M/Ph&bdZddlZddlmZddlmZGddeZGddeZdS) z_ Penalty classes for Generalized Additive Models Author: Luca Puggini Author: Josef Perktold N) block_diag)Penaltyc:eZdZdZd dZd dZd dZd dZd dZdS) UnivariateGamPenaltya  Penalty for smooth term in Generalized Additive Models Parameters ---------- univariate_smoother : instance instance of univariate smoother or spline class alpha : float default penalty weight, alpha can be provided to each method weights: TODO: not used and verified, might be removed Attributes ---------- Parameters are stored, additionally nob s: The number of samples used during the estimation n_columns : number of columns in smoother basis ct||_||_||_|jj|_|jj|_dSN)weightsalphaunivariate_smoothernobs dim_basis n_columns)selfr r r s ]/var/www/html/test/jupyter/venv/lib/python3.11/site-packages/statsmodels/gam/gam_penalties.py__init__zUnivariateGamPenalty.__init__"s6  #6 ,1 1;Nc||j}||jj|}||z|jz S)aRevaluate penalization at params Parameters ---------- params : ndarray coefficients for the spline basis in the regression model alpha : float default penalty weight Returns ------- func : float value of the penalty evaluated at params )r dotr cov_der2r )rparamsr fs rfunczUnivariateGamPenalty.func)sF =JE JJt/8<s@ =JE It7@&II I TYrcP||j}d|z|jjz}||jz}|S)aevaluate second derivative of penalty with respect to params Parameters ---------- params : ndarray coefficients for the spline basis in the regression model alpha : float default penalty weight Returns ------- deriv2 : ndarray, 2-Dim second derivative, hessian of the penalty with respect to params Nr)r r rr )rrr d2s rderiv2zUnivariateGamPenalty.deriv2Ts4 =JE Y1: : di rc2||j}||jjzS)apenalty matrix for the smooth term of a GAM Parameters ---------- alpha : list of floats or None penalty weights Returns ------- penalty matrix square penalty matrix for quadratic penalization. The number of rows and columns are equal to the number of columns in the smooth terms, i.e. the number of parameters for this smooth term in the regression model )r r r)rr s rpenalty_matrixz#UnivariateGamPenalty.penalty_matrixjs! =JEt/888r)rrr __name__ __module__ __qualname____doc__rrrr!r#rrrrs&<<<<%%%%*,,999999rrc>eZdZdZ d dZd dZd dZd dZd dZdS) MultivariateGamPenaltya^ Penalty for Generalized Additive Models Parameters ---------- multivariate_smoother : instance instance of additive smoother or spline class alpha : list of float default penalty weight, list with length equal to the number of smooth terms. ``alpha`` can also be provided to each method. weights : array_like currently not used is a list of doubles of the same length as alpha or a list of ndarrays where each component has the length equal to the number of columns in that component start_idx : int number of parameters that come before the smooth terms. If the model has a linear component, then the parameters for the smooth components start at ``start_index``. Attributes ---------- Parameters are stored, additionally nob s: The number of samples used during the estimation dim_basis : number of columns of additive smoother. Number of columns in all smoothers. k_variables : number of smooth terms k_params : total number of parameters in the regression model Nrct|jt|kr5dt|jt|fz}t||_jj_jj_jj_|_|_|jz_ |$dtjD_ n ddl }| d|_ fdtjD_|}tjjD])\}} dj|||| jz<|| jz }*g_tjD]T}t#jj|j |j|} j| UdS)NzTall the input values should be of the same length. len(smoothers)=%d, len(alphas)=%dcg|]}dS)g?r)).0_s r z3MultivariateGamPenalty.__init__..s@@@1B@@@rrzweights is currently ignoredcPg|]"}tjjt#S))dtype)rzerosk_paramsbool)r.r/rs rr0z3MultivariateGamPenalty.__init__..s:777Xdm4888777rT)r r )len smoothers ValueErrormultivariate_smootherr k_variablesr r start_idxr4ranger warningswarnmask enumerategprappend) rr9r r r;msgr= param_countismootherrAs ` rrzMultivariateGamPenalty.__init__s $. / /3u:: = =8,677U(?(?@@@DLL OOO MM8 9 9 9"DL7777#D$455777  $T%?%IJJ . .KAxKODIaLkH4F&FF G 8- -KKt'((  A%d&@&J1&M.2l1o,0JqM;;;B GNN2       rc| dg|jz}d}t|jD]@}||j|}||j||||z }A|S)aHevaluate penalization at params Parameters ---------- params : ndarray coefficients in the regression model alpha : float or list of floats penalty weights Returns ------- func : float value of the penalty evaluated at params Nrr )r:r<r?rAr)rrr costrEparams_is rrzMultivariateGamPenalty.funcsr =FT--Et'(( > >Adil+H DGAJOOHE!HO== =DD rcB| dg|jz}tj|jg}t |jD]P}||j|}||j||||Qtj |S)arevaluate derivative of penalty with respect to params Parameters ---------- params : ndarray coefficients in the regression model alpha : list of floats or None penalty weights Returns ------- deriv : ndarray derivative, gradient of the penalty with respect to params NrH) r:rr3r;r<r?rBrAr concatenate)rrr gradrErJs rrzMultivariateGamPenalty.derivs =FT--E(()t'(( D DAdil+H KK ((q(BB C C C C~d###rc:| dg|jz}tj|j|jfg}t |jD]P}||j|}||j||||Qt|S)aevaluate second derivative of penalty with respect to params Parameters ---------- params : ndarray coefficients in the regression model alpha : list of floats or None penalty weights Returns ------- deriv2 : ndarray, 2-Dim second derivative, hessian of the penalty with respect to params NrH) r:rr3r;r<r?rBrAr!r)rrr r!rErJs rr!zMultivariateGamPenalty.deriv2s =FT--E(DNDN;<<=t'(( G GAdil+H MM$'!*++HE!H+EE F F F F6""rc ||j}tj|j|jfg}t |jD]<}||j|||=t|S)apenalty matrix for generalized additive model Parameters ---------- alpha : list of floats or None penalty weights Returns ------- penalty matrix block diagonal, square penalty matrix for quadratic penalization. The number of rows and columns are equal to the number of parameters in the regression model ``k_params``. Notes ----- statsmodels does not support backwards compatibility when keywords are used as positional arguments. The order of keywords might change. We might need to add a ``params`` keyword if the need arises. NrH) r rr3r;r<r:rBrAr#r)rr s_allrEs rr#z%MultivariateGamPenalty.penalty_matrixs* =JE4>4>:;;<t'(( D DA LL22q2BB C C C C5!!r)Nrr r$r)rrr+r+s>>B))))V2$$$$2####2""""""rr+) r(numpyr scipy.linalgrstatsmodels.base._penaltiesrrr+r)rrrTs######//////o9o9o9o9o97o9o9o9dr"r"r"r"r"Wr"r"r"r"r"r