p7]c@ sndZddlmZddddddd gZdd lmZdd lmZdd lZ dd l m Z ddl m Z mZmZddlmZmZddlmZdd ljjZddlmZddlmZddlmZmZddlm Z dd l!j"j#Z"ddl$m%Z%dd l&j'j(Z)dd l*j"j+Z,ddl-m.Z.ddl/m0Z0ydd l1Z1e2Z3Wne4k re5Z3nXdd l6Z6e j7e8j9Z:dZ;dZ<dZ=dZ>dZ?dZ@dZAde"jBfd YZCd!eCfd"YZDd#eDfd$YZEd%eCfd&YZFd'eCfd(YZGdeFfd)YZHdeFfd*YZIdeDfd+YZJdeDfd,YZKdeEfd-YZLdeFfd.YZMd eFfd/YZNd0e"jOfd1YZPd2ePfd3YZQd4eQfd5YZRd6eRfd7YZSd8ePfd9YZTd:eQfd;YZUd<eTeUfd=YZVd>eTeRfd?YZWd@eTeSfdAYZXdBePfdCYZYdDePfdEYZZdFeZfdGYZ[dHeZfdIYZ\dJeZfdKYZ]dLePfdMYZ^dNe^fdOYZ_dPe)j`fdQYZae,jbeaeYdRe)j`fdSYZce,jbeceQdTe)j`fdUYZde,jbedeRdVe)j`fdWYZee,jbeeeSdXe)j`fdYYZfe,jbefeUdZe)j`fd[YZgd\e)j`fd]YZhe,jbeheVd^e)j`fd_YZie,jbeieWd`e)j`fdaYZje,jbejeXdbe)j`fdcYZke,jbekeZdde)j`fdeYZle,jbele]dfe)j`fdgYZme,jbeme^dhe)j`fdiYZne,jbene_d S(js Limited dependent variable and qualitative variables. Includes binary outcomes, count data, (ordered) ordinal data and limited dependent variables. General References -------------------- A.C. Cameron and P.K. Trivedi. `Regression Analysis of Count Data`. Cambridge, 1998 G.S. Madalla. `Limited-Dependent and Qualitative Variables in Econometrics`. Cambridge, 1983. W. Greene. `Econometric Analysis`. Prentice Hall, 5th. edition. 2003. i(tdivisiontPoissontLogittProbittMNLogittNegativeBinomialtGeneralizedPoissontNegativeBinomialP(trange(tloggammaN(t get_dummies(tgammalntdigammat polygamma(tstatstspecial(tnbinom(tdata(tcache_readonly(tPerfectSeparationErrortSpecificationWarning(tapprox_fprime_cs(t handle_data(t fit_l1_slsqp(t genpoisson_pt s %(one_line_description)s Parameters ---------- model : A DiscreteModel instance params : array-like The parameters of a fitted model. hessian : array-like The hessian of the fitted model. scale : float A scale parameter for the covariance matrix. Attributes ---------- df_resid : float See model definition. df_model : float See model definition. llf : float Value of the loglikelihood %(extra_attr)ss  nnz_params : Integer The number of nonzero parameters in the model. Train with trim_params == True or else numerical error will distort this. trimmed : Boolean array trimmed[i] == True if the ith parameter was trimmed from the model.s Compute one-step moment estimator for null (constant-only) model This is a preliminary estimator used as start_params. Returns ------- params : ndarray parameter estimate based one one-step moment matching cC s|jjdkr6tj|dtdt\}}nL|jdkra|}t|jd}n!tj|dtdt\}}||fS(NtStOtdropt dictnamesii(RR(tdtypetkindttoolst categoricaltTruetndimRtshape(tendogt endog_dummiestynames((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyt_numpy_to_dummieslscC s|jdkr`|jddkrQ|jd}t|jdddf}qud}|}n|j}t|}|jj}|||fS(Niiity(R#R$tcolumnsR tiloctnamettolist(R%tynameR&R'((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyt_pandas_to_dummiesys "   cC s+|dkr'tdjd|ndS(s  As of 0.10.0, the supported values for `method` in `fit_regularized` are "l1" and "l1_cvxopt_cp". If an invalid value is passed, raise with a helpful error message Parameters ---------- method : str Raises ------ ValueError tl1t l1_cvxopt_cpsG`method` = {method} is not supported, use either "l1" or "l1_cvxopt_cp"tmethodN(R0R1(t ValueErrortformat(R2((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyt_validate_l1_methods  t DiscreteModelc B seZdZdZdZdZdZdZddddddd Z e je j j j7_dd d de dd d ddde d ZdZde dZddddZdZRS(s Abstract class for discrete choice models. This class does not do anything itself but lays out the methods and call signature expected of child classes in addition to those of statsmodels.model.LikelihoodModel. cK s)tt|j|||t|_dS(N(tsuperR6t__init__R"traise_on_perfect_prediction(tselfR%texogtkwargs((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyR8scC sItjj|j}t|d|_t|jjd||_dS(s Initialize is called by statsmodels.model.LikelihoodModel.__init__ and should contain any preprocessing that needs to be done for a model. iiN(tnptlinalgt matrix_rankR;tfloattdf_modelR$tdf_resid(R:trank((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyt initializescC s tdS(sD The cumulative distribution function of the model. N(tNotImplementedError(R:tX((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pytcdfscC s tdS(sG The probability density (mass) function of the model. N(RE(R:RF((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pytpdfscG sm|j}|jtj|j||jjd }|jritj||drid}t|ndS(Niis2Perfect separation detected, results not available( R%RGR=tdotR;R$R9tallcloseR(R:tparamstargsR%t fittedvaluestmsg((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyt_check_perfect_preds  , tnewtoni#ic K sX|dkr|j}ntt|jd|d|d|d|d|d||}|S(s Fit the model using maximum likelihood. The rest of the docstring is from statsmodels.base.model.LikelihoodModel.fit t start_paramsR2tmaxitert full_outputtdisptcallbackN(tNoneROR7R6tfit( R:RQR2RRRSRTRUR<tmlefit((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyRWs   R0tdefined_by_methoditautog{Gz?g-C6?gQ?c K st||j}tj|}|jdks:ty|| dtinvtzerostnantonesR$( R:tlikelihood_modeltxopttretvalstHRptnz_idxt nnz_paramst H_restrictedtH_restricted_invt cov_params((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyRess      cC s tdS(sQ Predict response variable of a model given exogenous variables. N(RE(R:RKR;tlinear((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pytpredictscC s tdS(sQ This should implement the derivative of the non-linear function N(RE(R:RKR;t dummy_idxt count_idx((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyt_derivative_exogsc C snddlm}m}|dk r@|||||||}n|dk rj|||||||}n|S(sK Helper for _derivative_exog to wrap results appropriately i(t_get_count_effectst_get_dummy_effectsN(tdiscrete_marginsRRRV( R:tmargeffRKR;RRt transformRR((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyt_derivative_exog_helpers  N(t__name__t __module__t__doc__R8RDRGRHRORVRWtbasetLikelihoodModelR"tFalseRoReRRR(((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyR6s&          t BinaryModelc B s}eZdZdedZddddddddddd d Zejje_dd d Z dd ddd Z RS(cK shtt|j|||t|jt rdtj|jdk|jdk@ rdt dndS(Niis#endog must be in the unit interval.( R7RR8t issubclasst __class__tMultinomialModelR=tallR%R3(R:R%R;R<((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyR8s&cC sK|dkr|j}n|s7|jtj||Stj||SdS(sV Predict response variable of a model given exogenous variables. Parameters ---------- params : array-like Fitted parameters of the model. exog : array-like 1d or 2d array of exogenous values. If not supplied, the whole exog attribute of the model is used. linear : bool, optional If True, returns the linear predictor dot(exog,params). Else, returns the value of the cdf at the linear predictor. Returns ------- array Fitted values at exog. N(RVR;RGR=RI(R:RKR;R((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyRs   R0RYiiRZg{Gz?g-C6?gQ?c K s}t|tt|jd|d|d|d|d|d|d|d|d | d | d | | } t|| }t|S( NRQR2RRRSRTRUR[R]R_R^R`(R5R7RRotL1BinaryResultstL1BinaryResultsWrapper(R:RQR2RRRSRTRUR[R]R_R^R`R<tbnryfitt discretefit((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyRos tdydxcC s|dkr|j}n|jtj||dddf|}d|kr|||j||dddf}n|S(s} For computing marginal effects standard errors. This is used only in the case of discrete and count regressors to get the variance-covariance of the marginal effects. It returns [d F / d params] where F is the predict. Transform can be 'dydx' or 'eydx'. Checking is done in margeff computations for appropriate transform. Ntey(RVR;RHR=RIR(R:RKR;RtdF((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyt_derivative_predicts  / )cC s|dkr|j}ntj|jtj||dddf|dddf}d|krx||9}nd|kr||j||dddf}n|j||||||S(s; For computing marginal effects returns dF(XB) / dX where F(.) is the predicted probabilities transform can be 'dydx', 'dyex', 'eydx', or 'eyex'. Not all of these make sense in the presence of discrete regressors, but checks are done in the results in get_margeff. NtexR(RVR;R=RIRHRR(R:RKR;RRRR((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyRs  .   )N( RRR8RVRRRoR6RRR(((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyRs    Rc B seZdZdZdedZdddddddZejj e_ ddddddd d d d d d Z ej j e _ dddZ dddddZ RS(c K sLtj|dr-t|\}}d}nQtj|drWt|\}}}n'tj|}t|\}}d}t|t st t t |j d|}n||_ t|||||} || _|| _| j|_xT|D]L} | dkrqny t|| | jj| Wqtk rCqXqW| S(NR)it design_infotformula(RR(t data_toolst_is_using_ndarray_typeRVR(t_is_using_pandasR/R=tasarrayt isinstanceRjtzipRR$t _ynames_mapRR't orig_endogR%twendogtsetattrt__dict__tpoptKeyError( R:R%R;tmissingthasconstR<R&R'R.Rtkey((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyt _handle_datas. %        cC stt|j|jjd|_|jjd|_|jjd|_ |j |jd9_ |jjd|j |jd|_ dS(s4 Preprocesses the data for MNLogit. iiN( R7RRDR%targmaxRR$tJR;tKRARB(R:((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyRD-s cC s|dkr|j}n|jdkr4|d}ntt|j|||}|rtjtjt ||f}n|S(s Predict response variable of a model given exogenous variables. Parameters ---------- params : array-like 2d array of fitted parameters of the model. Should be in the order returned from the model. exog : array-like 1d or 2d array of exogenous values. If not supplied, the whole exog attribute of the model is used. If a 1d array is given it assumed to be 1 row of exogenous variables. If you only have one regressor and would like to do prediction, you must provide a 2d array with shape[1] == 1. linear : bool, optional If True, returns the linear predictor dot(exog,params). Else, returns the value of the cdf at the linear predictor. Notes ----- Column 0 is the base case, the rest conform to the rows of params shifted up one for the base case. iN( RVR;R#R7RRR=t column_stackRutlen(R:RKR;Rtpred((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyR9s   'RPi#ic K s|dkr,tj|j|jd}ntj|}d}tjj|d|d|d|d|d|d||}|j j |jd d d |_ t ||}t |S( NicS sdS(N(RV(tx((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyt`tRQR2RRRSRTRUitordertF( RVR=RuRRRRRRWRKtreshapetMultinomialResultstMultinomialResultsWrapper( R:RQR2RRRSRTRUR<tmnfit((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyRWZs   !R0RYiRZg{Gz?g-C6?gQ?c K s|dkr,tj|j|jd}ntj|}tj|d|d|d|d|d|d|d|d |d | d | d | |  } | jj |jd dd| _t || } t | S(NiRQR2RRRSRTRUR[R]R_R^R`iRR( RVR=RuRRRR6RoRKRtL1MultinomialResultstL1MultinomialResultsWrapper(R:RQR2RRRSRTRUR[R]R_R^R`R<R((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyRojs  !RcC s2|dkr|j}n|jdkrL|j|j|jddd}ntjtj||}d|j ddddf}t |j}t |j}tj ||dd}tj ||d} | | |d} |j dddddf| |||d} | jd } tjtj|dtj|jt} |j dddddf | ||djd dd| f| dd| f 0) constaint. In case use_transparams=True and method="newton" or "ncg" transformation is ignored. RPtncgs&Parameter "use_transparams" is ignoredRiRiRTt skip_hessianR toptim_kwds_prelimRBgcW s|S(N((R((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyRRRQRRR2RSRUiRtuse_selftuse_t(RPR<N(RPR<(R"R/twarningstwarntRuntimeWarningRRVRR=tsizetupdateRRR%R;RWRKR6RR7RBR.tmaxR7RRRtGeneralizedPoissonResultst GeneralizedPoissonResultsWrappert_get_robustcov_results(R:RQR2RRRSRTRUtuse_transparamsRRR@R<RR>tmod_poitres_poiR8RXtgpfittresult((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyRWsF      "!         &   R0RYiRZg{Gz?g-C6?gQ?c K st|tj|dkrb|dkrb|jjd|j} |tj| }d|d|jtkrtj|dt}tj|d|dRBg?RicW s|S(N((R((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyR RRQRRR2RSRURR?R@(RPR<(RPR<(RR/R}t startswithR"RVRR=RDRERRR%R;RWRKR6RR7RBR.RFRfRR7RRRtNegativeBinomialResultstNegativeBinomialResultsWrapperRI(R:RQR2RRRSRTRURRR@R<RR>RKRLR8RXtnbinfitRN((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyRW sR     "!    !      &  R0RYRZg{Gz?g-C6?gQ?c K st||jjdrttj|dkrt|dkrt|jjd|j} |tj| }d|d 0) constaint. In case use_transparams=True and method="newton" or "ncg" transformation is ignored. RPR<s&Parameter "use_transparams" is ignoredRiRiRTR=R R>RBg?cW s|S(N((R((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyR RRQRRR2RSRUiRR?R@(RPR<N(RPR<(R"R/RARBRCRRVRR=RDRERRR%R;RWRKR6RR7RBR.RFR7RRRRRRI(R:RQR2RRRSRTRURJRRR@R<RR>RKRLR8RXRRN((sBlib/python2.7/site-packages/statsmodels/discrete/discrete_model.pyRWd sF      "!         &   R0RYiRZg{Gz?g-C6?gQ?c K st|tj|dkrb|dkrb|jjd|j} |tj| }d|ds           cCp 4