@@ -476,178 +476,6 @@ def _check_multi_class(multi_class, solver, n_classes):
476476 return multi_class
477477
478478
479- @deprecated ('logistic_regression_path was deprecated in version 0.21 and '
480- 'will be removed in version 0.23.0' )
481- def logistic_regression_path (X , y , pos_class = None , Cs = 10 , fit_intercept = True ,
482- max_iter = 100 , tol = 1e-4 , verbose = 0 ,
483- solver = 'lbfgs' , coef = None ,
484- class_weight = None , dual = False , penalty = 'l2' ,
485- intercept_scaling = 1. , multi_class = 'auto' ,
486- random_state = None , check_input = True ,
487- max_squared_sum = None , sample_weight = None ,
488- l1_ratio = None ):
489- """Compute a Logistic Regression model for a list of regularization
490- parameters.
491-
492- This is an implementation that uses the result of the previous model
493- to speed up computations along the set of solutions, making it faster
494- than sequentially calling LogisticRegression for the different parameters.
495- Note that there will be no speedup with liblinear solver, since it does
496- not handle warm-starting.
497-
498- .. deprecated:: 0.21
499- ``logistic_regression_path`` was deprecated in version 0.21 and will
500- be removed in 0.23.
501-
502- Read more in the :ref:`User Guide <logistic_regression>`.
503-
504- Parameters
505- ----------
506- X : array-like or sparse matrix, shape (n_samples, n_features)
507- Input data.
508-
509- y : array-like, shape (n_samples,) or (n_samples, n_targets)
510- Input data, target values.
511-
512- pos_class : int, None
513- The class with respect to which we perform a one-vs-all fit.
514- If None, then it is assumed that the given problem is binary.
515-
516- Cs : int | array-like, shape (n_cs,)
517- List of values for the regularization parameter or integer specifying
518- the number of regularization parameters that should be used. In this
519- case, the parameters will be chosen in a logarithmic scale between
520- 1e-4 and 1e4.
521-
522- fit_intercept : bool
523- Whether to fit an intercept for the model. In this case the shape of
524- the returned array is (n_cs, n_features + 1).
525-
526- max_iter : int
527- Maximum number of iterations for the solver.
528-
529- tol : float
530- Stopping criterion. For the newton-cg and lbfgs solvers, the iteration
531- will stop when ``max{|g_i | i = 1, ..., n} <= tol``
532- where ``g_i`` is the i-th component of the gradient.
533-
534- verbose : int
535- For the liblinear and lbfgs solvers set verbose to any positive
536- number for verbosity.
537-
538- solver : {'lbfgs', 'newton-cg', 'liblinear', 'sag', 'saga'}
539- Numerical solver to use.
540-
541- coef : array-like, shape (n_features,), default None
542- Initialization value for coefficients of logistic regression.
543- Useless for liblinear solver.
544-
545- class_weight : dict or 'balanced', optional
546- Weights associated with classes in the form ``{class_label: weight}``.
547- If not given, all classes are supposed to have weight one.
548-
549- The "balanced" mode uses the values of y to automatically adjust
550- weights inversely proportional to class frequencies in the input data
551- as ``n_samples / (n_classes * np.bincount(y))``.
552-
553- Note that these weights will be multiplied with sample_weight (passed
554- through the fit method) if sample_weight is specified.
555-
556- dual : bool
557- Dual or primal formulation. Dual formulation is only implemented for
558- l2 penalty with liblinear solver. Prefer dual=False when
559- n_samples > n_features.
560-
561- penalty : str, 'l1', 'l2', or 'elasticnet'
562- Used to specify the norm used in the penalization. The 'newton-cg',
563- 'sag' and 'lbfgs' solvers support only l2 penalties. 'elasticnet' is
564- only supported by the 'saga' solver.
565-
566- intercept_scaling : float, default 1.
567- Useful only when the solver 'liblinear' is used
568- and self.fit_intercept is set to True. In this case, x becomes
569- [x, self.intercept_scaling],
570- i.e. a "synthetic" feature with constant value equal to
571- intercept_scaling is appended to the instance vector.
572- The intercept becomes ``intercept_scaling * synthetic_feature_weight``.
573-
574- Note! the synthetic feature weight is subject to l1/l2 regularization
575- as all other features.
576- To lessen the effect of regularization on synthetic feature weight
577- (and therefore on the intercept) intercept_scaling has to be increased.
578-
579- multi_class : {'ovr', 'multinomial', 'auto'}, default='auto'
580- If the option chosen is 'ovr', then a binary problem is fit for each
581- label. For 'multinomial' the loss minimised is the multinomial loss fit
582- across the entire probability distribution, *even when the data is
583- binary*. 'multinomial' is unavailable when solver='liblinear'.
584- 'auto' selects 'ovr' if the data is binary, or if solver='liblinear',
585- and otherwise selects 'multinomial'.
586-
587- .. versionadded:: 0.18
588- Stochastic Average Gradient descent solver for 'multinomial' case.
589- .. versionchanged:: 0.22
590- Default changed from 'ovr' to 'auto' in 0.22.
591-
592- random_state : int, RandomState instance or None, optional, default None
593- The seed of the pseudo random number generator to use when shuffling
594- the data. If int, random_state is the seed used by the random number
595- generator; If RandomState instance, random_state is the random number
596- generator; If None, the random number generator is the RandomState
597- instance used by `np.random`. Used when ``solver`` == 'sag' or
598- 'liblinear'.
599-
600- check_input : bool, default True
601- If False, the input arrays X and y will not be checked.
602-
603- max_squared_sum : float, default None
604- Maximum squared sum of X over samples. Used only in SAG solver.
605- If None, it will be computed, going through all the samples.
606- The value should be precomputed to speed up cross validation.
607-
608- sample_weight : array-like, shape(n_samples,) optional
609- Array of weights that are assigned to individual samples.
610- If not provided, then each sample is given unit weight.
611-
612- l1_ratio : float or None, optional (default=None)
613- The Elastic-Net mixing parameter, with ``0 <= l1_ratio <= 1``. Only
614- used if ``penalty='elasticnet'``. Setting ``l1_ratio=0`` is equivalent
615- to using ``penalty='l2'``, while setting ``l1_ratio=1`` is equivalent
616- to using ``penalty='l1'``. For ``0 < l1_ratio <1``, the penalty is a
617- combination of L1 and L2.
618-
619- Returns
620- -------
621- coefs : ndarray, shape (n_cs, n_features) or (n_cs, n_features + 1)
622- List of coefficients for the Logistic Regression model. If
623- fit_intercept is set to True then the second dimension will be
624- n_features + 1, where the last item represents the intercept. For
625- ``multiclass='multinomial'``, the shape is (n_classes, n_cs,
626- n_features) or (n_classes, n_cs, n_features + 1).
627-
628- Cs : ndarray
629- Grid of Cs used for cross-validation.
630-
631- n_iter : array, shape (n_cs,)
632- Actual number of iteration for each Cs.
633-
634- Notes
635- -----
636- You might get slightly different results with the solver liblinear than
637- with the others since this uses LIBLINEAR which penalizes the intercept.
638-
639- .. versionchanged:: 0.19
640- The "copy" parameter was removed.
641- """
642-
643- return _logistic_regression_path (
644- X , y , pos_class = None , Cs = 10 , fit_intercept = True , max_iter = 100 ,
645- tol = 1e-4 , verbose = 0 , solver = 'lbfgs' , coef = None , class_weight = None ,
646- dual = False , penalty = 'l2' , intercept_scaling = 1. , multi_class = 'auto' ,
647- random_state = None , check_input = True , max_squared_sum = None ,
648- sample_weight = None , l1_ratio = None )
649-
650-
651479def _logistic_regression_path (X , y , pos_class = None , Cs = 10 , fit_intercept = True ,
652480 max_iter = 100 , tol = 1e-4 , verbose = 0 ,
653481 solver = 'lbfgs' , coef = None ,
0 commit comments