o tBh>p@sdZddlZddlZddlZddlmZddlm Z m Z m Z m Z m Z ddlmZmZddlmZddlmZmZmZdd lmZmZdd lmZd d lmZGd ddeeZGdddeZGdddeZ GdddeZ!dS)z> Generalized Linear Models with Exponential Dispersion Family N)TweedieDistribution) HalfGammaLossHalfPoissonLossHalfSquaredErrorHalfTweedieLossHalfTweedieLossIdentity) BaseEstimatorRegressorMixin)_check_optimize_result) check_scalar check_array deprecated)check_is_fitted_check_sample_weight)_openmp_effective_n_threads)LinearModelLossc@sreZdZdZdddddddd d d Zdd dZddZddZdddZddZ ddZ e de ddZ d S)_GeneralizedLinearRegressoraRegression via a penalized Generalized Linear Model (GLM). GLMs based on a reproductive Exponential Dispersion Model (EDM) aim at fitting and predicting the mean of the target y as y_pred=h(X*w) with coefficients w. Therefore, the fit minimizes the following objective function with L2 priors as regularizer:: 1/(2*sum(s_i)) * sum(s_i * deviance(y_i, h(x_i*w)) + 1/2 * alpha * ||w||_2^2 with inverse link function h, s=sample_weight and per observation (unit) deviance deviance(y_i, h(x_i*w)). Note that for an EDM, 1/2 * deviance is the negative log-likelihood up to a constant (in w) term. The parameter ``alpha`` corresponds to the lambda parameter in glmnet. Instead of implementing the EDM family and a link function separately, we directly use the loss functions `from sklearn._loss` which have the link functions included in them for performance reasons. We pick the loss functions that implement (1/2 times) EDM deviances. Read more in the :ref:`User Guide `. .. versionadded:: 0.23 Parameters ---------- alpha : float, default=1 Constant that multiplies the penalty term and thus determines the regularization strength. ``alpha = 0`` is equivalent to unpenalized GLMs. In this case, the design matrix `X` must have full column rank (no collinearities). Values must be in the range `[0.0, inf)`. fit_intercept : bool, default=True Specifies if a constant (a.k.a. bias or intercept) should be added to the linear predictor (X @ coef + intercept). solver : 'lbfgs', default='lbfgs' Algorithm to use in the optimization problem: 'lbfgs' Calls scipy's L-BFGS-B optimizer. max_iter : int, default=100 The maximal number of iterations for the solver. Values must be in the range `[1, inf)`. tol : float, default=1e-4 Stopping criterion. For the lbfgs solver, the iteration will stop when ``max{|g_j|, j = 1, ..., d} <= tol`` where ``g_j`` is the j-th component of the gradient (derivative) of the objective function. Values must be in the range `(0.0, inf)`. warm_start : bool, default=False If set to ``True``, reuse the solution of the previous call to ``fit`` as initialization for ``coef_`` and ``intercept_``. verbose : int, default=0 For the lbfgs solver set verbose to any positive number for verbosity. Values must be in the range `[0, inf)`. Attributes ---------- coef_ : array of shape (n_features,) Estimated coefficients for the linear predictor (`X @ coef_ + intercept_`) in the GLM. intercept_ : float Intercept (a.k.a. bias) added to linear predictor. n_iter_ : int Actual number of iterations used in the solver. _base_loss : BaseLoss, default=HalfSquaredError() This is set during fit via `self._get_loss()`. A `_base_loss` contains a specific loss function as well as the link function. The loss to be minimized specifies the distributional assumption of the GLM, i.e. the distribution from the EDM. Here are some examples: ======================= ======== ========================== _base_loss Link Target Domain ======================= ======== ========================== HalfSquaredError identity y any real number HalfPoissonLoss log 0 <= y HalfGammaLoss log 0 < y HalfTweedieLoss log dependend on tweedie power HalfTweedieLossIdentity identity dependend on tweedie power ======================= ======== ========================== The link function of the GLM, i.e. mapping from linear predictor `X @ coeff + intercept` to prediction `y_pred`. For instance, with a log link, we have `y_pred = exp(X @ coeff + intercept)`. ?Tlbfgsd-C6?Fralpha fit_interceptsolvermax_itertol warm_startverbosecCs.||_||_||_||_||_||_||_dSNr)selfrrrrrrr r#t/var/www/html/riverr-enterprise-integrations-main/venv/lib/python3.10/site-packages/sklearn/linear_model/_glm/glm.py__init__}s  z$_GeneralizedLinearRegressor.__init__Nc Cst|jdtjdddt|jtstd|j|j dvr*t|j j d|j |j }t|j dtj d d t|jd tjdd dt|jd tj dd t|jtsZtd|j|j||ddgtjtjgddd\}}|dkrutj}n tt|j|jtj}t||ddd}t|||d}|j\}}||_t|j|jd}|j|std|jj j d|| }|jrt!|dr|jrt"|j#t$|j%gf} n|j#} | j&|dd} n"|jrtj'|d |d} |jj((tj)||d| d<ntj'||d} |dkr8|j*} |j} t+} t,j-j.| | d d|j |jdkd |jd!t/t0j1d"|||| | fd#} t2d| |_3| j4} |jrJ| d|_%| d$d|_#|Sd|_%| |_#|S)%aFit a Generalized Linear Model. Parameters ---------- X : {array-like, sparse matrix} of shape (n_samples, n_features) Training data. y : array-like of shape (n_samples,) Target values. sample_weight : array-like of shape (n_samples,), default=None Sample weights. Returns ------- self : object Fitted model. rleft)name target_typemin_valinclude_boundariesz0The argument fit_intercept must be bool; got {0})rz$ supports only solvers 'lbfgs'; got r)r(r)r*rneitherr rz-The argument warm_start must be bool; got {0}csccsrTF) accept_sparsedtype y_numeric multi_outputrCr1order ensure_2dr1) base_lossr:Some value(s) of y are out of the valid range of the loss .coef_)copyweightszL-BFGS-Bg@@)maxiteriprintgtolftol)methodjacoptionsargsN)5r rnumbersReal isinstancerbool ValueErrorformatr __class____name__rIntegralrr r_validate_datanpfloat64float32minmaxr1r rshape _get_loss _base_lossrr9in_y_true_rangesumhasattr concatenater<array intercept_astypezeroslinkaverage loss_gradientrscipyoptimizeminimizefinfofloatepsr n_iter_x)r"Xy sample_weightr loss_dtype n_samples n_features linear_losscoeffuncl2_reg_strength n_threadsopt_resr#r#r$fits                   z_GeneralizedLinearRegressor.fitcCs:t||j|gdtjtjgdddd}||j|jS)aCompute the linear_predictor = `X @ coef_ + intercept_`. Note that we often use the term raw_prediction instead of linear predictor. Parameters ---------- X : {array-like, sparse matrix} of shape (n_samples, n_features) Samples. Returns ------- y_pred : array of shape (n_samples,) Returns predicted values of linear predictor. )r/r.cooTF)r0r1r7allow_ndreset)rrRrSrTrUr<r`)r"rnr#r#r$_linear_predictor3s z-_GeneralizedLinearRegressor._linear_predictorcCs||}|jj|}|S)a*Predict using GLM with feature matrix X. Parameters ---------- X : {array-like, sparse matrix} of shape (n_samples, n_features) Samples. Returns ------- y_pred : array of shape (n_samples,) Returns predicted values. )r~rZrcinverse)r"rnraw_predictiony_predr#r#r$predictMs z#_GeneralizedLinearRegressor.predictc Cs||}t||jddd}|durt|||jd}|j}||s+td|jdt |j |d}|durD||j d t |9}||||d d }|j tj||d }||t||j d |d d } d ||| |S) aJCompute D^2, the percentage of deviance explained. D^2 is a generalization of the coefficient of determination R^2. R^2 uses squared error and D^2 uses the deviance of this GLM, see the :ref:`User Guide `. D^2 is defined as :math:`D^2 = 1-\frac{D(y_{true},y_{pred})}{D_{null}}`, :math:`D_{null}` is the null deviance, i.e. the deviance of a model with intercept alone, which corresponds to :math:`y_{pred} = \bar{y}`. The mean :math:`\bar{y}` is averaged by sample_weight. Best possible score is 1.0 and it can be negative (because the model can be arbitrarily worse). Parameters ---------- X : {array-like, sparse matrix} of shape (n_samples, n_features) Test samples. y : array-like of shape (n_samples,) True values of target. sample_weight : array-like of shape (n_samples,), default=None Sample weights. Returns ------- score : float D^2 of self.predict(X) w.r.t. y. r4Fr5Nr8r:r;)y_truerr,)rrrprxr>)r~r r1rrZr[rMrPrSmeanconstant_to_optimal_zerorXr\rcrdtile) r"rnrorprr9constantdeviancey_mean deviance_nullr#r#r$score_s8 % z!_GeneralizedLinearRegressor.scorecCs|}d|d iS)Nrequires_positive_yg)rYr[)r"r9r#r#r$ _more_tagssz&_GeneralizedLinearRegressor._more_tagscCtS)a This is only necessary because of the link and power arguments of the TweedieRegressor. Note that we do not need to pass sample_weight to the loss class as this is only needed to set loss.constant_hessian on which GLMs do not rely. )rr"r#r#r$rYsz%_GeneralizedLinearRegressor._get_losszLAttribute `family` was deprecated in version 1.1 and will be removed in 1.3.cCs:t|trdSt|trdSt|trt|jdStd)z:Ensure backward compatibility for the time of deprecation.poissongammapowerzThis should never happen. You presumably accessed the deprecated `family` attribute from a subclass of the private scikit-learn class _GeneralizedLinearRegressor.)rKPoissonRegressorGammaRegressorTweedieRegressorrrrMrr#r#r$familys    z"_GeneralizedLinearRegressor.familyr!)rP __module__ __qualname____doc__r%rzr~rrrrYrpropertyrr#r#r#r$rs,a $ K rc8eZdZdZdddddddfd d Zd d ZZS) ra Generalized Linear Model with a Poisson distribution. This regressor uses the 'log' link function. Read more in the :ref:`User Guide `. .. versionadded:: 0.23 Parameters ---------- alpha : float, default=1 Constant that multiplies the penalty term and thus determines the regularization strength. ``alpha = 0`` is equivalent to unpenalized GLMs. In this case, the design matrix `X` must have full column rank (no collinearities). Values must be in the range `[0.0, inf)`. fit_intercept : bool, default=True Specifies if a constant (a.k.a. bias or intercept) should be added to the linear predictor (X @ coef + intercept). max_iter : int, default=100 The maximal number of iterations for the solver. Values must be in the range `[1, inf)`. tol : float, default=1e-4 Stopping criterion. For the lbfgs solver, the iteration will stop when ``max{|g_j|, j = 1, ..., d} <= tol`` where ``g_j`` is the j-th component of the gradient (derivative) of the objective function. Values must be in the range `(0.0, inf)`. warm_start : bool, default=False If set to ``True``, reuse the solution of the previous call to ``fit`` as initialization for ``coef_`` and ``intercept_`` . verbose : int, default=0 For the lbfgs solver set verbose to any positive number for verbosity. Values must be in the range `[0, inf)`. Attributes ---------- coef_ : array of shape (n_features,) Estimated coefficients for the linear predictor (`X @ coef_ + intercept_`) in the GLM. intercept_ : float Intercept (a.k.a. bias) added to linear predictor. n_features_in_ : int Number of features seen during :term:`fit`. .. versionadded:: 0.24 feature_names_in_ : ndarray of shape (`n_features_in_`,) Names of features seen during :term:`fit`. Defined only when `X` has feature names that are all strings. .. versionadded:: 1.0 n_iter_ : int Actual number of iterations used in the solver. See Also -------- TweedieRegressor : Generalized Linear Model with a Tweedie distribution. Examples -------- >>> from sklearn import linear_model >>> clf = linear_model.PoissonRegressor() >>> X = [[1, 2], [2, 3], [3, 4], [4, 3]] >>> y = [12, 17, 22, 21] >>> clf.fit(X, y) PoissonRegressor() >>> clf.score(X, y) 0.990... >>> clf.coef_ array([0.121..., 0.158...]) >>> clf.intercept_ 2.088... >>> clf.predict([[1, 1], [3, 4]]) array([10.676..., 21.875...]) rTrrFrrrrrrr ctj||||||ddSNrsuperr%r"rrrrrr rOr#r$r%%  zPoissonRegressor.__init__cCrr!)rrr#r#r$rY8zPoissonRegressor._get_lossrPrrrr%rY __classcell__r#r#rr$rsXrcr) ra Generalized Linear Model with a Gamma distribution. This regressor uses the 'log' link function. Read more in the :ref:`User Guide `. .. versionadded:: 0.23 Parameters ---------- alpha : float, default=1 Constant that multiplies the penalty term and thus determines the regularization strength. ``alpha = 0`` is equivalent to unpenalized GLMs. In this case, the design matrix `X` must have full column rank (no collinearities). Values must be in the range `[0.0, inf)`. fit_intercept : bool, default=True Specifies if a constant (a.k.a. bias or intercept) should be added to the linear predictor (X @ coef + intercept). max_iter : int, default=100 The maximal number of iterations for the solver. Values must be in the range `[1, inf)`. tol : float, default=1e-4 Stopping criterion. For the lbfgs solver, the iteration will stop when ``max{|g_j|, j = 1, ..., d} <= tol`` where ``g_j`` is the j-th component of the gradient (derivative) of the objective function. Values must be in the range `(0.0, inf)`. warm_start : bool, default=False If set to ``True``, reuse the solution of the previous call to ``fit`` as initialization for ``coef_`` and ``intercept_`` . verbose : int, default=0 For the lbfgs solver set verbose to any positive number for verbosity. Values must be in the range `[0, inf)`. Attributes ---------- coef_ : array of shape (n_features,) Estimated coefficients for the linear predictor (`X * coef_ + intercept_`) in the GLM. intercept_ : float Intercept (a.k.a. bias) added to linear predictor. n_features_in_ : int Number of features seen during :term:`fit`. .. versionadded:: 0.24 n_iter_ : int Actual number of iterations used in the solver. feature_names_in_ : ndarray of shape (`n_features_in_`,) Names of features seen during :term:`fit`. Defined only when `X` has feature names that are all strings. .. versionadded:: 1.0 See Also -------- PoissonRegressor : Generalized Linear Model with a Poisson distribution. TweedieRegressor : Generalized Linear Model with a Tweedie distribution. Examples -------- >>> from sklearn import linear_model >>> clf = linear_model.GammaRegressor() >>> X = [[1, 2], [2, 3], [3, 4], [4, 3]] >>> y = [19, 26, 33, 30] >>> clf.fit(X, y) GammaRegressor() >>> clf.score(X, y) 0.773... >>> clf.coef_ array([0.072..., 0.066...]) >>> clf.intercept_ 2.896... >>> clf.predict([[1, 0], [2, 8]]) array([19.483..., 35.795...]) rTrrFrrcrrrrrr#r$r%rzGammaRegressor.__init__cCrr!)rrr#r#r$rYrzGammaRegressor._get_lossrr#r#rr$r<sYrc s<eZdZdZdddddddd d fd d Zd dZZS)ra<Generalized Linear Model with a Tweedie distribution. This estimator can be used to model different GLMs depending on the ``power`` parameter, which determines the underlying distribution. Read more in the :ref:`User Guide `. .. versionadded:: 0.23 Parameters ---------- power : float, default=0 The power determines the underlying target distribution according to the following table: +-------+------------------------+ | Power | Distribution | +=======+========================+ | 0 | Normal | +-------+------------------------+ | 1 | Poisson | +-------+------------------------+ | (1,2) | Compound Poisson Gamma | +-------+------------------------+ | 2 | Gamma | +-------+------------------------+ | 3 | Inverse Gaussian | +-------+------------------------+ For ``0 < power < 1``, no distribution exists. alpha : float, default=1 Constant that multiplies the penalty term and thus determines the regularization strength. ``alpha = 0`` is equivalent to unpenalized GLMs. In this case, the design matrix `X` must have full column rank (no collinearities). Values must be in the range `[0.0, inf)`. fit_intercept : bool, default=True Specifies if a constant (a.k.a. bias or intercept) should be added to the linear predictor (X @ coef + intercept). link : {'auto', 'identity', 'log'}, default='auto' The link function of the GLM, i.e. mapping from linear predictor `X @ coeff + intercept` to prediction `y_pred`. Option 'auto' sets the link depending on the chosen `power` parameter as follows: - 'identity' for ``power <= 0``, e.g. for the Normal distribution - 'log' for ``power > 0``, e.g. for Poisson, Gamma and Inverse Gaussian distributions max_iter : int, default=100 The maximal number of iterations for the solver. Values must be in the range `[1, inf)`. tol : float, default=1e-4 Stopping criterion. For the lbfgs solver, the iteration will stop when ``max{|g_j|, j = 1, ..., d} <= tol`` where ``g_j`` is the j-th component of the gradient (derivative) of the objective function. Values must be in the range `(0.0, inf)`. warm_start : bool, default=False If set to ``True``, reuse the solution of the previous call to ``fit`` as initialization for ``coef_`` and ``intercept_`` . verbose : int, default=0 For the lbfgs solver set verbose to any positive number for verbosity. Values must be in the range `[0, inf)`. Attributes ---------- coef_ : array of shape (n_features,) Estimated coefficients for the linear predictor (`X @ coef_ + intercept_`) in the GLM. intercept_ : float Intercept (a.k.a. bias) added to linear predictor. n_iter_ : int Actual number of iterations used in the solver. n_features_in_ : int Number of features seen during :term:`fit`. .. versionadded:: 0.24 feature_names_in_ : ndarray of shape (`n_features_in_`,) Names of features seen during :term:`fit`. Defined only when `X` has feature names that are all strings. .. versionadded:: 1.0 See Also -------- PoissonRegressor : Generalized Linear Model with a Poisson distribution. GammaRegressor : Generalized Linear Model with a Gamma distribution. Examples -------- >>> from sklearn import linear_model >>> clf = linear_model.TweedieRegressor() >>> X = [[1, 2], [2, 3], [3, 4], [4, 3]] >>> y = [2, 3.5, 5, 5.5] >>> clf.fit(X, y) TweedieRegressor() >>> clf.score(X, y) 0.839... >>> clf.coef_ array([0.599..., 0.299...]) >>> clf.intercept_ 1.600... >>> clf.predict([[1, 1], [3, 4]]) array([2.500..., 4.599...]) r&rTautorrFr)rrrrcrrrr c s(tj||||||d||_||_dSr)rr%rcr) r"rrrrcrrrr rr#r$r%s  zTweedieRegressor.__init__cCsj|jdkr|jdkrt|jdSt|jdS|jdkr!t|jdS|jdkr,t|jdStd|jd)NrrrlogidentityzFThe link must be an element of ['auto', 'identity', 'log']; got (link=))rcrrrrMrr#r#r$rY6s        zTweedieRegressor._get_lossrr#r#rr$rswr)"rrInumpyrSscipy.optimizerf_loss.glm_distributionr _loss.lossrrrrrbaser r utils.optimizer utilsr r rutils.validationrrutils._openmp_helpersr _linear_lossrrrrrr#r#r#r$s&    4mn