ó áp7]c@sMdZddlmZddlZddljZddlm Z ddl m Z de fd„ƒYZ edkrId gZd ekrId Zejejjd ed fƒejed fƒfZejejejd ƒd d ƒejdƒddd…fjZejddddgddddgddddggƒZejddddgddddgddddggƒZejeeƒZedejjd ejƒ7ZejedddgƒZedejjd ejƒZ e e eƒZ!e"e!j#ƒƒe"dƒe!j$ddde&ddƒe"e!j#ƒƒqIndS(sQ Created on Sun Nov 14 08:21:41 2010 Author: josef-pktd License: BSD (3-clause) iÿÿÿÿ(tprint_functionN(tpca(t LeaveOneOuttFactorModelUnivariatecBsMeZdZd„Zdded„Zd„Zdedd„Zd„Z RS(s  Todo: check treatment of const, make it optional ? add hasconst (0 or 1), needed when selecting nfact+hasconst options are arguments in calc_factors, should be more public instead cross-validation is slow for large number of observations cCs(tj|ƒ|_tj|ƒ|_dS(N(tnptasarraytendogtexog(tselfRR((sHlib/python2.7/site-packages/statsmodels/sandbox/datarich/factormodels.pyt__init__sicCs¦|dkr|j}ntj|ƒ}t|d|ddƒ\}}}}||_|r~tj|dtƒ|_ d|_ n||_ d|_ ||_ ||_ dS(sÜget factor decomposition of exogenous variables This uses principal component analysis to obtain the factors. The number of factors kept is the maximum that will be considered in the regression. tkeepdimt normalizeitprependiN( tNoneRRRRt exog_reducedtsmt add_constanttTruetfactorsthasconsttevalstevecs(RtxR taddconsttxredtfactRR((sHlib/python2.7/site-packages/statsmodels/sandbox/datarich/factormodels.pyt calc_factors"s  $     cCsRt|dƒs|jƒntj|j|jdd…d|d…fƒjƒS(Ntfactors_wconsti(thasattrRRtOLSRRtfit(Rtnfact((sHlib/python2.7/site-packages/statsmodels/sandbox/datarich/factormodels.pytfit_fixed_nfact9s c CsDt|dƒs|jƒn|j}|dkrH|jjd|}n||dkrktddƒ‚nt|dƒ}|j}g}x+t d||ƒD]}|jdd…d|…f}t j ||ƒj ƒ} |s…|dkrt t|ƒƒ}nd} x„|D]p\} } t j || || dd…fƒj ƒ} | || | jj| j|| dd…fƒd7} qWn tj} |j|| j| j| j| gƒqWtj|ƒ|_}tjtj|dd…dd …fd ƒtj|dd…d fd ƒtj|dd…d fd ƒf|_dS( sWestimate the model and selection criteria for up to maxfact factors The selection criteria that are calculated are AIC, BIC, and R2_adj. and additionally cross-validation prediction error sum of squares if `skip_crossval` is false. Cross-validation is not used by default because it can be time consuming to calculate. By default the cross-validation method is Leave-one-out on the full dataset. A different cross-validation sample can be specified as an argument to cv_iter. Results are attached in `results_find_nfact` Ris9nothing to do, number of factors (incl. constant) should s be at least 1i Ngg@iiiÿÿÿÿ(RRRR Rtshapet ValueErrortminRtrangeRRRRtlentmodeltpredicttparamsRtnantappendtaictbict rsquared_adjtarraytresults_find_nfacttr_targmintargmaxt best_nfact(Rtmaxfactt skip_crossvaltcv_iterRty0tresultstkRtrestprederr2tinidxtoutidxtres_l1o((sHlib/python2.7/site-packages/statsmodels/sandbox/datarich/factormodels.pytfit_find_nfact>s6       , 5 )JcCst|dƒs|jƒn|j}d}|dd7}|dddt|jƒ7}dd lm}d jd ƒ}d gd gd}td|ƒ}|||dd|ƒ}|dd7}|dd7}|d|j ƒ7}|dd7}|dd7}|dd7}|dd7}|S(sÇprovides a summary for the selection of the number of factors Returns ------- sumstr : string summary of the results for selecting the number of factors R/ts s+Best result for k, by AIC, BIC, R2_adj, L1Ot is%5d %4d %6d %5diÿÿÿÿ(t SimpleTablesk, AIC, BIC, R2_adj, L1Os, s%6ds%10.3fit data_fmtsttxt_fmts!PCA regression on simulated data,s*DGP: 2 factors and 4 explanatory variabless(Notes: k is number of components of PCA,s% constant is added additionallys, k=0 means regression on constant onlys> L1O: sum of squared prediction errors for leave-one-outs N( RR?R/ttupleR3tstatsmodels.iolib.tableRBtsplittdictR t__str__(RR8tsumstrRBtheaderst numformatttxt_fmt1ttabl((sHlib/python2.7/site-packages/statsmodels/sandbox/datarich/factormodels.pytsummary_find_nfact€s&   N( t__name__t __module__t__doc__R R RRR R?RO(((sHlib/python2.7/site-packages/statsmodels/sandbox/datarich/factormodels.pyRs   Bt__main__iiôtsizeiiigð?gg@g@gš™™™™™¹?gø?swith cross validation - slowerR4R5R6('RRt __future__RtnumpyRtstatsmodels.apitapiRtstatsmodels.sandbox.toolsRt#statsmodels.sandbox.tools.cross_valRtobjectRRPtexamplestnobstc_trandomtnormaltonestf0trepeatteyetarangetTtf2xcoefR.tdottx0R!tytrueR7tmodtprintROR?R tFalse(((sHlib/python2.7/site-packages/statsmodels/sandbox/datarich/factormodels.pyts6 “   7D