ó áp7]c@sˆdZddlmZmZddlZddlZddlmZddl m Z yddl Z e Z Wnek rƒeZ nXddlmZedejd ejd ejd ejd ejd ejdejdejdejdejdejdejƒ Zd„Z d„Z!de"fd„ƒYZ#de"fd„ƒYZ$de"fd„ƒYZ%d„Z&d„Z'd dd e d„Z(dS(sh Module containing the base object for multivariate kernel density and regression, plus some utilities. iÿÿÿÿ(tranget string_typesN(toptimize(t mquantilesi(tkernelst wangryzintaitchisonaitkentgaussiantaitchison_aitken_regt wangryzin_regtgauss_convolutiontwangryzin_convolutiontaitchisonaitken_convolutiont gaussian_cdftaitchisonaitken_cdft wangryzin_cdft d_gaussiancCsqtj|ddƒ}t|dddƒjd}t|dddƒjd}||d}tj||ƒ}|S(s1Compute minimum of std and IQR for each variable.taxisigè?gÐ?g/Ý$•õ?(tnptstdRtdatatminimum(Rts1tq75tq25ts2t dispersion((sElib/python2.7/site-packages/statsmodels/nonparametric/_kernel_base.pyt_compute_min_std_IQR#s c Csƒ| r5tjj|ƒ|d|…dd…f} n$|| d| d…dd…f} |dkr¦ddlm} | d}| | |d|dtdtƒƒ}nD|d kr6dd lm}| \}}}| dd…d|…f}| dd…|d…f}|||||d|dtdtƒƒ}n´|d krÞdd lm }| \}}}t | dd…dfdƒ}t | dd…dd…f|ƒ}|d |d|d|d|d|dtdtƒƒ}n t dƒ‚|d kr| dd…dd…f} nt | ƒ}||d||}|d||||<|d||||<|j |}|j }||fS(s¾"Compute bw on subset of data. Called from ``GenericKDE._compute_efficient_*``. Notes ----- Needs to be outside the class in order for joblib to be able to pickle it. NiitKDEMultivariate(Rtbwtdefaultst efficienttKDEMultivariateConditional(R t KernelReg(R!tendogtexogtreg_typetvar_typesdclass_type not recognized, should be one of {KDEMultivariate, KDEMultivariateConditional, KernelReg}gð¿gÀ(Rtrandomtshuffletkernel_densityRtEstimatorSettingstFalseR tkernel_regressionR!t _adjust_shapet ValueErrorRR(t class_typeRRtcotdotn_cvarstix_ordtix_unordtn_subt class_varst randomizetboundtsub_dataRR%t sub_modelR tk_deptdep_typet indep_typeR"R#R!tk_varsR$Rtfcttsample_scale_subtbw_sub((sElib/python2.7/site-packages/statsmodels/nonparametric/_kernel_base.pyt_compute_subset-sD $    ! %      t GenericKDEcBsheZdZd„Zd„Zd„Zd„Zd„Zd„Zd„Z d„Z d „Z d „Z RS( sG Base class for density estimation and regression KDE classes. cCsŽ|dkrd}nt|tƒs?d|_tj|ƒ}nK||_|dkr`|j}n!|dkrx|j}n |j}|ƒ}|S(sé Computes the bandwidth of the data. Parameters ---------- bw: array_like or str If array_like: user-specified bandwidth. If a string, should be one of: - cv_ml: cross validation maximum likelihood - normal_reference: normal reference rule of thumb - cv_ls: cross validation least squares Notes ----- The default values for bw is 'normal_reference'. tnormal_referencesuser-specifiedtcv_mlN( tNonet isinstanceRt _bw_methodRtasarrayt_normal_referencet_cv_mlt_cv_ls(tselfRtrestbwfunc((sElib/python2.7/site-packages/statsmodels/nonparametric/_kernel_base.pyt _compute_bwhs          cCs t|ƒS(sß Computes the measure of dispersion. The minimum of the standard deviation and interquartile range / 1.349 Notes ----- Reimplemented in `KernelReg`, because the first column of `data` has to be removed. References ---------- See the user guide for the np package in R. In the notes on bwscaling option in npreg, npudens, npcdens there is a discussion on the measure of dispersion (R(RLR((sElib/python2.7/site-packages/statsmodels/nonparametric/_kernel_base.pyt_compute_dispersionŽscCsdS(spHelper method to be able to pass needed vars to _compute_subset. Needs to be implemented by subclasses.N((RL((sElib/python2.7/site-packages/statsmodels/nonparametric/_kernel_base.pyt_get_class_vars_type¡sc s6ˆd krdˆ _ntˆtƒr3ˆˆ _n dˆ _ˆSˆ j}ˆ j‰ tjˆ jƒ‰ˆ j j dƒ‰ d‰d‰t ˆ j ƒ\}‰‰ˆ j r¶d gˆ j ‰ncgt|ˆ ƒD]}|ˆ |dˆ f^qlj|ˆ dkrˆj||ˆ |fƒnˆ j r+ˆ j n tˆƒ}tj|ˆ jfƒ}tj|ˆ jfƒ}ˆ jƒ\‰‰trÔtjdˆ jƒ‡‡‡‡‡‡‡‡‡‡ ‡ ‡ f d†t|ƒDƒƒ}nXg}xOt|ƒD]A}|jtˆˆˆˆˆˆ ˆˆˆ ˆˆ j ˆ|ƒ ƒqçWxPt|ƒD]B}||d||d d …f<||d||d d …fÔsNRgð¿gÀ(RERGRFRtnobsR4tcopytdeepcopyRt data_typetcountt _get_type_posR6tn_resRtappendtlenRtemptyR=RQt has_joblibRTtParallelRSRARPt return_mediantmediantmeantreturn_only_bw( RLRRYt_RWtn_blockst sample_scaletonly_bwRMtst order_functm_scale(( RXRR.R5R/RR0R2R3R1R4RLsElib/python2.7/site-packages/statsmodels/nonparametric/_kernel_base.pyt_compute_efficient§sX       5-  " cCsX|j|_|j|_|j|_|j|_|j|_|j|_|j|_dS(s4Sets the default values for the efficient estimationN(R_R4R6ReRRhRS(RLR((sElib/python2.7/site-packages/statsmodels/nonparametric/_kernel_base.pyt _set_defaultsñs      cCs=tj|jddƒ}d||jdd|jjdS(s\ Returns Scott's normal reference rule of thumb bandwidth parameter. Notes ----- See p.13 in [2] for an example and discussion. The formula for the bandwidth is .. math:: h = 1.06n^{-1/(4+q)} where ``n`` is the number of observations and ``q`` is the number of variables. Rigö(\Âõð?gð¿ii(RRRRYtshape(RLtX((sElib/python2.7/site-packages/statsmodels/nonparametric/_kernel_base.pyRIûscCs`d||dk> 300) and/or multiple variables (k_vars > 3). If False (default), all data is used at the same time. randomize: bool, optional If True, the bandwidth estimation is to be performed by taking `n_res` random resamples (with replacement) of size `n_sub` from the full sample. If set to False (default), the estimation is performed by slicing the full sample in sub-samples of size `n_sub` so that all samples are used once. n_sub: int, optional Size of the sub-samples. Default is 50. n_res: int, optional The number of random re-samples used to estimate the bandwidth. Only has an effect if ``randomize == True``. Default value is 25. return_median: bool, optional If True (default), the estimator uses the median of all scaling factors for each sub-sample to estimate the bandwidth of the full sample. If False, the estimator uses the mean. return_only_bw: bool, optional If True, the estimator is to use the bandwidth and not the scaling factor. This is *not* theoretically justified. Should be used only for experimenting. n_jobs : int, optional The number of jobs to use for parallel estimation with ``joblib.Parallel``. Default is -1, meaning ``n_cores - 1``, with ``n_cores`` the number of available CPU cores. See the `joblib documentation `_ for more details. Examples -------- >>> settings = EstimatorSettings(randomize=True, n_jobs=3) >>> k_dens = KDEMultivariate(data, var_type, defaults=settings) ii2iÿÿÿÿcCsC||_||_||_||_||_||_||_dS(N(RR6R_R4ReRhRS(RLRR6R_R4ReRhRS((sElib/python2.7/site-packages/statsmodels/nonparametric/_kernel_base.pyt__init__ƒs      (RR‚RƒR*tTrueR„(((sElib/python2.7/site-packages/statsmodels/nonparametric/_kernel_base.pyR)Ss/ t LeaveOneOutcBs eZdZd„Zd„ZRS(sŽ Generator to give leave-one-out views on X. Parameters ---------- X : array-like 2-D array. Examples -------- >>> X = np.random.normal(0, 1, [10,2]) >>> loo = LeaveOneOut(X) >>> for x in loo: ... print x Notes ----- A little lighter weight than sklearn LOO. We don't need test index. Also passes views on X, not the index. cCstj|ƒ|_dS(N(RRHRs(RLRs((sElib/python2.7/site-packages/statsmodels/nonparametric/_kernel_base.pyR„£sccsp|j}tj|ƒ\}}xKt|ƒD]=}tj|dtjƒ}t||<||dd…fVq+WdS(Ntdtype(RsRRrRtonestboolR*(RLRsRYR=RWtindex((sElib/python2.7/site-packages/statsmodels/nonparametric/_kernel_base.pyt__iter__¦s   (RR‚RƒR„R‹(((sElib/python2.7/site-packages/statsmodels/nonparametric/_kernel_base.pyR†Žs cCs…tjg|D]}|dk^q ƒ}tjg|D]}|dk^q5ƒ}tjg|D]}|dk^q]ƒ}|||fS(NRRtotu(Rtarray(R%RRtix_contR2R3((sElib/python2.7/site-packages/statsmodels/nonparametric/_kernel_base.pyR^°s(((cCsëtj|ƒ}|jdkr0tj|ƒ}n|jdkrT|dkrTd}n{|jdkr~|dkr~t|ƒ}nQtj|ƒd|kr¼tj|ƒd|kr¼|j}ntj|ƒd}tj|||fƒ}|S(s> Returns an array of shape (nobs, k_vars) for use with `gpke`.iii(RRHtndimtsqueezeRaRrtTtreshape(tdatR=RY((sElib/python2.7/site-packages/statsmodels/nonparametric/_kernel_base.pyR,·s 2 cCsûtd|d|d|ƒ}tj|jƒ} xbt|ƒD]T\} } t|| } | || |dd…| f|| ƒ| dd…| f(#()Rƒtstatsmodels.compat.pythonRRRZtnumpyRtscipyRtscipy.stats.mstatsRRTR…Rct ImportErrorR*tRR•t wang_ryzintaitchison_aitkenRRtwang_ryzin_regtgaussian_convolutiontwang_ryzin_convolutiontaitchison_aitken_convolutionR taitchison_aitken_cdftwang_ryzin_cdfRR—RRAtobjectRBR)R†R^R,R¦(((sElib/python2.7/site-packages/statsmodels/nonparametric/_kernel_base.pyts@                  7ï;"