M/Ph=dZddlZddlZddlZddlZddlmZddl m Z dZ dZ dZ dZd Ze e  ddZdZdS)ae A predict-like function that constructs means and pointwise or simultaneous confidence bands for the function f(x) = E[Y | X*=x, X1=x1, ...], where X* is the focus variable and X1, X2, ... are non-focus variables. This is especially useful when conducting a functional regression in which the role of x is modeled with b-splines or other basis functions. N) ValueWarning)Appendera Predictions and contrasts of a fitted model as a function of a given covariate. The value of the focus variable varies along a sequence of its quantiles, calculated from the data used to fit the model. The other variables are held constant either at given values, or at values obtained by applying given summary functions to the data used to fit the model. Optionally, a second specification of the non-focus variables is provided and the contrast between the two specifications is returned. Parameters ---------- result : statsmodels result object A results object for the fitted model. focus_var : str The name of the 'focus variable'. summaries : dict-like A map from names of non-focus variables to summary functions. Each summary function is applied to the data used to fit the model, to obtain a value at which the variable is held fixed. values : dict-like Values at which a given non-focus variable is held fixed. summaries2 : dict-like A second set of summary functions used to define a contrast. values2 : dict-like A second set of fixed values used to define a contrast. alpha : float `1 - alpha` is the coverage probability. ci_method : str The method for constructing the confidence band, one of 'pointwise', 'scheffe', and 'simultaneous'. num_points : int The number of equally-spaced quantile points where the prediction is made. exog : array_like Explicitly provide points to cover with the confidence band. exog2 : array_like Explicitly provide points to contrast to `exog` in a functional confidence band. kwargs : Arguments passed to the `predict` method. Returns ------- pred : array_like The predicted mean values. cb : array_like An array with two columns, containing respectively the lower and upper limits of a confidence band. fvals : array_like The values of the focus variable at which the prediction is made. Notes ----- All variables in the model except for the focus variable should be included as a key in either `summaries` or `values` (unless `exog` is provided). If `summaries2` and `values2` are not provided, the returned value contains predicted conditional means for the outcome as the focus variable varies, with the other variables fixed as specified. If `summaries2` and/or `values2` is provided, two sets of predicted conditional means are calculated, and the returned value is the contrast between them. If `exog` is provided, then the rows should contain a sequence of values approximating a continuous path through the domain of the covariates. For example, if Z(s) is the covariate expressed as a function of s, then the rows of exog may approximate Z(g(s)) for some continuous function g. If `exog` is provided then neither of the summaries or values arguments should be provided. If `exog2` is also provided, then the returned value is a contrast between the functionas defined by `exog` and `exog2`. Examples -------- Fit a model using a formula in which the predictors are age (modeled with splines), ethnicity (which is categorical), gender, and income. Then we obtain the fitted mean values as a function of age for females with mean income and the most common ethnicity. >>> model = sm.OLS.from_formula('y ~ bs(age, df=4) + C(ethnicity) + gender + income', data) >>> result = model.fit() >>> mode = lambda x : x.value_counts().argmax() >>> summaries = {'income': np.mean, ethnicity=mode} >>> values = {'gender': 'female'} >>> pr, cb, x = predict_functional(result, 'age', summaries, values) Fit a model using arrays. Plot the means as a function of x3, holding x1 fixed at its mean value in the data used to fit the model, and holding x2 fixed at 1. >>> model = sm.OLS(y ,x) >>> result = model.fit() >>> summaries = {'x1': np.mean} >>> values = {'x2': 1} >>> pr, cb, x = predict_functional(result, 'x3', summaries, values) Fit a model usng a formula and construct a contrast comparing the female and male predicted mean functions. >>> model = sm.OLS.from_formula('y ~ bs(age, df=4) + gender', data) >>> result = model.fit() >>> values = {'gender': 'female'} >>> values2 = {'gender': 'male'} >>> pr, cb, x = predict_functional(result, 'age', values=values, values2=values2) c|j}|jj|i}|i}|jt jdurt dt |t |z|gz}fd|D}tj |j hz }|t|z } t | } t| dkr:tjddd| Dzt t#|} t%j| | } t)||D]\} } t%j| | | | <t jdd | }t j||}t j|}|| jdd|f<|D].}||jdd|f| jdd|f</|D] }||| |<t5j|jj| d }|| |fS)a Create dataframes for exploring a fitted model as a function of one variable. This works for models fit with a formula. Returns ------- dexog : data frame A data frame in which the focus variable varies and the other variables are fixed at specified or computed values. fexog : data frame The data frame `dexog` processed through the model formula. NOz'focus variable may not have object typec*g|]}|jS)dtype).0xexogs f/var/www/html/test/jupyter/venv/lib/python3.11/site-packages/statsmodels/sandbox/predict_functional.py z+_make_exog_from_formula..s . . .d1gm . . .rz0%s in data frame but not in summaries or values., cg|]}d|zSz'%s'rr r s r rz+_make_exog_from_formula.."A"A"A!6A:"A"A"Ar)indexcolumns)rr d dataframe return_type)modeldataframer np ValueErrorlistkeyssetrtolist endog_nameslenwarningswarnjoinrrangepd DataFramezipSerieslinspace percentileasarraylocpatsydmatrix design_info)result focus_var summariesvalues num_pointsrcolnamesdtypesvarl unmatchedixfexogdr pctlsfvalskydexogr s @r _make_exog_from_formularEs LE : D  ~ I --BCCCINN$$%%V[[]](;(;;ykIH . . . .X . . .F t|""$$ % %):(; ;Ds8}}$IYI 9~~ H "A"Ay"A"A"ABBC" $ $ $ z  B Lr8 4 4 4EFH%%00192Q///a K3 + + 2 2 4 4E M$y/5 1 1E Ju  E#EIaaalnn::(9R=!!!R%99 !!!R%kkmm2Jb M%*0%&1 3 3 3E % rc|j}|j}|j}|i}|i}tj||jdf}t |t |z|gz} t|t| z } t | } t| dkr:tj dd d| Dzttjdd|} ||} tj|dd| f| } | |dd| f<|D]9}||} |||dd| f|dd| f<:|D]&}||} |||dd| f<'|| fS)a; Create dataframes for exploring a fitted model as a function of one variable. This works for models fit without a formula. Returns ------- exog : data frame A data frame in which the focus variable varies and the other variables are fixed at specified or computed values. Nrz/%s in model but not in `summaries` or `values`.rcg|]}d|zSrrrs r rz*_make_exog_from_arrays..rrr)rr exog_namesrzerosshaper r!r"r%r&r'r(rr.r#rr/)r5r6r7r8r9r model_exogrIr r:r=rAr>rBrCs r _make_exog_from_arraysrMs LEJ!J  ~ 8Z!1!!45 6 6DFKKMM""T)..*:*:%;%;;ykIHJ#h--/IYI 9~~ G "A"Ay"A"A"ABBC" $ $ $ K3 + + 2 2 4 4E   ) $ $B M*QQQU+U 3 3EDBKnn77   b ! !#imJqqq"u$566QQQU kkmm!!   b ! !RjQQQU ;rct|jjdrt|||||\}}}nt |||||\}}||}}|||fS)Nr)hasattrrrrErM) r5r6r7r8r9rDr?rBr s r _make_exogrPsvv| '**"5fi'0&*FFuee-VY !'55 eTu % rc^|i}|i}|i}|i}||f||ffD]\}}t|t|z}t|}t|dkr%t dd|z||||fS)NrzJOne or more variable names are contained in both `summaries` and `values`:r)r"r!r r%rr()r8r7values2 summaries2svrCs r _check_argsrV s ~  V$z7&;;,,1 ]]S]] * "XX r77Q;;i!YYr]]+,, ,  9gz 11r皙? pointwiseT c  >|dvrtd|dup|du} | r|std|j}| td||||fDrtd| }tj|jj|d}| |}| %| }tj|jj|d}|}nht||||\}}}}t||||| \}}}t|t|zdkrt||||| \}}}dd l m }dd l m }t|j||fr,| }|d d in| }|jdd |i|}| r|jdd |i|}||z }||z }|dkr-||}||}n|dkr||}|j}t+j| df}ddlm}|jjjd}|jjjd|z }|d|z ||} |t+j|| zz}!||!z |dddf<||!z|dddf<nY|dkrSt;|||\}"}#t+j|jddf}||#|"zz |dddf<||#|"zz|dddf<|s6|jj}$|$ |}|$ |}|||fS)N)rXscheffe simultaneouszQconfidence band method must be one of `pointwise`, `scheffe`, and `simultaneous`.z-`linear` must be True for computing contrastsc3K|]}|duV dS)Nrrs r z%predict_functional..0s&OOq}OOOOOOrzAif `exog` is provided then do not provide `summaries` or `values`rrr)GLM)GEEwhichlinearr rX)alphar[)frGr\r)!rranyr2r3rr4rVrPr%+statsmodels.genmod.generalized_linear_modelr_3statsmodels.genmod.generalized_estimating_equationsr` isinstancecopyupdatepredictt_testconf_intsdrrJscipy.stats.distributionsrer rKcdfsqrt_glm_basic_scrfamilylinkinverse)%r5r6r7r8rSrRrc ci_methodrbr9r exog2kwargscontrastrr?rDrBfexog2dexog2fvals2r_r` kwargs_predpredpred2rmcbrofdistdf1df2qffxsigmacrus% r predict_functionalrs @@@GHH Ht#@4)?HJJHIII LE  OOy*fg&NOOO O O @?@@ @ ej4#>>>Y  F]5:#9#){DDDFF2=V=F=D=G2I2I. 7J )I)/==ue z??S\\ )A - -%/ :07&E&E "FFF@?????GGGGGG&,c ++kkmm GX.//// 6> 4 4u 4 4 4D::F:k::e|Ku%% __5_ ) ) i  u%% Y Xz1o & & 988888l%a(l%a(3. YYq5y#s + + "'#(## #"9111a4"9111a4 n $ $!&%77q Xu{1~q) * *!E'>111a4!E'>111a4 }!||D!! \\"   U?rcf|j}|jjd}|}tj|}||z }tj|j}t j |||z d} t j t j | } tj |j|jj} | t j |z} | | dddfz} t j| dd} | dz d} t j |  ddlmfd}ddlm}||dd d \}}|jst)d | |fS) a The basic SCR from (Sun et al. Annals of Statistics 2000). Computes simultaneous confidence regions (SCR). Parameters ---------- result : results instance The fitted GLM results instance exog : array_like The exog values spanning the interval alpha : float `1 - alpha` is the coverage probability. Returns ------- An array with two columns, containing the lower and upper confidence bounds, respectively. Notes ----- The rows of `exog` should be a sequence of covariate values obtained by taking one 'free variable' x and varying it over an interval. The matrix `exog` is thus the basis functions and any other covariates evaluated as x varies. rrGN)axisrd)normctj|dz dz ztjz dd|z zzz S)NrdrG)rexppirq)rrckappa_0rs r funcz_glm_basic_scr..funcsDAa(2501a$((1++o3FFNNr)brentqrYT) full_outputzRoot finding error in basic SCR)rr rK cov_paramsrlinalginvcholeskyTdotsumr0rrsolvediffrprscipy.optimizer convergedr)r5r rcrncovhessABsigma2rbzbzdbzdnrrrrsltrrs ` @@r rsrss8 LE A     C 9==  D qA 1AfT3$& + +A . .F Jrwv ' 'E df % % 'B"'!**B%4.B '"aa C F<<??Dgdmm!!G......OOOOOOO&%%%%%fT1bd333GAt ><:;;; !8Or) NNNNrWrXTrYNN)__doc__pandasr*r2numpyrr&statsmodels.tools.sm_exceptionsrstatsmodels.compat.pandasr_predict_functional_docrErMrPrVrrsrrr rs 888888......nd@@@F222j   222*  !""AE<@FH(,]]]#"]@@@@@@r