ttearly_stoppingtvalidation_fractiontn_iter_no_changet warm_starttaveragetn_itertmax_iterttolt_validate_paramstFalse(R4RFRGRJRKRLRMRYRZRNRPRIRORHRQRRRSRTRURVRWRX((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyR6Gs,                     cOs-tt|j|||jdt|S(NRE(tsuperRDt set_paramsR[R\(R4targstkwargs((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyR^gscCsdS(s Fit model.N((R4tXty((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pytfitlscCst|jts!tdnt|jtsBtdn|jr`|r`tdn|jdk r|jdkrtd|jnd|jkodknstdn|jdkrtd n|j d krtd nd|j kodkns,td n|j dkr\|j dkr\tdq\n|j dkr|jdkrtdn|j |j|j|j |j|jkrtd|jn|sdS|j|_|jdk rtjdt|j}d|_n|jdkrh|jdkrh|s_tjdt|jtnd}nM|jdkrtjdt|jtn|jdk r|jnd}||_dS(sValidate input params. s$shuffle must be either True or Falses+early_stopping must be either True or Falses/early_stopping should be False with partial_fitismax_iter must be > zero. Got %fgg?sl1_ratio must be in [0, 1]salpha must be >= 0isn_iter_no_change must be >= 1s%validation_fraction must be in ]0, 1[R"R$R%seta0 must be > 0R#sgalpha must be > 0 since learning_rate is 'optimal'. alpha is used to compute the optimal learning rate.sThe loss %s is not supported. Nsan_iter parameter is deprecated in 0.19 and will be removed in 0.21. Use max_iter and tol instead.smax_iter and tol parameters have been added in %s 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.ismax_iter and tol parameters have been added in %s in 0.19. If max_iter is set but tol is left unset, the default value for tol in 0.19 and 0.20 will be None (which is equivalent to -infinity, so it has no effect) but will change in 0.21 to 1e-3. Specify tol to silence this warning.i(R"R$R%(t isinstanceRNtboolt ValueErrorRSRYR/RLRJRURTRHRQt_get_penalty_typeRGt_get_learning_rate_typeRFtloss_functionsRZt_tolRXtwarningstwarntDeprecationWarningttypeRAt FutureWarningt _max_iter(R4REtfor_partial_fitRY((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyR[ps^         cCssyK|j|}|d|d}}|dkr@|jf}n||SWn!tk rntd|nXdS( s7Get concrete ``LossFunction`` object for str ``loss``. iithubertepsilon_insensitivetsquared_epsilon_insensitivesThe loss %s is not supported. N(RrRsRt(RiRItKeyErrorRf(R4RFtloss_t loss_classR_((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyt_get_loss_functions   cCs4y t|SWn!tk r/td|nXdS(Ns#learning rate %s is not supported. (tLEARNING_RATE_TYPESRuRf(R4RH((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyRhs   cCsFt|j}y t|SWn!tk rAtd|nXdS(NsPenalty %s is not supported. (tstrtlowert PENALTY_TYPESRuRf(R4RG((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyRgs   cCsq|dkr-tj|dtjdd}ntj|dtjdd}|jd|krmtdn|S(sSet the sample weight array.tdtypetorderRKis+Shapes of X and sample_weight do not match.N(R/R9tonestfloat64tasarraytshapeRf(R4t sample_weightt n_samples((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyt_validate_sample_weights ! cCs}|dkr|d k r]tj|dd}|j||fkrQtdn||_n'tj||fdtjdd|_|d k rtj|dd}|j|fkrtdn||_qtj|dtjdd|_n |d k rYtj|dtjdd}|j }|j|fkrMtdn||_n!tj|dtjdd|_|d k rtj|dtj}|jd kr|jd krtdn|j d|_n!tjddtjdd|_|j d kry|j|_ |j|_ tj|jjdtjdd|_tj|j jdtjdd|_nd S( s4Allocate mem for parameters; initialize if provided.iR~RKs+Provided ``coef_`` does not match dataset. R}s/Provided intercept_init does not match dataset.s*Provided coef_init does not match dataset.iiN(i((R/R9RRRfR8tzerosRR;travelR7RWtstandard_coef_tstandard_intercept_t average_coef_taverage_intercept_(R4t n_classest n_featurest coef_inittintercept_init((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyt_allocate_parameter_memsP             !     cCs|jd}tj|dtj}|js2|St|rGt}nt}|d|jd|j }t |j tjd|jddf|\}}|jddks|jddkrt d||j|jd|jdfnd||<|S(sGSplit the dataset between training set and validation set. Parameters ---------- y : array, shape (n_samples, ) Target values. Returns ------- validation_mask : array, shape (n_samples, ) Equal to 1 on the validation set, 0 on the training set. iR}t test_sizeRORisSplitting %d samples into a train set and a validation set with validation_fraction=%r led to an empty set (%d and %d samples). Please either change validation_fraction, increase number of samples, or disable early_stopping.( RR9Rtuint8RSRRRRTROtnexttsplitRf(R4RbRtvalidation_maskt splitter_typetcvt idx_traintidx_val((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyt_make_validation_splits"     7& cCs2|js dSt|||||||d|S(NR5(RSR/R,(R4RRaRbRR5((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyt_make_validation_score_cbEs  N(RARBRCtTrueR/R\R6R^RRcR[RxRhRgRRRR(((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyRDDs&       K    < 'cCs)tj|jdtjdd}d|||j|k((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyt_prepare_fit_binaryOs(!        c Cst|||\} } }}}| jd|jdkoM| jdknsXtt|| | \}}|j|j}|j|}| dkr|j| } nt j ddgd| j }|j | || | d|}t |j}|jdt jt jj}|jdk r3|jnt j }|jst| ||j||||j|| |j|t|j||t|jt|jt|j||| ||j|j |j!|}nt"| ||||j||||j|| |j|t|j||t|jt|jt|j||| ||j|j |j!||j\}}}}}t#|j$dkr||j%dRRtdatasettintercept_decayt penalty_typetlearning_rate_typeR5tvalidation_score_cbROtseedRZtresultt standard_coeftstandard_intercepttn_iter_((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyt fit_binaryosN1!7 !"        tBaseSGDClassifiercBs4eZi edfd6edfd6edfd6efd6efd6efd6eefd6e efd 6e efd 6Z e dd d d e dde dedddddedddeeddZeeddZdZddddZdZdZdddZddddZRS(g?thinget squared_hingegt perceptrontlogtmodified_hubert squared_lossRrRsRtR)g-C6?g333333?iR#g?g?ic)Cstt|jd|d|d|d|d|d|d|d|d | d | d | d | d |d|d|d|d|d|d|d|||_| |_dS(NRFRGRJRLRMRYRZRNRPRIRORHRQRRRSRTRURVRWRX(R]RR6t class_weighttn_jobs(R4RFRGRJRLRMRYRZRNRPRIRRORHRQRRRSRTRURRVRWRX((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyR6s     srAttribute loss_function was deprecated in version 0.19 and will be removed in 0.21. Use ``loss_function_`` insteadcCs|jS(N(R(R4((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyt loss_functionsc  Cst||ddtjdddt\}}|j\} } t|||jjd}t|j|j||_ |j | | } t |dddks| dk r|j || | | n9| |jjdkrtd | |jjdfn|j||_t|d s)d |_n|d krf|j||d |d|d|d| d|nM|d kr|j||d |d|d|d| d|ntd||S(NtcsrR}R~RKtaccept_large_sparseiR8is6Number of features %d does not match previous data %d.R.g?iRJRHRRYs>The number of classes has to be greater than one; got %d class(R R9RR\RRR0RRt_expanded_class_weightRtgetattrR/RR8RfRxRthasattrR.t_fit_multiclasst _fit_binary(R4RaRbRJRKRFRHRYR5RRRRRR((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyt _partial_fits>  $        c Cs|jt|dr%d|_nt||ddtjdddt\}}|j\} } tj |} |j rt|dr|dkr|j }n|dkr|j }qnd|_ d|_ |j dkr |j |_|j |_d|_d|_nd |_|j|||||||j| | || |jdk r|jtj kr|j|jkrtjd tn|S( NR0RR}R~RKRR8ig?shMaximum number of iteration reached before convergence. Consider increasing max_iter to improve the fit.(R[RR/R0R R9RR\RtuniqueRVR8R;RWRRRRR.RRpRjRRRkRlR( R4RaRbRJRKRFRHRRRRRR5((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyt_fit3s6            "  c Cst|d|||||||jd|jd| \}} } |j| |jd7_| |_|jdkr|j|jdkr|jjdd|_|j |_ q|j jdd|_t j | |_|j|_ n'|jdd|_t j | |_ dS(s$Fit a binary classifier on X and y. iiiN(RRR.RRRWRR7R8RR;RR9R:R( R4RaRbRJRKRRHRYR=R>R((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyR^s    c sRjtdjdjtddf dttjD}d} x?t|D]1\} \} } } | j | sgig?N(RRRRPR!trangeRR0t enumerateR;RR.RRRWRR8RRR9R:R(R4RaRbRJRKRHRRYRRRt_R>tn_iter_i(( RKRaRJRHRYRR4RRbsGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyRxs$ $"    cCs|jdt|jdkr:tdj|jn|j||d|jddd|jd|jd d d |d |d ddd S(sPerform one epoch of stochastic gradient descent on given samples. Internally, this method uses ``max_iter = 1``. Therefore, it is not guaranteed that a minimum of the cost function is reached after calling it once. Matters such as objective convergence and early stopping should be handled by the user. Parameters ---------- X : {array-like, sparse matrix}, shape (n_samples, n_features) Subset of the training data y : numpy array, shape (n_samples,) Subset of the target values classes : array, shape (n_classes,) Classes across all calls to partial_fit. Can be obtained by via `np.unique(y_all)`, where y_all is the target vector of the entire dataset. This argument is required for the first call to partial_fit and can be omitted in the subsequent calls. Note that y doesn't need to contain all labels in `classes`. sample_weight : array-like, shape (n_samples,), optional Weights applied to individual samples. If not provided, uniform weights are assumed. Returns ------- self : returns an instance of self. RqtbalancedsFclass_weight '{0}' is not supported for partial_fit. In order to use 'balanced' weights, use compute_class_weight('{0}', classes, y). In place of y you can us a large enough sample of the full training set target to properly estimate the class frequency distributions. Pass the resulting weights as the class_weight parameter.RJRKg?RFRHRYiR5RRR(RN( R[RRRftformatRRJRFRHR/(R4RaRbR5R((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyt partial_fits  ' cCsC|j||d|jddd|jd|jd|d|d|S( sFit linear model with Stochastic Gradient Descent. Parameters ---------- X : {array-like, sparse matrix}, shape (n_samples, n_features) Training data y : numpy array, shape (n_samples,) Target values coef_init : array, shape (n_classes, n_features) The initial coefficients to warm-start the optimization. intercept_init : array, shape (n_classes,) The initial intercept to warm-start the optimization. sample_weight : array-like, shape (n_samples,), optional Weights applied to individual samples. If not provided, uniform weights are assumed. These weights will be multiplied with class_weight (passed through the constructor) if class_weight is specified Returns ------- self : returns an instance of self. RJRKg?RFRHRRR(RRJRFRH(R4RaRbRRR((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyRcs N(RARBRRRRRRtDEFAULT_EPSILONRR RiRRR/R\R6tpropertyRRRRRRRRc(((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyRs<               /*  &/t SGDClassifiercBseZdZddddeddededdddded d deedd Zd Ze d Z dZ e dZ dZ RS(s&Linear classifiers (SVM, logistic regression, a.o.) with SGD training. This estimator implements regularized linear models with stochastic gradient descent (SGD) learning: 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). SGD allows minibatch (online/out-of-core) learning, see the partial_fit method. For best results using the default learning rate schedule, the data should have zero mean and unit variance. This implementation works with data represented as dense or sparse arrays of floating point values for the features. The model it fits can be controlled with the loss parameter; by default, it fits a linear support vector machine (SVM). 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. Read more in the :ref:`User Guide `. Parameters ---------- loss : str, default: 'hinge' The loss function to be used. Defaults to 'hinge', which gives a linear SVM. The possible options are 'hinge', 'log', 'modified_huber', 'squared_hinge', 'perceptron', or a regression loss: 'squared_loss', 'huber', 'epsilon_insensitive', or 'squared_epsilon_insensitive'. The 'log' loss gives logistic regression, a probabilistic classifier. 'modified_huber' is another smooth loss that brings tolerance to outliers as well as probability estimates. 'squared_hinge' is like hinge but is quadratically penalized. 'perceptron' is the linear loss used by the perceptron algorithm. The other losses are designed for regression but can be useful in classification as well; see SGDRegressor for a description. 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. n_jobs : int or None, optional (default=None) The number of CPUs to use to do the OVA (One Versus All, for multi-class problems) computation. ``None`` means 1 unless in a :obj:`joblib.parallel_backend` context. ``-1`` means using all processors. See :term:`Glossary ` for more details. 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': [default] eta = 1.0 / (alpha * (t + t0)) where t0 is chosen by a heuristic proposed by Leon Bottou. 'invscaling': eta = eta0 / pow(t, power_t) 'adaptive': eta = eta0, as long as the training keeps decreasing. Each time n_iter_no_change consecutive epochs fail to decrease the training loss by tol or fail to increase validation score by tol if early_stopping is True, the current learning rate is divided by 5. eta0 : double The initial learning rate for the 'constant', 'invscaling' or 'adaptive' schedules. The default value is 0.0 as eta0 is not used by the default schedule 'optimal'. power_t : double The exponent for inverse scaling learning rate [default 0.5]. early_stopping : bool, default=False Whether to use early stopping to terminate training when validation score is not improving. If set to True, it will automatically set aside a fraction of training data as validation and terminate training when validation score is not improving by at least tol for n_iter_no_change consecutive epochs. .. versionadded:: 0.20 validation_fraction : float, default=0.1 The proportion of training data to set aside as validation set for early stopping. Must be between 0 and 1. Only used if early_stopping is True. .. versionadded:: 0.20 n_iter_no_change : int, default=5 Number of iterations with no improvement to wait before early stopping. .. versionadded:: 0.20 class_weight : dict, {class_label: weight} or "balanced" or None, optional Preset for the class_weight fit parameter. Weights associated with classes. If not given, all classes are supposed to have weight one. 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))`` 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. See :term:`the Glossary `. Repeatedly calling fit or partial_fit when warm_start is True can result in a different solution than when calling fit a single time because of the way the data is shuffled. If a dynamic learning rate is used, the learning rate is adapted depending on the number of samples already seen. Calling ``fit`` resets this counter, while ``partial_fit`` will result in increasing the existing counter. 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 (1, n_features) if n_classes == 2 else (n_classes, n_features) Weights assigned to the features. intercept_ : array, shape (1,) if n_classes == 2 else (n_classes,) Constants in decision function. n_iter_ : int The actual number of iterations to reach the stopping criterion. For multiclass fits, it is the maximum over every binary fit. loss_function_ : concrete ``LossFunction`` Examples -------- >>> import numpy as np >>> from sklearn import linear_model >>> X = np.array([[-1, -1], [-2, -1], [1, 1], [2, 1]]) >>> Y = np.array([1, 1, 2, 2]) >>> clf = linear_model.SGDClassifier(max_iter=1000, tol=1e-3) >>> clf.fit(X, Y) ... #doctest: +NORMALIZE_WHITESPACE SGDClassifier(alpha=0.0001, average=False, class_weight=None, early_stopping=False, epsilon=0.1, eta0=0.0, fit_intercept=True, l1_ratio=0.15, learning_rate='optimal', loss='hinge', max_iter=1000, n_iter=None, n_iter_no_change=5, n_jobs=None, penalty='l2', power_t=0.5, random_state=None, shuffle=True, tol=0.001, validation_fraction=0.1, verbose=0, warm_start=False) >>> print(clf.predict([[-0.8, -1]])) [1] See also -------- sklearn.svm.LinearSVC, LogisticRegression, Perceptron RR)g-C6?g333333?iR#gg?g?ic-Cstt|jd|d|d|d|d|d|d|d|d | d | d | d | d | d|d|d|d|d|d|d|d|d|dS(NRFRGRJRLRMRYRZRNRPRIRRORHRQRRRSRTRURRVRWRX(R]RR6(R4RFRGRJRLRMRYRZRNRPRIRRORHRQRRRSRTRURRVRWRX((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyR6s  cCs)|jdkr%td|jndS(NRRs3probability estimates are not available for loss=%r(RR(RFtAttributeError(R4((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyt _check_probascCs|j|jS(sLProbability estimates. This method is only available for log loss and modified Huber loss. Multiclass probability estimates are derived from binary (one-vs.-rest) estimates by simple normalization, as recommended by Zadrozny and Elkan. Binary probability estimates for loss="modified_huber" are given by (clip(decision_function(X), -1, 1) + 1) / 2. For other loss functions it is necessary to perform proper probability calibration by wrapping the classifier with :class:`sklearn.calibration.CalibratedClassifierCV` instead. Parameters ---------- X : {array-like, sparse matrix}, shape (n_samples, n_features) Returns ------- array, shape (n_samples, n_classes) Returns the probability of the sample for each class in the model, where classes are ordered as they are in `self.classes_`. References ---------- Zadrozny and Elkan, "Transforming classifier scores into multiclass probability estimates", SIGKDD'02, http://www.research.ibm.com/people/z/zadrozny/kdd2002-Transf.pdf The justification for the formula in the loss="modified_huber" case is in the appendix B in: http://jmlr.csail.mit.edu/papers/volume2/zhang02c/zhang02c.pdf (Rt_predict_proba(R4((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyt predict_probas$ cCst|d|jdkr)|j|S|jdkrot|jdk}|j|}|rtj|jddf}|dddf}n|}tj |dd||d7}|d :}|r|dddfc|8<|}ny|j d d}|dk}tj |rKd||ddf`. 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)) where t0 is chosen by a heuristic proposed by Leon Bottou. 'invscaling': [default] eta = eta0 / pow(t, power_t) 'adaptive': eta = eta0, as long as the training keeps decreasing. Each time n_iter_no_change consecutive epochs fail to decrease the training loss by tol or fail to increase validation score by tol if early_stopping is True, the current learning rate is divided by 5. eta0 : double The initial learning rate for the 'constant', 'invscaling' or 'adaptive' schedules. The default value is 0.0 as eta0 is not used by the default schedule 'optimal'. power_t : double The exponent for inverse scaling learning rate [default 0.5]. early_stopping : bool, default=False Whether to use early stopping to terminate training when validation score is not improving. If set to True, it will automatically set aside a fraction of training data as validation and terminate training when validation score is not improving by at least tol for n_iter_no_change consecutive epochs. .. versionadded:: 0.20 validation_fraction : float, default=0.1 The proportion of training data to set aside as validation set for early stopping. Must be between 0 and 1. Only used if early_stopping is True. .. versionadded:: 0.20 n_iter_no_change : int, default=5 Number of iterations with no improvement to wait before early stopping. .. versionadded:: 0.20 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. See :term:`the Glossary `. Repeatedly calling fit or partial_fit when warm_start is True can result in a different solution than when calling fit a single time because of the way the data is shuffled. If a dynamic learning rate is used, the learning rate is adapted depending on the number of samples already seen. Calling ``fit`` resets this counter, while ``partial_fit`` will result in increasing the existing counter. 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(max_iter=1000, tol=1e-3) >>> clf.fit(X, y) ... #doctest: +NORMALIZE_WHITESPACE SGDRegressor(alpha=0.0001, average=False, early_stopping=False, epsilon=0.1, eta0=0.01, fit_intercept=True, l1_ratio=0.15, learning_rate='invscaling', loss='squared_loss', max_iter=1000, n_iter=None, n_iter_no_change=5, penalty='l2', power_t=0.25, random_state=None, shuffle=True, tol=0.001, validation_fraction=0.1, verbose=0, warm_start=False) See also -------- Ridge, ElasticNet, Lasso, sklearn.svm.SVR RR)g-C6?g333333?iR$g{Gz?g?g?ic)Cstt|jd|d|d|d|d|d|d|d|d | d | d | d | d | d|d|d|d|d|d|d|dS(NRFRGRJRLRMRYRZRNRPRIRORHRQRRRSRTRURVRWRX(R]RR6(R4RFRGRJRLRMRYRZRNRPRIRORHRQRRRSRTRURVRWRX((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyR6/s  N(RARBRCRR/RR\R6(((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pyRfs      (@RCtnumpyR9RktabcRRt utils._joblibRRtbaseRRRRRR R tutilsR R R t utils.extmathRtutils.multiclassRtutils.validationRt exceptionsRt externalsRtmodel_selectionRRtsgd_fastRRRRRRRRRRRR t utils.fixesR!RyR|RtobjectR,twith_metaclassRDRR/RRRRR(((sGlib/python2.7/site-packages/sklearn/linear_model/stochastic_gradient.pytsX  "% ! hj