'YHh;ZdZddlZddlmZddlmZddlmZddl m Z ddl m Z m Z mZdZd Zd Zd Zd Zd ZGddeZedk(r`dZgdZgdZej4j7deeeedZeej;zZeeZd\e_e_ e!ee_ejEgdZ#e$deee$e#jJddl&m'Z'e$e'edeee#jJdd\Z(Z)e jTZ+ejYe#jJddee#jJddzzZ-e$eee-eee-Z.e$e.j_e.j`e$eee-e$eee-yy)zMultivariate Normal Model with full covariance matrix toeplitz structure is not exploited, need cholesky or inv for toeplitz Author: josef-pktd N)linalg)toeplitz)GenericLikelihoodModel)sunspots) ArmaProcess arma_acovfarma_generate_samplect|}|dz }|dzj}tj| |z}|dtjtj|z z|zz}tj |rG|j dk(r8|dtjtjj|zz}|S)zxloglike multivariate normal copied from GLS and adjusted names not sure why this differes from mvn_loglike @?) lensumnplogpianyndimrdet)xsigmanobsnobs2SSRllfs Z/var/www/html/planif/env/lib/python3.12/site-packages/statsmodels/miscmodels/try_mlecov.pymvn_loglike_sumrs q6D 3JE a4**,C 66#;, CAbffRUU5[! !5 ((C vve}q r"&&u-... Jcltj|}tjtjj |}t |}tj |tj || }||tjdtjzzz}||z}|dz}|S)loglike multivariate normal assumes x is 1d, (nobs,) and sigma is 2d (nobs, nobs) brute force from formula no checking of correct inputs use of inv and log-det should be replace with something more efficient r r)rinvrrrrdotr)rrsigmainv logdetsigmarrs r mvn_logliker&%szz% H&&u-.K q6D FF1bffXq) * *C4"&&RUU# ##C;C3JC Jrc tjj|}tjj|j}tj ||}tj tjj|}t|}ddl m }tdttj |jj|jtj |j| }||tj dtjzzz}||z}|dz}||dtjtj tj |zfS)r!r)statsz scipy.statsr r)rrr"choleskyTr#rrrscipyr(printnormpdfrrdiagonal) rrr$ cholsigmainv x_whitenedr%rr(rs rmvn_loglike_cholr2:s yy}}U#H99%%h/11L a(J&&u-.K q6D - "&& + , 0 0 23 FF:<< , ,C4"&&RUU# ##C;C3JC  Qr{{t|}|jdS)N)maF)retnew)r invertroots)r9procs rinvertiblerootsr=vs " D   5  ))rctjdg|d|j f}tjdg||j df}ddlm}|j ||j |fS)Nr r)rr_narnmanumpy.polynomial polynomial Polynomial)selfparamsarr9polys rgetpolyrI{sg sVITXX&&& 'B sFDHH9:&& 'B# ??2  3 33rc"eZdZdZdZdZdZy)MLEGLSaARMA model with exact loglikelhood for short time series Inverts (nobs, nobs) matrix, use only for nobs <= 200 or so. This class is a pattern for small sample GLS-like models. Intended use for loglikelihood of initial observations for ARMA. TODO: This might be missing the error variance. Does it assume error is distributed N(0,1) Maybe extend to mean handling, or assume it is already removed. ctjdg|d|j f}tjdg||j df}t |||}|d|}t |}|S)zget autocovariance matrix from ARMA regression parameter ar parameters are assumed to have rhs parameterization r N)r)rr?r@rArr)rErFrrGr9autocovrs r _params2covzMLEGLS._params2covsu UUA3 *** + UUA3yz** +R$/%4.! rc|j|dd|j}||ddzz}t|j|}|S)Nr )rNrr&endog)rErFsiglogliks rloglikezMLEGLS.loglikesEvcr{DII6F2JM!TZZ- rc||j|i|}tjdg|j|j|j|j zf}t |\}}|sU|jj}|dd||j|j|j z|j|}|S)Nr  start_params)fitrr?rFr@rAr=copy)rEargskwdsresr9mainv wasinvertiblerWs rfit_invertiblezMLEGLS.fit_invertiblesdhh%% UUA3 488TXXdhh->?? @.r2}::??,L8=ab L488DHH#4 5(( (5C rN)__name__ __module__ __qualname____doc__rNrTr_rrrKrKs  & rrK__main__2)r4皙皙?)r4rh皙?iMr )r r )rhrgrirhr4rVDGP) yule_walkerrP)1rcnumpyrr+r scipy.linalgrstatsmodels.base.modelrstatsmodels.datasetsrstatsmodels.tsa.arima_processrrr rr&r2r7r=rIrKr`rrGr9randomseedymeanmodr@rArrXr\r,rFstatsmodels.regressionrkarpolymapolyloaddatarNrllorshaperdrrr}s!9) *K8@* 4 3 #3n z D B BIINN7R4*AMA )CCGSW1vCH ''8' 9C %R #**2 +a S#**Sb/2NFF 8==?D OOCJJsOT 2SZZ^Q5F FE +a  1e $C #'')SYY 1e $% /!U #$Or