M/Ph4rddlZddlZddlmZGddZGddZGddeZdS) N)LikelihoodModelResultsc2eZdZdZd dZdZdZdZdZdS) BayesGaussMIa Bayesian Imputation using a Gaussian model. The approach is Bayesian. The goal is to sample from the joint distribution of the mean vector, covariance matrix, and missing data values given the observed data values. Conjugate priors for the population mean and covariance matrix are used. Gibbs sampling is used to update the mean vector, covariance matrix, and missing data values in turn. After burn-in, the imputed complete data sets from the Gibbs chain can be used in multiple imputation analyses (MI). Parameters ---------- data : ndarray The array of data to be imputed. Values in the array equal to NaN are imputed. mean_prior : ndarray, optional The covariance matrix of the Gaussian prior distribution for the mean vector. If not provided, the identity matrix is used. cov_prior : ndarray, optional The center matrix for the inverse Wishart prior distribution for the covariance matrix. If not provided, the identity matrix is used. cov_prior_df : positive float The degrees of freedom of the inverse Wishart prior distribution for the covariance matrix. Defaults to 1. Examples -------- A basic example with OLS. Data is generated assuming 10% is missing at random. >>> import numpy as np >>> x = np.random.standard_normal((1000, 2)) >>> x.flat[np.random.sample(2000) < 0.1] = np.nan The imputer is used with ``MI``. >>> import statsmodels.api as sm >>> def model_args_fn(x): ... # Return endog, exog from x ... return x[:, 0], x[:, 1:] >>> imp = sm.BayesGaussMI(x) >>> mi = sm.MI(imp, sm.OLS, model_args_fn) Ncd|_t|tjur |j|_t j|d}||_||_t j ||_ |j j d|_ |j j d|_ dt jdt j|j j dzz}t j|j |}i}t#|D]0\}} | dkr | |vrg|| <|| |1d|D|_|jj d} t j| |_g} t/| D]{}|jdd|f} | t j| } t3| dkrd|z} t5| | | |t j| |_|t j| }||_|t j| }||_||_dS)NW) requirementsrrc6g|]}tj|S)npasarray).0vs _/var/www/html/test/jupyter/venv/lib/python3.11/site-packages/statsmodels/imputation/bayes_mi.py z)BayesGaussMI.__init__..Os @@@1A@@@z Column %d has no observed values) exog_namestypepd DataFramecolumnsr requiredata_dataisnanmaskshapenobsnvarlogarangedot enumerateappendvaluespatternseyecovrangeisfinitelen ValueErrormeanr mean_prior cov_prior cov_prior_df) selfrr.r/r0zcrowmapirpr-msgs r__init__zBayesGaussMI.__init__7s& :: % %"lDOz$S111  HTNN IOA& IOA&  q29TY_Q%788899 9 F49a aLL  DAqAvvq 1I  Q    @@ @@@  J Q 6!99q " "A 111a4 A"+a..!A1vv{{81< oo% KK ! ! ! !Jt$$   J$  q I")rc~|||dS)z2 Cycle through all Gibbs updates. N) update_data update_mean update_cov)r1s rupdatezBayesGaussMI.updateks@   rc `|jD]}|d}tj|j|ddf}tj|j|ddf}|j|}|j|}|j|ddfdd|f}|j|ddfdd|f}|j|ddfdd|f} |j|ddfdd|f|z } |tj| tj || j j z} |tj| tj || j z } tj | } tj t|t|f}| tj|| j z|jtj||<|j(t#j|j|jd|_dS|j|_dS)z: Gibbs update of the missing data values. rNsizeF)rcopy)r&r flatnonzerorr-r(rr"linalgsolveTcholeskyrandomnormalr+ix_rrrr)r1ixr5ix_missix_obsmmmovoovmmvmorcmcvcsus rr:zBayesGaussMI.update_datavs - C CB1AnTYq!!!t_55G^TYq!!!t_$455F7#B6"B(61119%aaai0C(7AAA:&qqq'z2C(7AAA:&qqq&y1C 2qqq5!!!!V),r1AbfS")//#qs";";<<>>Brvc29??3#>#>???B##B''B   s2wwG &= >>A.026!RT??.BDJrvb'** + + ? & :#'? %'''DIII  DIIIrc "tj|j|jz |jz|j|jz }tj|j|}tj|j|jd}tj||}tj |}|tj|tj dd|j z|_ dS)zo Gibbs update of the mean vector. Do not call until update_data has been called once. rrN)r rCrDr(rr.r"rsumrFrGrHrr-)r1rSvmrRs rr;zBayesGaussMI.update_meansY__TXdi/$/A!_ty8:: VDHb ! !Y__TXtz~~a'8'8 9 9 VB^^ I  r " "29#3#3Aq$)#D#DEEE rc:|j|jz }tj|j|}||jz}t tj|j|j z}tj tj |}tjtj ||jf|j}tj|j|}tj ||_dS)zu Gibbs update of the covariance matrix. Do not call until update_data has been called once. r?N)rr-r r"rEr/intceilrr0rCrFinvrGrHrr()r1rRgradfxmas rr<zBayesGaussMI.update_covs J " VAC^^   T%6677 8 8 I  ry}}Q// 0 0 F29##"di#9913 ? ? VAC^^9==$$r)NNr) __name__ __module__ __qualname____doc__r8r=r:r;r<r rrrrst..`2)2)2)2)h   !#!#!#FFFF*%%%%%rrc.eZdZdZ d dZd dZdZdS) MIa{ MI performs multiple imputation using a provided imputer object. Parameters ---------- imp : object An imputer class, such as BayesGaussMI. model : model class Any statsmodels model class. model_args_fn : function A function taking an imputed dataset as input and returning endog, exog. If the model is fit using a formula, returns a DataFrame used to build the model. Optional when a formula is used. model_kwds_fn : function, optional A function taking an imputed dataset as input and returning a dictionary of model keyword arguments. formula : str, optional If provided, the model is constructed using the `from_formula` class method, otherwise the `__init__` method is used. fit_args : list-like, optional List of arguments to be passed to the fit method fit_kwds : dict-like, optional Keyword arguments to be passed to the fit method xfunc : function mapping ndarray to ndarray A function that is applied to the complete data matrix prior to fitting the model burn : int Number of burn-in iterations nrep : int Number of imputed data sets to use in the analysis skip : int Number of Gibbs iterations to skip between successive multiple imputation fits. Notes ----- The imputer object must have an 'update' method, and a 'data' attribute that contains the current imputed dataset. xfunc can be used to introduce domain constraints, e.g. when imputing binary data the imputed continuous values can be rounded to 0/1. Nd c $||_| |_||_||_|d} | }||_|d} | }||_|d} | }||_|d} | }||_||_| |_ | |_t| D]} | dS)NcgSNr ras rfzMI.__init__..f rciSrnr ros rrpzMI.__init__..frqrcgSrnr ros rrpzMI.__init__..f rqrciSrnr ros rrpzMI.__init__..frqr) impskipmodelformula model_args_fn model_kwds_fnfit_argsfit_kwdsxfuncnrepr)r=)r1rurwryrzrxr{r|r}burnr~rvrpks rr8z MI.__init__s        M*     M*     H      H     t  A JJLLLL  rcgg}}g}t|jD]}t|jdzD]}|j|jj}|j||}|j4|j| |i| |}nA|jj |jg| |Ri| |}|j | |i||}|||||t!j|j|t!j||||\} } } t-||| | } | | _|| _| S)a Impute datasets, fit models, and pool results. Parameters ---------- results_cb : function, optional If provided, each results instance r is passed through `results_cb`, then appended to the `results` attribute of the MIResults object. To save complete results, use `results_cb=lambda x: x`. The default behavior is to save no results. Returns ------- A MIResults object. r)r)r~rvrur=rr}rxrwryrz from_formulafitr{r|r$r r paramsrA cov_params_combine MIResultsfmiresults) r1 results_cbparr( all_resultsrdarwresultrrrrRs rrzMI.fits"rS ty!! ? ?A49Q;'' " "!!!!Bz%ZZ^^|#" D$6$6r$:$:=%)%7%7%;%;==0 /,4)-););B)?)?444 ..r2244UY b 1 1GT]]25F5FGGF%""::f#5#5666 JJrz&-"4"4"6"677 8 8 8 JJrz&"3"3"5"5":":"<"<== > > > >"&--S"9"9 C dE6: 6 6 rctj|}|jd}|d}t |t |z }tj|j}tj|}|ddt|z z|zz}ddt|z ztj |ztj |z }|||fS)Nrr) r r rr-rXr+r(rE atleast_2dfloatdiag) r1rr(mrwcovbcovcovprs rrz MI._combineUsjoo IaL!3xx#c(("vce}}}T""q1U1XX:~t++1U1XX:~.>tS  r) NNNNNNrirjrkrn)rcrdrerfr8rrr rrrhrhse++ZFJCG)+----^5555n!!!!!rrhc*eZdZdZfdZddZxZS)ra; A results class for multiple imputation (MI). Parameters ---------- mi : MI instance The MI object that produced the results model : instance of statsmodels model class This can be any instance from the multiple imputation runs. It is used to get class information, the specific parameter and data values are not used. params : array_like The overall multiple imputation parameter estimates. normalized_cov_params : array_like (2d) The overall variance covariance matrix of the estimates. cjt|||||_||_dSrn)superr8mi_model)r1rrwrnormalized_cov_params __class__s rr8zMIResults.__init__s2 (=>>> rN皙?cddlm}|}d}i}d|d<|jjj|d<|jj|d<d|jjj j dz|d <d|jj z|d <| |d | | || }|j|d<|||||||S)a Summarize the results of running multiple imputation. Parameters ---------- title : str, optional Title for the top table. If not None, then this replaces the default title alpha : float Significance level for the confidence intervals Returns ------- smry : Summary instance This holds the summary tables and text, which can be printed or converted to various output formats. r)summary2z%8.3frhzMethod:zModel:zDependent variable:z%dz Sample size:zNum. imputationsl)align float_format)alphaFMI)r)titler)statsmodels.iolibrSummaryrrwrcr endog_namesrurrr~add_dictsummary_paramsradd_df add_title)r1rrrsmryrinfoparams rsummaryzMIResults.summarys& /.....!! Y/X&*k&= "##dgk&6&rs999999x%x%x%x%x%x%x%x%vl!l!l!l!l!l!l!l!^?????&?????r