:iOSSKrSSKrSSKrSSKJr SSKJr SSK J r J r SSK J r JrJr SSKJr "S S \5rS rg) N)issparse)tqdm)linksmaskers)DimensionErrorInvalidFeaturePerturbationErrorInvalidModelError) Explainerc|^\rSrSrSr\R SS4U4SjjrSr\ S5r \ S5r S r S r S rU=r$) LinearExplainera Computes SHAP values for a linear model, optionally accounting for inter-feature correlations. This computes the SHAP values for a linear model and can account for the correlations among the input features. Assuming features are independent leads to interventional SHAP values which for a linear model are ``coef[i] * (x[i] - X.mean(0)[i])`` for the ith feature. If instead we account for correlations, then we prevent any problems arising from collinearity and share credit among correlated features. Accounting for correlations can be computationally challenging, but ``LinearExplainer`` uses sampling to estimate a transform that can then be applied to explain any prediction of the model. Parameters ---------- model : (coef, intercept) or sklearn.linear_model.* User supplied linear model either as either a parameter pair or sklearn object. masker : function, numpy.array, pandas.DataFrame, tuple of (mean, cov), shap.maskers.Masker A callable Python object used to "mask" out hidden features of the form ``masker(binary_mask, x)``. It takes a single input sample and a binary mask and returns a matrix of masked samples. These masked samples are evaluated using the model function and the outputs are then averaged. As a shortcut for the standard masking using by SHAP you can pass a background data matrix instead of a function and that matrix will be used for masking. You can also provide a tuple of ``(mean, covariance)``, or pass in a masker meant for tabular data (i.e., :class:`.maskers.Independent`, :class:`.maskers.Impute`, or :class:`.maskers.Partition`) directly. data : (mean, cov), numpy.array, pandas.DataFrame, iml.DenseData or scipy.csr_matrix The background dataset to use for computing conditional expectations. Note that only the mean and covariance of the dataset are used. This means passing a raw data matrix is just a convenient alternative to passing the mean and covariance directly. nsamples : int Number of samples to use when estimating the transformation matrix used to account for feature correlations. feature_perturbation : None (default), "interventional" or "correlation_dependent" DEPRECATED: this option is now deprecated in favor of using the appropriate tabular masker and will be removed in a future release. There are two ways we might want to compute SHAP values, either the full conditional SHAP values or the interventional SHAP values. - For interventional SHAP values we break any dependence structure between features in the model and so uncover how the model would behave if we intervened and changed some of the inputs. This approach is used by the Independent and Partition maskers. - For the full conditional SHAP values we respect the correlations among the input features, so if the model depends on one input but that input is correlated with another input, then both get some credit for the model's behavior. This approach is used by the Impute masker. The interventional option stays "true to the model" meaning it will only give credit to features that are actually used by the model, while the correlation option stays "true to the data" in the sense that it only considers how the model would behave when respecting the correlations in the input data. For sparse case only interventional option is supported. Examples -------- See `Linear explainer examples `_ Nc >SU;a Sn[U5eUbSn[R"U[5 OSnUS;a Sn[ U5eXPl[ U[R5(dH[ U[R5(d[U5(aW[UR5S:Xa>UR S:Xa[R"U5nO[R "U5nO{[#[%U5[&5(a][U5S:XaNUR S:Xa[R"US US S .S S 9nO[R "US US S .5n[(TU]T"X4SU0UD6 X@l[.R1U5uUlUl[#[%UR65[R [R845(aSUlOE[#[%UR65[R5(aSUlO [;S5e[=UR6SS5n [ U [R5(a U R>n [=UR6SS5b8UR6R@Ul UR6RBUl!GOX[ U [D5(a[U 5S:XaU SUl [ UR@[RF5(aUR@R>Ul U SUl![ URB[R5(aURBR>Ul!GO[ U [&5(a[U 5S:XaU S Ul [ UR@[RF5(aUR@R>Ul U S Ul![ URB[R5(aURBR>Ul!OU c [S5e[U 5(aO[RH"[R@"U S 55S Ul UR S:wa [;S5eOh[RH"[R@"U S 55RK5Ul UR S:Xa[RB"U SS9Ul![UR@5(d2[M[%UR@55ROS5(a[RP"UR2UR@5UR4-Ul)[URR5S :XaURRSUl)Of[RH"URR5S Ul)O=[RP"UR2UR@5UR4-Ul)[UR@5Ul*UR S:XGa9[RV"[RX"URB5S:5S Ul-UR@URZUl URBSS2URZ4URZSS24Ul!UR2URZUl[]URB5uUl/n [R`"[R`"UR^URB5UR^Rb5Ul![R`"UR^UR@5Ul [R`"XR25Ul[RdRgURB5upU Ri5S:aBURB[Rj"URBRS 5S--Ul!URmU5up[R`"XR@5Ul7Xl8gUR S:XaUS:wa[R"S5 gg[ SUR -5e)Nfeature_dependencezGThe option feature_dependence has been renamed to feature_perturbation!zThe feature_perturbation option is now deprecated in favor of using the appropriate masker (maskers.Independent, maskers.Partition or maskers.Impute).interventional)rcorrelation_dependentzOfeature_perturbation must be one of 'interventional' or 'correlation_dependent'rrrr )meancovlinear)methodlinkz\The Linear explainer only supports the Independent, Partition, and Impute maskers right now!datarrz0A background data distribution must be provided!IOnly feature_perturbation = 'interventional' is supported for sparse dataF)rowvarzmatrix'>rr:0yE>gHz>gư>rzLSetting nsamples has no effect when feature_perturbation = 'interventional'!z/Unknown type of feature_perturbation provided: )9 ValueErrorwarningswarn FutureWarningr feature_perturbation isinstancepd DataFramenpndarrayrlenshaperImpute Independent issubclasstypetuplesuper__init__nsamplesr _parse_modelcoef interceptmasker PartitionNotImplementedErrorgetattrvaluesrrdictSeriesarrayflattenstrendswithdotexpected_valueMwherediag valid_indsduplicate_componentsavg_projmatmulTlinalgeigmineye_estimate_transformsmean_transformed x_transform)selfmodelr6rr2r#kwargsemsgwmsgrsum_proje_mean_transformrQ __class__s Z/srv/projetos/modelo_ml_acdoc/venv/lib/python3.13/site-packages/shap/explainers/_linear.pyr1LinearExplainer.__init__YsR 6 )\DT" "  +U  MM$ .#3 'R RdD1$7 7$8! fbll + +  + +x/?/?SEVZ[E[((,CC / ,,V4 V e , ,V1A((,CC 6!9(MV^_ ,,fQiq -RS  d4;;''*=*=w?P?P)Q R R(8D % T[[)7>> : :(?D %%n t{{FD1 dBLL ) );;D 4;; - 9 ((DI{{DH d # #D QV DI$))RYY// II,, E{DH$((BLL1188?? e $ $TaQDI$))RYY// II,, AwDH$((BLL1188?? \OP P~~HHRWWT1%56q9 ,,0@@-cA HHRWWT1%56>>@ ,,0GG!vvd59DH DII  #d499o"6"?"? "K"K#%&&DII">"OD 4&&'1,&*&9&9$&?#&(hht/B/B&CA&F#"$&&DII">"OD TYY  $ $(? ? hhrwwtxx'84'?@CDO $//2DIxx4?? 23DOOQ4FGDH $//2DI';488&D #DM8yy4==$((!CT]]__UDH $--;DI (II6DI99==*DAuuw~88bffTXX^^A->&?$&FF*.*C*CH*M 'N$&IInii$HD !*   & &*: :4 lm 2ADD]D]] c [UR5n[R"X"45n[R"X"45n[R"U[ S9n[ [U5S5GHMn[RRU5 [R"S5n[R"US45n[U5GHn XYn Un Un URSS2USU S-4nXSS24Rn [R"X5nURX4nU[R"URU 5- n[R"U S-U S-45nU S:a;U [R"X5U- -USS2SS24'U*U- =USS2S4'USSS24'SU- US'X:U 4==URU - ss'[R"URXYS-S[R"X5XYS-S5nX:USU S-4==U- ss'[R"URXYS[R"X5XYS5nX:USU 4==U-ss'XJU 4==URU - ss'XJUSU S-4==U- ss'XJUSU 4==U-ss'GM GMP X1-nXA-nX44$) a Uses block matrix inversion identities to quickly estimate transforms. After a bit of matrix math we can isolate a transform matrix (# features x # features) that is independent of any sample we are explaining. It is the result of averaging over all feature permutations, but we just use a fixed number of samples to estimate the value. TODO: Do a brute force enumeration when # feature subsets is less than nsamples. This could happen through a recursive method that uses the same block matrix inversion as below. dtypezEstimating transformsrrNr )rbrb)r)r4r'zerosarangeintrrangerandomshufflerrJrIouter)rRr2rCrZrQindsrY cov_inv_SiSicov_Sijicov_S cov_inv_SSdtZu coef_R_Sicoef_R_Ss r\rO$LinearExplainer._estimate_transformss  N1&)hhv& yy#&eHo'>?A II  d #88F+LXXq!f%F1XG) !T'AE]"23crc6N$$IIj,HHQTN !##q))!xxQA7 q5-7"((1.1:L-LL"crc*EFBFJL"b)LSbS,A'(1u V$!t$ ! 4$IIdiiUW &> &@_`dijejel`m@no $wQ-/0I=099TYYtBx%8"))E:VW[\^W_:`a$r({+x7+qD!TYYq\1!tGa!e},-:-tBQxK(H4(O @Z " **r^c[U[5(a[U5S:Xa USnUSnX4$[US5(a[US5(a{[URR 5S:a=URR SS:Xa URSnUR SnOURnUR nX4$[S[[U55-5e![a UR nN;f=f)zOAttempt to pull out the coefficients and intercept from the given model object.rrr coef_ intercept_z"An unknown model type was passed: ) r$r/r)hasattrryr*rz TypeErrorr r?r.)rSr4r5s r\r3LinearExplainer._parse_model"s eU # #E a8DaI"UG $ $ )E)E5;;$$%)ekk.?.?.Ba.G{{1~1 % 0 0 3I{{!,, $$H3tTY{K[$[\ \ !1 % 0 0I1s$C//DDc[U[R[R[R45(dg[ R U5 g![a gf=f)z+Determines if we can parse the given model.FT)r$rr,r7r+rr3 Exception)rSr6s r\supports_model_with_masker*LinearExplainer.supports_model_with_masker;sW&7#6#68I8I7>>"Z[[   ( ( /  sA A$#A$c[U5S:XdS5eUSn[UR5S:XaURSS5n[U[R [R 45(a URn[UR5S;a![S[UR535eURS:Xa[U5(a [S5e[R"[R"US S 2UR4URR 5UR"R 5UR$- n [R"XR5n [R&"U RSUR(45n XS S 2UR4'U n OURS :Xa[U5(aB[R*"[R,"XR.- UR055n O/[R*"XR.- 5UR0-n W R UR2URSS 4U R S S .$) z_Explains a single row and returns the tuple (row_values, row_expected_values, row_mask_shapes).r zEOnly single-argument functions are supported by the Linear explainer!rrbr r+Instance must have 1 or 2 dimensions! Not: rrNr)r:expected_values mask_shapes main_effects clustering)r)r*reshaper$r%r<r&r:rr#rr r'rIrFrHrJrQrPrcrCr=multiplyrr4rB) rR max_evalsr error_bounds batch_sizeoutputssilentrow_argsXphifull_phis r\ explain_rowLinearExplainer.explain_rowGs8}!j#jj! QK qww<1  !R A a"))R\\2 3 3A qwwEE   r^c [U[R[R45(a URn[ UR 5S;a![S[ UR 535eURS:Xa[U5(a [S5e[R"[R"USS2UR4URR5UR R5UR"- n[R"X R5n[R$"UR SUR&45nX#SS2UR4'U$URS:XGa[U5(a[ UR(R 5S:XaA[R*"[R,"XR.- UR(55$[R0"[3UR(R S5Vs/sHGn[R*"[R,"XR.- UR(U55PMI snS S 9$[ UR(R 5S:Xa/[R*"XR.- 5UR(-$[R0"[3UR(R S5Vs/sH5n[R*"XR.- 5UR(U-PM7 snS S 9$gs snfs snf) agEstimate the SHAP values for a set of samples. Parameters ---------- X : numpy.array, pandas.DataFrame or scipy.csr_matrix A matrix of samples (# samples x # features) on which to explain the model's output. Returns ------- np.array Estimated SHAP values, usually of shape ``(# samples x # features)``. Each row sums to the difference between the model output for that sample and the expected value of the model output (which is stored as the ``expected_value`` attribute of the explainer). The shape of the returned array depends on the number of model outputs: * one output: array of shape ``(#num_samples, *X.shape[1:])``. * multiple outputs: array of shape ``(#num_samples, *X.shape[1:], #num_outputs)``. .. versionchanged:: 0.45.0 Return type for models with multiple outputs changed from list to np.ndarray. rrrrNrrr rb)axis)r$r%r<r&r:r)r*rr#rr r'rIrFrHrJrQrPrcrCr4r=rrstackrf)rRrrrrns r\ shap_valuesLinearExplainer.shap_values|sg8 a"))R\\2 3 3A qww>88INtyy_`OaIbcIbA!ii-0499Q<?Ibcjl; pds /AM  >D_cGR>+@0  3 j@@r^rcz[R"S[R"[R"U55- 5n[R"[R"X5U5n[R"UR S[ S9*nSn[UR S5HqnSn[UR S5HPnX&S:dM [R"SXU4-XU4- XU4- 5S:dM>U(dUS- nSnX2U'MR Ms [R"[[R"U55UR S45nSUS '[SUR S5H nSXrUU4'M URURS5- RU4$) Nr rr`rbFrrTr)r'rEsqrtrIonesr*rerfabsrcr)uniquerJsum)CD componentscountrn found_grouprmprojs r\rGrGsc BGGBGGAJ''(A "))A/1%A''!''!*C00J E 1771:  qwwqz"A}q RVVAQ$K!qD',AAdG,K%Lt%S"QJE"&K %1 # 88S:./< =DDJ 1aggaj !!"]A " FFTXXa[ # #T ))r^)r numpyr'pandasr% scipy.sparser tqdm.autorrrutils._exceptionsrr r _explainerr rrGr^r\rs:! "kik\ *r^