U Kv fY@s.ddlmZddlZddlZddlmZmZddlZddl m Z ddl m Z m Z mZddlmZddlmZdd Zd d Zd d ZddZddZddZddZddZGdddZGdddZedkr*ddlZddlmZej ddd d!d"d#gd$Z!ed%e!d&"Z#ed'e!d&"Z$ee#d(d)Z%dS)*)lrangeN) DataFrameIndex)stats)_has_intercept_intercept_idx_remove_intercept_patsy)summary2)OLScCsX|dkr|S|dkr|jS|dkr,|jS|dkr:|jS|dkrH|jStd|dS)NZhc0Zhc1Zhc2Zhc3z robust options %s not understood)Z cov_paramsZcov_HC0Zcov_HC1Zcov_HC2Zcov_HC3 ValueError)modelrobustr;/tmp/pip-unpacked-wheel-2v6byqio/statsmodels/stats/anova.py_get_covariancesrc Ks2|dd}|dd}|dd}|dd}|r<|}|jj}|jj}|jd}|jj} |jjj} |jj } t | j t | d} d |} d d d || g}t t| d f|d}|dkrt||||| || || | S|dkrt|| | || |S|dkr t|| | || |S|dkrtdntdt|dS)a9 Anova table for one fitted linear model. Parameters ---------- model : fitted linear model results instance A fitted linear model typ : int or str {1,2,3} or {"I","II","III"} Type of sum of squares to use. **kwargs** scale : float Estimate of variance, If None, will be estimated from the largest model. Default is None. test : str {"F", "Chisq", "Cp"} or None Test statistics to provide. Default is "F". Notes ----- Use of this function is discouraged. Use anova_lm instead. testFscaleNtypr rzPR(>%s)dfsum_sqmean_sqcolumnsrI)ZII)ZIII)ZIVzType IV not yet implementedzType %s not understood)getlowerr endogexogshapeZ endog_namesdata design_info exog_nameslentermsrrnpzerosanova1_lm_singleanova2_lm_singleanova3_lm_singleNotImplementedErrorr str)r kwargsrrrr r#r$nobsZ response_namer'r(n_rowspr_testnamestablerrr anova_single#sD           r8c st|dd} | dkr2tj|\} } t| j|} ttjtj f} fddj D}t |D]\}}d| ||f<qht| | d}t }||}t j }||}|}t|dg|_tj| |d|f|j|dd gf<|j|jf|jdd dgf<|d krr|d |d|j|j||<tj|d |d|j||<tjtjf|jd||gf<|d |d|d <|S) a Anova table for one fitted linear model. Parameters ---------- model : fitted linear model results instance A fitted linear model **kwargs** scale : float Estimate of variance, If None, will be estimated from the largest model. Default is None. test : str {"F", "Chisq", "Cp"} or None Test statistics to provide. Default is "F". Notes ----- Use of this function is discouraged. Use anova_lm instead. effectsNcsg|]}|qSr)slice).0namer'rr |sz$anova1_lm_single..rrResidualrrrr)getattrr+linalgqrdotTr,r)r*Z column_names term_names enumeraterarraytolistrindexZc_sumlocssrdf_residrfsfnan)r r#r$r3r'r7r4rr5r r9qrZarrslicesiZslice_ridxrErIrr=rr-_s6    (  r-cCs|jdd}t|}dd||g}tt|df|d}t|d} t||} g} g} t|D]\} }||}t|j |j }g}t |j }|D]R}t |j }| |r||ks||}|t|j |j |t|j |j qt|jjjd|}t|jjjd|}|jrtt|| |j}ddlm}||\}}|jd|jd}t|dd| dfj|}n|}|jd}|d kr|j|| d }|j|j|j| |f<}|j|j|j| |f<||j|j| df<| |j | |q\t | d g|_|j!t"| |jjjddg}|||d|j#|j$}||d<|j#|j$tj%tj%f|jd dd||gf<|S) a Anova type II table for one fitted linear model. Parameters ---------- model : fitted linear model results instance A fitted linear model **kwargs** scale : float Estimate of variance, If None, will be estimated from the largest model. Default is None. test : str {"F", "Chisq", "Cp"} or None Test statistics to provide. Default is "F". Notes ----- Use of this function is discouraged. Use anova_lm instead. Type II Sum of Squares compares marginal contribution of terms. Thus, it is not particularly useful for models with significant interaction terms. Nrrr rrr)rArZcov_pr?)&r*rrr+r,rrFr:rstartstopsetfactorsissubsetextendeyer r$r%sizerCrDscipyrArBf_testfvaluerKrIpvalueappendr<rilocZargsortrLrMrP)r r'r4rr5r terms_infor6r7covZ robust_cov col_orderrIrTtermcolsL1ZL2Zterm_settZ other_setcolZLVLrAZ orth_compl_rRL12rN test_valuerLrrrr.s\        "   $r.cCsN|t|7}|j}dd||g}tt|df|d}t||} g} g} t|D]\} } || }t|j j j d|}|}|j d}|dkr|j || d}|j |j|j| |f<}|j|j|j| |f<||j|j| df<| | qNt| d g|_|||d|j|j}||d<|j|jtjtjf|jd dd||gf<|S) Nrrr rrrrrVr?)rr*rr+r,rrFr:r]r r$r%r`rarKrIrbrcr<rrLrMrP)r r'r4rr5r rer6r7rfrgrIrTrhrirjrnrRrNrorLrrrr/s6     r/c OsZ|dd}t|dkr,|d}t|f|S|dkrDtdt||dd}|dd }t|}d |}d d d d||g}tt|df|d} |s|dj}dd|D| d <dd|D| d <t | d j  | j | j dd d f<| d  | d<|dkrV| d| d || d<t j| d| d | d | |<tj| j | d|f<| S)a Anova table for one or more fitted linear models. Parameters ---------- args : fitted linear model results instance One or more fitted linear models scale : float Estimate of variance, If None, will be estimated from the largest model. Default is None. test : str {"F", "Chisq", "Cp"} or None Test statistics to provide. Default is "F". typ : str or int {"I","II","III"} or {1,2,3} The type of Anova test to perform. See notes. robust : {None, "hc0", "hc1", "hc2", "hc3"} Use heteroscedasticity-corrected coefficient covariance matrix. If robust covariance is desired, it is recommended to use `hc3`. Returns ------- anova : DataFrame When args is a single model, return is DataFrame with columns: sum_sq : float64 Sum of squares for model terms. df : float64 Degrees of freedom for model terms. F : float64 F statistic value for significance of adding model terms. PR(>F) : float64 P-value for significance of adding model terms. When args is multiple models, return is DataFrame with columns: df_resid : float64 Degrees of freedom of residuals in models. ssr : float64 Sum of squares of residuals in models. df_diff : float64 Degrees of freedom difference from previous model in args ss_dff : float64 Difference in ssr from previous model in args F : float64 F statistic comparing to previous model in args PR(>F): float64 P-value for significance comparing to previous model in args Notes ----- Model statistics are given in the order of args. Models must have been fit using the formula api. See Also -------- model_results.compare_f_test, model_results.compare_lm_test Examples -------- >>> import statsmodels.api as sm >>> from statsmodels.formula.api import ols >>> moore = sm.datasets.get_rdataset("Moore", "carData", cache=True) # load >>> data = moore.data >>> data = data.rename(columns={"partner.status" : ... "partner_status"}) # make name pythonic >>> moore_lm = ols('conformity ~ C(fcategory, Sum)*C(partner_status, Sum)', ... data=data).fit() >>> table = sm.stats.anova_lm(moore_lm, typ=2) # Type 2 Anova DataFrame >>> print(table) rrrrz6Multiple models only supported for type I. Got type %srrrNzPr(>%s)rMrLZdf_diffZss_diffrcSsg|] }|jqSr)rLr;Zmdlrrrr>qszanova_lm..cSsg|] }|jqSr)rMrrrrrr>rs)r!r)r8r r1rr+r,rZdiffvaluesrKrIrrNrOrPZisnull) argsr2rr rrZn_modelsr5r6r7rrranova_lms6F      & rucCs.tdg|}|D]}||}d||<q|S)NTF)r+rG)rSZslices_to_excludenindrhsrrr _not_slices  ryc Cs\t|||jd}||}t||dd|f|}|j|}t|t|}||fS)ah Residual sum of squares of OLS model excluding factors in `keys` Assumes x matrix is orthogonal Parameters ---------- y : array_like dependent variable x : array_like independent variables term_slices : a dict of slices term_slices[key] is a boolean array specifies the parameters associated with the factor `key` params : ndarray OLS solution of y = x * params keys : keys for term_slices factors to be excluded Returns ------- rss : float residual sum of squares df : int degrees of freedom rN)ryr%r+subtractrCrDr)) yx term_slicesparamskeysrwZparams1rLrMrrr_ssr_reduced_models  rc@s2eZdZdZd ddZddZddZd d ZdS) AnovaRMa Repeated measures Anova using least squares regression The full model regression residual sum of squares is used to compare with the reduced model for calculating the within-subject effect sum of squares [1]. Currently, only fully balanced within-subject designs are supported. Calculation of between-subject effects and corrections for violation of sphericity are not yet implemented. Parameters ---------- data : DataFrame depvar : str The dependent variable in `data` subject : str Specify the subject id within : list[str] The within-subject factors between : list[str] The between-subject factors, this is not yet implemented aggregate_func : {None, 'mean', callable} If the data set contains more than a single observation per subject and cell of the specified model, this function will be used to aggregate the data before running the Anova. `None` (the default) will not perform any aggregation; 'mean' is s shortcut to `numpy.mean`. An exception will be raised if aggregation is required, but no aggregation function was specified. Returns ------- results : AnovaResults instance Raises ------ ValueError If the data need to be aggregated, but `aggregate_func` was not specified. Notes ----- This implementation currently only supports fully balanced designs. If the data contain more than one observation per subject and cell of the design, these observations need to be aggregated into a single observation before the Anova is calculated, either manually or by passing an aggregation function via the `aggregate_func` keyword argument. Note that if the input data set was not balanced before performing the aggregation, the implied heteroscedasticity of the data is ignored. References ---------- .. [*] Rutherford, Andrew. Anova and ANCOVA: a GLM approach. John Wiley & Sons, 2011. NcCs||_||_||_d|kr"td||_|dk r8td||_|dkrRtjj |_ n||_ | |j |g|ds|j dk r| n d}t||dS)NCzSFactor name cannot be 'C'! This is in conflict with patsy's contrast function name.z)Between subject effect not yet supported!mean)ZsubsetzThe data set contains more than one observation per subject and cell. Either aggregate the data manually, or pass the `aggregate_func` parameter.)r&depvarwithinr betweenr0subjectpdZSeriesraggregate_funcequalsZdrop_duplicates _aggregate_check_data_balanced)selfr&rrrrrmsgrrr__init__s$   zAnovaRM.__init__cCs.|jj|jg|jdd|j|j|_dS)NF)Zas_index)r&groupbyrrrZaggrrrrrrs zAnovaRM._aggregatec Csd}|jD]}|t|j|9}q i}t|jjdD]T}g}|jD]}||j|j|qHt|}||kr||d||<q:d||<q:d}t||krt |||}|D]}|||krt |q|jjd||krt ddS)zraise if data is not balanced This raises a ValueError if the data is not balanced, and returns None if it is balance Return might change rrzData is unbalanced.z9There are more than 1 element in a cell! Missing factors?N) rr)r&uniqueranger%rcrdtupler ) rZ factor_levelsZwiZ cell_countrIkeyrl error_messagecountrrrrs*      zAnovaRM._check_data_balancedcCsf|j|jj}dd|jD}d|j}||g}tjd||jd}|jj }|D]4}t dg|j d}d|||<t |||<qTd |g} t || |j d}|d d |f}t||} | } | j|j dkrtd | D]} || q|D]}|||||<q| j} | j}| j}d d ddg}tjt d|d}|D]}|j|krF|dkrFt|||| |g\}}||}|||}|d |d dks|d ||kr||}|}n2t|||| |d |g\}}||}|||}||}tj|||}|dddd}||j|d f<||j|d f<||j|df<||j|df<qFt|S)zvestimate the model and compute the Anova table Returns ------- AnovaResults instance cSsg|] }d|qS) C(%s, Sum)r)r;rTrrrr>0szAnovaRM.fit..r*r&FrT:Nz$Independent variables are collinear.zF ValuezNum DFzDen DFzPr > F)rr rZ InterceptrqzC(z, Sum)) r&rrsrrpatsyZdmatrixjoinr'Zterm_name_slicesr+rGr%ryr fitZrankr popr~rMrLrrr,rrrNrOreplacerK AnovaResults)rr{rrrZr|r}rrwZ term_excluder resultsrTr~rMrLr anova_tableZssr1Z df_resid1Zdf1ZmsmZmseZdf2rprhrrrr&sv           z AnovaRM.fit)NNN)__name__ __module__ __qualname____doc__rrrrrrrrrs7 !rc@s(eZdZdZddZddZddZdS) rzX Anova results class Attributes ---------- anova_table : DataFrame cCs ||_dSN)r)rrrrrrqszAnovaResults.__init__cCs |Sr)summary__str__rrrrrtszAnovaResults.__str__cCs"t}|d||j|S)zlcreate summary results Returns ------- summary : summary2.Summary instance ZAnova)r ZSummaryZ add_titleZadd_dfr)rZsummrrrrws  zAnovaResults.summaryN)rrrrrrrrrrrrisr__main__)olsz moore.csvrZpartner_statusZ conformityZ fcategoryZfscore)Zskiprowsr6z5conformity ~ C(fcategory, Sum)*C(partner_status, Sum)rz#conformity ~ C(partner_status, Sum)r)r)&Zstatsmodels.compat.pythonrZnumpyr+Zpandasrrrrr_rZ statsmodels.formula.formulatoolsrrrZstatsmodels.iolibr Z#statsmodels.regression.linear_modelr rr8r-r.r/ruryrrrrZstatsmodels.formula.apirZread_csvZmoorerZmoore_lmZmooreBr7rrrrsB    <7X'j"A