ó áp7]c@sÂddlZddlZddlmZmZddlmZe e e e d„Z e d„Z d„Z defd„ƒYZd efd „ƒYZe e e d „Ze d „Ze d „ZdS(iÿÿÿÿN(tchi2tnorm(tutilscCs‹|dkr*tj|dtƒ\}}nCtj||fƒ}tj|dtƒ\}}|dt|ƒ!}t|ƒ} |dkr£tj|d|d| ƒ} ntj|d||d| ƒ} |dkrætj|d| ƒ} ntj|d|d| ƒ} |dk r—tj| ƒ| } tj||ddƒ} |dkr\tj| d| ƒ} ntj| d|d| ƒ} tj| ƒ| } | | } n)tj| ddd…ƒddd…} |rütj | dkƒ}| |} | |} ||}nd | | j tj ƒ}|d k}d ||= 1 indicates which event occured at time t. If status = 0, the subject was censored at time t. title : string Optional title used for plots and summary output. freq_weights : array-like Optional frequency weights exog : array-like Optional, if present used to account for violation of independent censoring. bw_factor : float Band-width multiplier for kernel-based estimation. Only used if exog is provided. dimred : boolean If True, proportional hazards regression models are used to reduce exog to two columns by predicting overall events and censoring in two separate models. If False, exog is used directly for calculating kernel weights without dimension reduction. Attributes ---------- times : array-like The distinct times at which the incidence rates are estimated cinc : list of arrays cinc[k-1] contains the estimated cumulative incidence rates for outcome k=1,2,... cinc_se : list of arrays The standard errors for the values in `cinc`. Not available when exog and/or frequency weights are provided. Notes ----- When exog is provided, a local estimate of the cumulative incidence rate around each point is provided, and these are averaged to produce an estimate of the marginal cumulative incidence functions. The procedure is analogous to that described in Zeng (2004) for estimation of the marginal survival function. The approach removes bias resulting from dependent censoring when the censoring becomes independent conditioned on the columns of exog. References ---------- The Stata stcompet procedure: http://www.stata-journal.com/sjpdf.html?articlenum=st0059 Dinse, G. E. and M. G. Larson. 1986. A note on semi-Markov models for partially censored data. Biometrika 73: 379-386. Marubini, E. and M. G. Valsecchi. 1995. Analysing Survival Data from Clinical Trials and Observational Studies. Chichester, UK: John Wiley & Sons. D. Zeng (2004). Estimating marginal survival function by adjusting for dependent censoring using many covariates. Annals of Statistics 32:4. http://arxiv.org/pdf/math/0409180.pdf gð?c sRt||d|dƒtj|ƒ}|_tj|ƒ}|_|dk rgtj|ƒ}|_n|dk rddlm}tj|ƒ}|_ |j d} | dd|‰‡fd†} ||||| ||ƒ} | d|_ | d|_ dSt |||ƒ} | d|_ | d|_| d|_ |sEdn||_dS( Ni(t_kernel_cumincidenceiiÿÿÿÿg@cs#tj|d ˆdƒjdƒS(Nii(R RR4(tx(tkw(s<lib/python2.7/site-packages/statsmodels/duration/survfunc.pytùtiRL(RFRR tasarrayRRRCt_kernel_estimatesRHRDRBttimestcincR?tcinc_settitle( tselfRRRRRCRDt bw_factortdimredRHtnobstkfuncRI((RJs<lib/python2.7/site-packages/statsmodels/duration/survfunc.pyt__init__ës*         N(t__name__t __module__t__doc__RR RX(((s<lib/python2.7/site-packages/statsmodels/duration/survfunc.pyRG sIt SurvfuncRightcBseeZdZd d d d dd„Zd d„Zd„Zddd„Zd„Zdd d d „Z RS( s Estimation and inference for a survival function. The survival function S(t) = P(T > t) is the probability that an event time T is greater than t. This class currently only supports right censoring. Parameters ---------- time : array-like An array of times (censoring times or event times) status : array-like Status at the event time, status==1 is the 'event' (e.g. death, failure), meaning that the event occurs at the given value in `time`; status==0 indicates that censoring has occured, meaning that the event occurs after the given value in `time`. entry : array-like, optional An array of entry times for handling left truncation (the subject is not in the risk set on or before the entry time) title : string Optional title used for plots and summary output. freq_weights : array-like Optional frequency weights exog : array-like Optional, if present used to account for violation of independent censoring. bw_factor : float Band-width multiplier for kernel-based estimation. Only used if exog is provided. Attributes ---------- surv_prob : array-like The estimated value of the survivor function at each time point in `surv_times`. surv_prob_se : array-like The standard errors for the values in `surv_prob`. Not available if exog is provided. surv_times : array-like The points where the survival function changes. n_risk : array-like The number of subjects at risk just before each time value in `surv_times`. Not available if exog is provided. n_events : array-like The number of events (e.g. deaths) that occur at each point in `surv_times`. Not available if exog is provided. Notes ----- If exog is None, the standard Kaplan-Meier estimator is used. If exog is not None, a local estimate of the marginal survival function around each point is constructed, and these are then averaged. This procedure gives an estimate of the marginal survival function that accounts for dependent censoring as long as the censoring becomes independent when conditioning on the covariates in exog. See Zeng et al. (2004) for details. References ---------- D. Zeng (2004). Estimating marginal survival function by adjusting for dependent censoring using many covariates. Annals of Statistics 32:4. http://arxiv.org/pdf/math/0409180.pdf gð?c s²t|||||ƒtj|ƒ}|_tj|ƒ}|_|dk rgtj|ƒ}|_n|dk rŒtj|ƒ}|_n|dk r=|dk r³tdƒ‚nddl m }tj|ƒ}|_ |j d} | dd|‰‡fd†} ||||| |ƒ} | d|_ | d|_dSt||d|d |ƒ} | d|_ | d|_| d |_| d |_| d |_|s¥d n||_dS(Ns%exog and entry cannot both be presenti(t_kernel_survfunciiÿÿÿÿg@cs#tj|d ˆdƒjdƒS(Nii(R RR4(RI(RJ(s<lib/python2.7/site-packages/statsmodels/duration/survfunc.pyRK^RLRRiiiRL(RFR RMRRRRCRR@RNR]RDRBt surv_probt surv_timesR-t surv_prob_setn_risktn_eventsRR( RSRRRRRRCRDRTR]RVRWRI((RJs<lib/python2.7/site-packages/statsmodels/duration/survfunc.pyRXKs6             cCs t||ƒS(s¦ Plot the survival function. Examples -------- Change the line color: >>> import statsmodels.api as sm >>> data = sm.datasets.get_rdataset("flchain", "survival").data >>> df = data.loc[data.sex == "F", :] >>> sf = sm.SurvfuncRight(df["futime"], df["death"]) >>> fig = sf.plot() >>> ax = fig.get_axes()[0] >>> li = ax.get_lines() >>> li[0].set_color('purple') >>> li[1].set_color('purple') Don't show the censoring points: >>> fig = sf.plot() >>> ax = fig.get_axes()[0] >>> li = ax.get_lines() >>> li[1].set_visible(False) (t plot_survfunc(RStax((s<lib/python2.7/site-packages/statsmodels/duration/survfunc.pytplotnscCsDtj|jd|kƒ}t|ƒdkr5tjS|j|dS(sõ Estimated quantile of a survival distribution. Parameters ---------- p : float The probability point at which the quantile is determined. Returns the estimated quantile. ii(R RR^R RR_(RStpR(((s<lib/python2.7/site-packages/statsmodels/duration/survfunc.pytquantileŠsgš™™™™™©?tcloglogc Cs¤tjd|dƒ}|jƒ}|dkrDd„}d„}n|dkred„}d„}no|d kr†d „}d „}nN|d kr§d „}d„}n-|dkrÈd„}d„}n tdƒ‚||jƒ|d|ƒ}|||jƒ|j:}tjtj|ƒ|kƒ}t |ƒdkrKtj tj fS|j |d} |dt |j ƒdkr…tj } n|j |dd} | | fS(så Returns a confidence interval for a survival quantile. Parameters ---------- p : float The probability point for which a confidence interval is determined. alpha : float The confidence interval has nominal coverage probability 1 - `alpha`. method : string Function to use for g-transformation, must be ... Returns ------- lb : float The lower confidence limit. ub : float The upper confidence limit. Notes ----- The confidence interval is obtained by inverting Z-tests. The limits of the confidence interval will always be observed event times. References ---------- The method is based on the approach used in SAS, documented here: http://support.sas.com/documentation/cdl/en/statug/68162/HTML/default/viewer.htm#statug_lifetest_details03.htm iiRhcSstjtj|ƒ ƒS(N(R R(RI((s<lib/python2.7/site-packages/statsmodels/duration/survfunc.pyRKÆRLcSsd|tj|ƒS(Niÿÿÿÿ(R R(RI((s<lib/python2.7/site-packages/statsmodels/duration/survfunc.pyRKÇRLtlinearcSs|S(N((RI((s<lib/python2.7/site-packages/statsmodels/duration/survfunc.pyRKÉRLcSsdS(Ni((RI((s<lib/python2.7/site-packages/statsmodels/duration/survfunc.pyRKÊRLRcSs tj|ƒS(N(R R(RI((s<lib/python2.7/site-packages/statsmodels/duration/survfunc.pyRKÌRLcSsd|S(Ni((RI((s<lib/python2.7/site-packages/statsmodels/duration/survfunc.pyRKÍRLtlogitcSstj|d|ƒS(Ni(R R(RI((s<lib/python2.7/site-packages/statsmodels/duration/survfunc.pyRKÏRLcSsd|d|S(Ni((RI((s<lib/python2.7/site-packages/statsmodels/duration/survfunc.pyRKÐRLtasinsqrtcSstjtj|ƒƒS(N(R tarcsinR(RI((s<lib/python2.7/site-packages/statsmodels/duration/survfunc.pyRKÒRLcSs&ddtj|ƒtjd|ƒS(Nii(R R(RI((s<lib/python2.7/site-packages/statsmodels/duration/survfunc.pyRKÓRLsunknown methodiiÿÿÿÿ( RtppftlowerR@R^R`R RtabsR RR_R( RSRftalphatmethodttrtgtgprimetrR(tlbtub((s<lib/python2.7/site-packages/statsmodels/duration/survfunc.pyt quantile_ciŸs8#                  cCsYtjd|jƒ}d|j_|j|d<|j|d<|j|d<|j|d<|S(sÐ Return a summary of the estimated survival function. The summary is a datafram containing the unique event times, estimated survival function values, and related quantities. tindextTimes Surv probs Surv prob SEs num at risks num events( tpdt DataFrameR_RytnameR^R`RaRb(RStdf((s<lib/python2.7/site-packages/statsmodels/duration/survfunc.pytsummaryçs     thwRcCsÖ|jƒ}|dkr-d}t|ƒ‚n|dkrHtdƒ‚n|jƒ}|jd|jd}|j}|dkrætj|ƒtj|jƒ}dd|||}tj|ƒ}|jd|} |j|} næ|d krÀd} | d||dtj|ƒ9} | tj|jd|jƒ9} tj tj|jƒƒ} tj | | d tj ƒ} tj | ƒd} tj | | tj tj dƒ} tj | ƒd} n td ƒ‚| | fS( s$ Returns a simultaneous confidence band for the survival function. Parameters ---------- alpha : float `1 - alpha` is the desired simultaneous coverage probability for the confidence region. Currently alpha must be set to 0.05, giving 95% simultaneous intervals. method : string The method used to produce the simultaneous confidence band. Only the Hall-Wellner (hw) method is currently implemented. transform : string The used to produce the interval (note that the returned interval is on the survival probability scale regardless of which transform is used). Only `log` and `arcsin` are implemented. Returns ------- lcb : array-like The lower confidence limits corresponding to the points in `surv_times`. ucb : array-like The upper confidence limits corresponding to the points in `surv_times`. R€s0only the Hall-Wellner (hw) method is implementedgš™™™™™©?salpha must be set to 0.05iRg_)ËǺõ?iRlisUnknown transform(RnR@R`R^RaR RRRRlRRtsintpi(RSRpRqt transformREts2tnnR*tthetatlcbtucbR7tfR>((s<lib/python2.7/site-packages/statsmodels/duration/survfunc.pytsimultaneous_cbøs4       #!$ N( RYRZR[RRXReRgRxRRŠ(((s<lib/python2.7/site-packages/statsmodels/duration/survfunc.pyR\sB"  H c KsXtj|ƒ}tj|ƒ}tj|ƒ}tj|ƒ}t|ƒdkr]tdƒ‚n|dkrt|||||||\}} n‹tj|ƒ}tj|ƒ} d\}} x^| D]V} || k} t|| || || ||||\} }|| 7}| |7} qÁW|tj| ƒ}|d}dtj |dƒ}||fS(s; Test for the equality of two survival distributions. Parameters ---------- time : array-like The event or censoring times. status : array-like The censoring status variable, status=1 indicates that the event occured, status=0 indicates that the observation was censored. group : array-like Indicators of the two groups weight_type : string The following weight types are implemented: None (default) : logrank test fh : Fleming-Harrington, weights by S^(fh_p), requires exponent fh_p to be provided as keyword argument; the weights are derived from S defined at the previous event time, and the first weight is always 1. gb : Gehan-Breslow, weights by the number at risk tw : Tarone-Ware, weights by the square root of the number at risk strata : array-like Optional stratum indicators for a stratified test entry : array-like Entry times to handle left truncation. The subject is not in the risk set on or before the entry time. Returns ------- chisq : The chi-square (1 degree of freedom) distributed test statistic value pvalue : The p-value for the chi^2 test is logrank only supports two groupsgiN(gg( R RMR R R@Rt _survdiffRRtcdf(RRtgroupt weight_typetstrataRtkwargstgrtobstvartstutstR(tobs1tvar1tzstattchisqtpvalue((s<lib/python2.7/site-packages/statsmodels/duration/survfunc.pytsurvdiff6s,)      c#Ks'|dkr*tj|dtƒ\}}n=tjtj||fƒdtƒ\}}|dt|ƒ!}|ddf|ddfg} |dk rÕx?dD]4} ||| k} || } || | f| | >> import statsmodels.api as sm >>> from statsmodels.duration.survfunc import plot_survfunc >>> data = sm.datasets.get_rdataset("flchain", "survival").data >>> df = data.loc[data.sex == "F", :] >>> sf0 = sm.SurvfuncRight(df["futime"], df["death"]) >>> sf1 = sm.SurvfuncRight(3.0 * df["futime"], df["death"]) >>> fig = plot_survfunc([sf0, sf1]) >>> ax = fig.get_axes()[0] >>> ax.set_position([0.1, 0.1, 0.64, 0.8]) >>> ha, lb = ax.get_legend_handles_labels() >>> leg = fig.legend((ha[0], ha[1]), (lb[0], lb[1]), 'center right') Change the line colors: >>> fig = plot_survfunc([sf0, sf1]) >>> ax = fig.get_axes()[0] >>> ax.set_position([0.1, 0.1, 0.64, 0.8]) >>> ha, lb = ax.get_legend_handles_labels() >>> ha[0].set_color('purple') >>> ha[1].set_color('orange') iiiÿÿÿÿRRsGroup %dt-tlabeltlwitwheretpostt+tmsi tcolors pointsg)\Âõ(ð?(Rt create_mpl_axttypeR\tAssertionErrort enumerateR R R_R^R0RtgetattrtstepRt logical_notRRRet get_colortset_ylim(t survfuncsRdtfigtgxtsfR_R^tmxtR°tliR(ttitjjR)((s<lib/python2.7/site-packages/statsmodels/duration/survfunc.pyRcÛs.(     $(tnumpyR tpandasR{tscipy.stats.distributionsRRtstatsmodels.graphicsRRR R-R?RFtobjectRGR\R›R‹Rc(((s<lib/python2.7/site-packages/statsmodels/duration/survfunc.pyts    N 6 gÿ0 H \