M/PhddlZGddZGddeZGddeZGdd eZGd d eZdS) NceZdZdZdZdS)RegressionEffectsat Base class for regression effects used in RegressionFDR. Any implementation of the class must provide a method called 'stats' that takes a RegressionFDR object and returns effect sizes for the model coefficients. Greater values for these statistics imply greater evidence that the effect is real. Knockoff effect sizes are based on fitting the regression model to an extended design matrix [X X'], where X' is a design matrix with the same shape as the actual design matrix X. The construction of X' guarantees that there are no true associations between the columns of X' and the dependent variable of the regression. If X has p columns, then the effect size of covariate j is based on the strength of the estimated association for coefficient j compared to the strength of the estimated association for coefficient p+j. ctN)NotImplementedError)selfparents e/var/www/html/test/jupyter/venv/lib/python3.11/site-packages/statsmodels/stats/knockoff_regeffects.pystatszRegressionEffects.statss!!N__name__ __module__ __qualname____doc__r r r rrs-$"""""r rceZdZdZdZdS)CorrelationEffectsa Marginal correlation effect sizes for FDR control. Parameters ---------- parent : RegressionFDR The RegressionFDR instance to which this effect size is applied. Notes ----- This class implements the marginal correlation approach to constructing test statistics for a knockoff analysis, as described under (1) in section 2.2 of the Barber and Candes paper. ctj|jj|j}tj|jj|j}tj|tj|z Sr)npdotexog1Tendogexog2abs)rr s1s2s r r zCorrelationEffects.stats-sK VFLNFL 1 1 VFLNFL 1 1vbzzBF2JJ&&r Nr rr r rrs-"'''''r rceZdZdZdZdZdS)ForwardEffectsa@ Forward selection effect sizes for FDR control. Parameters ---------- parent : RegressionFDR The RegressionFDR instance to which this effect size is applied. pursuit : bool If True, 'basis pursuit' is used, which amounts to performing a full regression at each selection step to adjust the working residual vector. If False (the default), the residual is adjusted by regressing out each selected variable marginally. Setting pursuit=True will be considerably slower, but may give better results when exog is not orthogonal. Notes ----- This class implements the forward selection approach to constructing test statistics for a knockoff analysis, as described under (5) in section 2.2 of the Barber and Candes paper. c||_dSr)pursuit)rr"s r __init__zForwardEffects.__init__Ls  r cjjd}jfdt |D}t j|}g}t |D]}t jfd|D}t j|}||z dz |||d<||d} ||=|j r5|D]} | t j | | | zz} | | t j | | zz|d|dz} ||dzd} t j | | k| | t j | | z z} | S)Nc6g|]}|jdd|ffSr)exog).0ir s r z(ForwardEffects.stats..Rs- : : :q&+aaad#$ : : :r c jg|]/}tjtj|d0S)r%)rrr)r(xrvs r r*z(ForwardEffects.stats..Vs3===Qrvb!A$//00===r r)r'shapercopyrangeremptyr_argmaxr"rappendwheresign)rr nvarvlzpastr)dpjr,vz1z2str-s ` @r r zForwardEffects.statsOsu{ # \    : : : :eDkk : : : HTNNt $ $A===="===>B " A(Q,AbeAhK1aA1| ((A1a'AA A "&Q--!# #BB qqy\ tQwxx[ Xb2gr2 & &b)9)9 9 r Nrrrrr#r rr r r r 3s<0r r ceZdZdZdZdS) OLSEffectsa OLS regression for knockoff analysis. Parameters ---------- parent : RegressionFDR The RegressionFDR instance to which this effect size is applied. Notes ----- This class implements the ordinary least squares regression approach to constructing test statistics for a knockoff analysis, as described under (2) in section 2.2 of the Barber and Candes paper. cddlm}||j|j}|}t |jdz}tj|jd|tj|j|dz }|S)Nr)OLSr.) #statsmodels.regression.linear_modelrFrr'fitlenparamsrr)rr rFmodelresultqr s r r zOLSEffects.statsxs;;;;;;FL&+..   ! #v}QqS)**RVFM!""4E-F-FF r Nr rr r rDrDfs-"r rDc$eZdZdZ ddZdZdS)RegModelEffectsa Use any regression model for Regression FDR analysis. Parameters ---------- parent : RegressionFDR The RegressionFDR instance to which this effect size is applied. model_cls : class Any model with appropriate fit or fit_regularized functions regularized : bool If True, use fit_regularized to fit the model model_kws : dict Keywords passed to model initializer fit_kws : dict Dictionary of keyword arguments for fit or fit_regularized FNcN||_||_||ni|_||ni|_dSr) model_cls regularized model_kwsfit_kws)rrQrRrSrTs r r#zRegModelEffects.__init__s5"&&/&;")"5ww2 r cD|j|j|jfi|j}|jr|jdi|jj}n|jdi|jj}t|dz}tj |d|tj ||dz }|S)Nr.rr) rQrr'rSrRfit_regularizedrTrJrHrIrr)rr rKrJrMr s r r zRegModelEffects.statssv|V[KKDNKK   6*U*::T\::AFFUY....5F KK1 vac{##bfVABBZ&8&88 r )FNNrBrr r rOrOsK&@D>>>>r rO)numpyrrrr rDrOrr r rXs"""""""".'''''*'''000000&000f"8#####'#####r