a h@ @s" ddlZddlmZddlZddlZddlmZddlm Z m Z ddl m Z ddl mZmZmZmZddlmZddlmZmZmZmZmZmZdd lmZmZmZmZm Z m!Z!m"Z"dd l#m$Z$m%Z%m&Z&dd l'm(Z(m)Z)m*Z*m+Z+m,Z,dd l-m.Z.dd l/m0Z0ddl1m2Z2m3Z3m4Z4m5Z5m6Z6ddl7m8Z8ddl9m:Z:m;Z;mZ>ddl?m@Z@mAZAddlBmCZCmDZDmEZEmFZFmGZGmHZHgdZIdZJdZKe LZMeMjNeMjOZPZQeRePjSdZTejUVdZWeWXeTeTddZTePeTeQeTZPZQe YZZeZjNeZjOZ[Z\ddZ]ddZ^ej_ddgdddZ`ejabd eIejabd!d"d#gd$d%Zcejabd eIejabd!d"d#gd&d'Zdejabd eIejabd!d"d#gd(d)Zeejabd eIejabd!d"d#gd*d+Zfejabd eIejabd!d"d#gd,d-Zgejabd eIejabd!d"d#gd.d/Zhejabd eIejabd!d"d#gejabd0dgeFejabd1d2d3gd4d5Zid6d7Zjd8d9Zkd:d;Zldd?Znd@dAZoejabdBgdCejabdDeFdEdFZpejabdGgdHejabdId"d#gejabdDeFdJdKZqejabdGgdHejabdId"d#gejabdDeFdLdMZrddTdUZsejabdVdWdXegdYdgeFDejabdZgd[ejabd\eRd]d^d_Ztejabd`dadbgejabdcejugeFejabdddedfgejabd!d"d#gejabdggdhdidjZvdkdlZwejabd`dadbgejabdcejugeFejabdmdndogejabdpgdqdrdsZxejabd0dgeFejabdtgdudvdwZydxdyZzdzd{Z{ejabd|ed#d}efed#d}efgd~dZ|ejabd|eefeefgejabddd]gddZ}ddZ~ddZddZddZejabddde]gejabdde*dgejabd0dgeFddZejabdde*dgejabd0dgeFddZddZddZejabde5ejajbdeAege8dejajbdedadge8dddZejabde6d#dddZejabdee2ddZejabdeze{eeeefejabdDeFddZddZejabdeefddZddZejabddde^gddZejabddde]gddZejabdeegddZejabddd]gejabdeegddZddZddZejabddd]gd]dggejabd0eDeEeFeGeHddZddZejabdeegejabdddiedfddiedfddiedfgddʄZejabdeegdd̄Zdd΄ZejadϡddфZejabd gdҢejabdd"d#gejabdDeFddՄZejabd gd֢ejabdDeFdd؄Zejabdd"d#gejabdDeFddڄZejabdd#d"gejabddedݡgejabdejgeFejabd gdߢddZejabd gdddZddZejabd gdejabd\edQddZddZejabdeifeddifedd]ifgddZejabd ddgejabd!d"d#gejabd1gdddZejabd!d"d#gejabd1gdddZejabd gdddZejabd1gdddZejabd1gdddZddZejabd!d#d"gejabd0dgeFejabdddgejabd eIdgddZddZddZejabdd#d"gejabd!d#d"gejabddQdgdd Zd d Zd d Zejabdeege d"dddZejabdeefeefge d"dddZdS(N)product)linalg)config_contextdatasets)clone)make_classificationmake_low_rank_matrixmake_multilabel_classificationmake_regression)ConvergenceWarning)LinearRegressionRidgeRidgeClassifierRidgeClassifierCVRidgeCVridge_regression)_check_gcv_mode _RidgeGCV_solve_cholesky_solve_cholesky_kernel _solve_lbfgs _solve_svd_X_CenterStackOp) get_scorer make_scorermean_squared_error) GridSearchCV GroupKFoldKFold LeaveOneOutcross_val_predict) minmax_scale)check_random_state)_NUMPY_NAMESPACE_NAMES_atol_for_type_convert_to_numpy)yield_namespace_device_dtype_combinationsyield_namespaces)_get_check_estimator_ids)assert_allcloseassert_almost_equalassert_array_almost_equalassert_array_equalignore_warnings)_array_api_for_tests check_array_api_input_and_values) _IS_32BITCOO_CONTAINERSCSC_CONTAINERSCSR_CONTAINERSDOK_CONTAINERSLIL_CONTAINERS)svd sparse_cgcholeskylsqrsagsaga)r7r:)r7r8r9r:r;cKst||kSN)npmeany_testy_predkwargsrDa/var/www/html/assistant/venv/lib/python3.9/site-packages/sklearn/linear_model/tests/test_ridge.py_accuracy_callableXsrFcCs||dS)N)r?)rArBrDrDrE_mean_squared_error_callable\srHlongZwide)paramscCs|jdkrd\}}nd\}}t||}tj|}t||||d}d|dddf<t|\}}} t|dkstJ|ddd|f|dd|df} } | d|ddf| |dddf} } |jdkr |j d d |d }||}|| |j ||d d 7}n.|j d d |d }| j t d|| j |}d}|t |}d|d<t|j |||j |}|||}|||}tj|tj|ksJ||||fS)aDDataset with OLS and Ridge solutions, well conditioned X. The construction is based on the SVD decomposition of X = U S V'. Parameters ---------- type : {"long", "wide"} If "long", then n_samples > n_features. If "wide", then n_features > n_samples. For "wide", we return the minimum norm solution w = X' (XX')^-1 y: min ||w||_2 subject to X w = y Returns ------- X : ndarray Last column of 1, i.e. intercept. y : ndarray coef_ols : ndarray of shape Minimum norm OLS solutions, i.e. min ||X w - y||_2_2 (with minimum ||w||_2 in case of ambiguity) Last coefficient is intercept. coef_ridge : ndarray of shape (5,) Ridge solution with alpha=1, i.e. min ||X w - y||_2_2 + ||w||_2^2. Last coefficient is intercept. rI) )rLrK) n_samples n_featuresZeffective_rank random_stateNMbP? lowhighsizerXrGr)rQrQ)paramminr>random RandomStaterrr6alluniformnormalTZdiagidentityZsolvenorm)global_random_seedrequestrMrNkrngXUsZVtZU1ZU2ZVt1_Zcoef_olsyalphadZ coef_ridgeZR_OLSZR_RidgerDrDrEols_ridge_dataset`s6    **   rosolver fit_interceptTFcCs|\}}}}d}t|d||dvr$dnd|d} |t|} |||} dt| dt| d} tfi| } |d d d d f}|r|d }n ||jd d }||}d }| |||d d }| jt|ksJt | j || ||t| ks Jtfi| j||t |j d d } | jt|ksHJt | j || ||t| kspJ| j|ksJd S)zTest that Ridge converges for all solvers to correct solution. We work with a simple constructed data set with known solution. ?Tr:r;V瞯<绽|=rmrqrptolrOrPrGNrQraxis sample_weight)dictr>r?sumr fit intercept_pytestapproxr)coef_scoreonesshapeZsolver_)rprqrordrhrlrkcoefrmrJZres_nullZ res_RidgeZR2_Ridgemodel interceptrDrDrEtest_ridge_regressions:         & rc Cs|\}}}}|j\}} d} t| d|||dvr2dnd|d} |ddddf}d tj||fd d }tj|t|| d ksJ|r|d} n ||jd d }||}d } | |||dd}| j t | ksJt | j tj||fd ddS)a Test that Ridge converges for all solvers to correct solution on hstacked data. We work with a simple constructed data set with known solution. Fit on [X] with alpha is the same as fit on [X, X]/2 with alpha/2. For long X, [X, X] is a singular matrix. rrrGrsrtrurvNrQ?rPrxr:0yE>atol)rr r> concatenater matrix_rankr[r?r~rrrr)rr_ rprqrordrhrlrkrrMrNrmrrrDrDrE test_ridge_regression_hstacked_Xs,      rc Cs|\}}}}|j\}} d} td| |||dvr2dnd|d} |ddddf}tj||fd d }tj|t|| ks|Jtj||f}|r|d} n ||jd d }||}d } | |||dd}| j t | ksJt | j|d d dS) aJTest that Ridge converges for all solvers to correct solution on vstacked data. We work with a simple constructed data set with known solution. Fit on [X] with alpha is the same as fit on [X], [y] [X], [y] with 2 * alpha. For wide X, [X', X'] is a singular matrix. rrrGrsrtrurvNrQrrxrr)rr r>rrrr[rr?r~rrrr)rrrDrDrE test_ridge_regression_vstacked_Xs.      rcCs<|\}}}}|j\}} d} t| |||dvr.dnd|d} tfi| } |rt|ddddf}|d} |dd}nd} | |||| ks|s| jt| ksJt| j|nt| ||t||| |t j t j | j| jft j t j | |fks Jtjdd | jt| ks,Jt| j|dS) aTest that unpenalized Ridge = OLS converges for all solvers to correct solution. We work with a simple constructed data set with known solution. Note: This checks the minimum norm solution for wide X, i.e. n_samples < n_features: min ||w||_2 subject to X w = y rrsrtrurvNrQ1Ridge does not provide the minimum norm solution.reason)rr|r r~rrrr)rpredictr>rrcrxfail)rprqrordrhrlrrkrMrNrmrJrrrDrDrE!test_ridge_regression_unpenalized's8      rc Csr|\}}}}|j\}} d} t| |||dvr.dnd|d} |rf|ddddf}|d} |dd}nd} dtj||fd d }tj|t|| ksJ| |||| ks|s| jt | ksJ|d krt t | j tj||fnt | ||tjtj| j| j ftjtj| ||fks6Jt jd d | jt | ksXJt | j tj||fdS)a^Test that unpenalized Ridge = OLS converges for all solvers to correct solution. We work with a simple constructed data set with known solution. OLS fit on [X] is the same as fit on [X, X]/2. For long X, [X, X] is a singular matrix and we check against the minimum norm solution: min ||w||_2 subject to min ||X w - y||_2 rrsrtrurvNrQrrPrxr8rr)rr r>rrrr[r~rrrskipr)rrrrcr rprqrordrhrlrrkrMrNrmrrrDrDrE,test_ridge_regression_unpenalized_hstacked_X^s<      rc CsV|\}}}}|j\}} d} t| |||dvr.dnd|d} |rf|ddddf}|d} |dd}nd} tj||fdd}tj|t|| ksJtj||f}| |||| ks|s| j t | ksJt | j |ntt | ||tjtj| j | j ftjtj| |fks$Jt jd d | j t | ksFJt | j |dS) aTest that unpenalized Ridge = OLS converges for all solvers to correct solution. We work with a simple constructed data set with known solution. OLS fit on [X] is the same as fit on [X], [y] [X], [y]. For wide X, [X', X'] is a singular matrix and we check against the minimum norm solution: min ||w||_2 subject to X w = y rrsrtrurvNrQrxrr)rr r>rrrr[rr~rrrr)rrrcrrrDrDrE,test_ridge_regression_unpenalized_vstacked_Xs:      rsparse_containerrmrr{Gz?cCsD|dur2|r|tvrtn|s2|tvr2t|\}}}} |j\} } tjdd| d} t||||dvrldndd|d } |dddd f}tj ||fdd }tj ||f}tj | d| f|} |r| d }n ||j dd }|| }d}|dur||}| j ||| d | dd } | j t|ks4Jt| j| dS) zTest that Ridge with sample weights gives correct results. We use the following trick: ||y - Xw||_2 = (z - Aw)' W (z - Aw) for z=[y, y], A' = [X', X'] (vstacked), and W[:n/2] + W[n/2:] = 1, W=diag(W) NrrPrUrsrtru順)rmrqrprwmax_iterrOrQrxrz)SPARSE_SOLVERS_WITH_INTERCEPTrr SPARSE_SOLVERS_WITHOUT_INTERCEPTrrgr_r r>rrr?r~rrr)r)rprqrrmrordrhrlrkrrMrNswrrrDrDrE$test_ridge_regression_sample_weightss>         rcCsXtdd}tt|dgd}tttj}t||dgd}ttj|j}t||dS)NrQrPrrm) y_diabetesreshaper X_diabetesr>dotrarr+)rlrKZ dual_coefZcoef2rDrDrEtest_primal_dual_relationships  rc Csntjd}|d}|dd}d}tjt|d&t||ddddd d Wdn1s`0YdS) NrrTz3sparse_cg did not converge after [0-9]+ iterations.matchrrr7rP)rmrprwrverbose)r>r\r]randnrwarnsr r)rgrlrhZwarning_messagerDrDrE&test_ridge_regression_convergence_fail s   rcCsVtjd}d\}}|||}||}|ddtjf}tj|d|f}t}||||jj |fksrJ|j j dksJt |jtj sJt |j t sJ||||jj |fksJ|j j dksJt |jtj sJt |j tj sJ||||jj d|fksJ|j j dks*Jt |jtj s>Jt |j tj sRJdS)NrrrTrPrDrPrG)rG)r>r\r]rnewaxisc_r r~rrr isinstanceZndarrayfloat)rgrMrNrhrlZY1YridgerDrDrEtest_ridge_shapes_types,      rcCstjd}d\}}|||}||}tj|d|f}t}||||j}|||t|jd|t|jd|ddS)NrrrrrP) r>r\r]rrr r~rr*)rgrMrNrhrlrrrrDrDrEtest_ridge_intercept4s     rcCstjd}d\}}||}|||}tddd}tdd}||||||t|j|j||||||t|j|jdS)Nr)rrLrFrmrqrq) r>r\r]rr r r~r*r)rgrMrNrlrhrZolsrDrDrEtest_ridge_vs_lstsqFs         rc stjd}d\}}}||||||t|tfddtjD}fdddD}|D]}t||qrt ddd}d }t j t |d | Wdn1s0YdS) N*)rTrcs&g|]\}}t|dd|jqS)r8rmrpr r~r).0rmtarget)rhrDrE gsz3test_ridge_individual_penalties..cs$g|]}t|ddjqS)-q=)rmrprwr)rrprhZ penaltiesrlrDrErms)r6r7r9r8r:r;rQrzCNumber of targets and number of penalties do not correspond: 4 != 5r)r>r\r]rarangearrayziprar+r rraises ValueErrorr~) rgrMrN n_targets coef_choleskyZcoefs_indiv_penZcoef_indiv_penrerr_msgrDrrEtest_ridge_individual_penalties[s&        rn_col)rDr) csr_containerc Cstjd}|dd}|d}|t|}|jdg|R}|jdg|R}t||||}t||dddf||dddfg} t| |||t| j ||j |dS)Nr  ) r>r\r]rlenrZhstackr)rra) rrrgrhZX_msqrt_swrAoperatorZreference_operatorrDrDrEtest_X_CenterStackOp{s   .rr))rTrP) r)r)rGrG)rruniform_weightscCstjd}|j|}|r,t|jd}n|d|d}t|}tj|d|d}|||dddf}| |j } |||dddf} t dd} | | |\} } t || t | | dSNrrP)ryweightsTr)r>r\r]rrr chisquaresqrtaveragerrarZ _compute_gramr))rrrrgrhrrX_mean X_centeredZ true_gramX_sparsegcvZ computed_gram computed_meanrDrDrEtest_compute_grams      rcCstjd}|j|}|r,t|jd}n|d|d}t|}tj|d|d}|||dddf}|j |} |||dddf} t dd} | | |\} } t || t | | dSr)r>r\r]rrrrrrrarrZ_compute_covariancer))rrrrgrhrrrrZtrue_covariancerrZ computed_covrrDrDrEtest_compute_covariances      rdrrTrP*@>@c  Cst|||||||d| d \} } }|dkr4t|g}| |7} tj| d|| jdk}| }d| |<d||<| ||8} | r| | t |d|7} t |d}|dkr|d}| r| | |fS| | fS)NT) rMrN n_informativerbiasnoiseshufflerrOrPrr) r r>asarrayr\r]Zbinomialrcopyrabs)rMrNproportion_nonzerorrrX_offsetrrrpositiverOrhrlcmaskZ removed_XrDrDrE_make_sparse_offset_regressions8    rzsolver, sparse_containerccs*|]"\}}|dus|dvr||fVqdS)N)r7ridgecvrD)rrprrDrDrE sr)r8r:r7r9r;rz"n_samples,dtype,proportion_nonzero))rfloat32皙?)(rrr)rfloat64皙?seedrc Csd}|dkrdnd}tdd||||d\}} t|}td|d || } |j|d d }| j|d d } |durt||}|d krt|gd } nt|d|d} | || t| j| jdddt| j| jddddS)Nrrg?gI@g@@rT)rrNrrrOrMr6)rprmF)rralphasru)rprwrmrRrrtol) rr!r r~astyperr)rr) rprrMdtyperrrmrrhrlZ svd_ridgerrDrDrEtest_solver_consistencys,  r gcv_moder6eigen X_containerX_shape)rr)rrzy_shape, noise))rrr)rrPr)rrb@c Cs|\}}t|dkr|dnd}t|||dd|dd\} } | |} gd} t||| d d } t||| d } | | | || }| || | jt| jksJt| j | j d d t| j | j d d dS)NrGrQrPrFrrMrNrrOrrrrRrrr$@g@@neg_mean_squared_errorcvrqrscoring)rrqrrRr ) rrrrr~alpha_rrr)rr)rrry_shaperqrrMrNrrhrlr loo_ridge gcv_ridgeX_gcvrDrDrEtest_ridge_gcv_vs_ridge_loo_cvs<   r#c Csd}d\}}d}t|||ddddd\}}gd}t|d ||d }td ||d }|||||||jt|jksJd |jd |jt|j|jddt|j|jdddS)NZexplained_variance)rTrrPrFrrrTr)rqrrzgcv_ridge.alpha_=z, loo_ridge.alpha_=rRr) rrr~rrrr)rr) rrMrNrrhrlrr r!rDrDrEtest_ridge_loo_cv_asym_scoringEs4   r$rNrrzy_shape, fit_intercept, noise))rTrr)rTg4@)rTr)rFrcs$gd}tjd}t|dkr(|dnd}td||dd|d\} } | |} d |t| } | | dt } t t | j d| | t } | | } } t| j dd }|j| | d }t||d |d }|| | t|j|d}|j| | d }t|| | |d}|j | j kr.|| j }| |dfddt | j dDt|| }t|d||d}|j|| | dt|dkr|jdddd||jf}n|jdd||jf}|jt|jksJt|ddt|j|jddt|j|jdddS)NrrrGrQrPrF)rMrNrrOrrr)Zn_splits)groupsr)rrrrqrrcs"g|]}tj|kddqS)rrx)r>r})riindicesZ kfold_errorsrDrErsz1test_ridge_gcv_sample_weights..T)rstore_cv_resultsrrqrzrRr)r>r\r]rrrrr[r intrepeatrrrrsplitrr~r rr r cv_results_indexrrr)rr)rrrqrNrrrrgrrhrlr{ZX_tiledZy_tiledrZsplitsZkfoldZ ridge_reg predictionsr"r!Z gcv_errorsrDr(rEtest_ridge_gcv_sample_weightsesf         "r1z2mode, mode_n_greater_than_p, mode_p_greater_than_n))Nr6r)autor6r)rrr)r6r6r6cCsJtddd\}}|dur ||}t|||ks2Jt|j||ksFJdS)NrrG)rMrN)r rra)rmodeZmode_n_greater_than_pZmode_p_greater_than_nrhrkrDrDrEtest_check_gcv_mode_choices r4cCstjd}g}|dur"td}}n|td}}t|d}||t|j}||t}tt dd}t d|d} || j|t| jt |ksJdd} t| }t d|d} || j|t| jt |ksJt d } t d| d} | |t| jt |ks J|durB|j|tt|d |jt |ksBJtttfj}|||||}||t||}tt||fj|d d |S) NrTFr)Zgreater_is_better)rqrcSs t|| Sr=)r)xrlrDrDrEfuncsz_test_ridge_loo..funcrrzh㈵>r)rrrr~rrappendr-rrrrrrr>rvstackrarr))rrMretrhrqZ ridge_gcvrfrZ ridge_gcv2r6Z ridge_gcv3ZscorerZ ridge_gcv4rY_predrBrDrDrE_test_ridge_loosB               r=cCs|dur tn|t}t}||t||t|jjdksDJt|j t j usXJt d}|j |d||t||t|jjdksJt|j t j usJdS)NrPrr&)rrr~rrrrrtyperr>rr set_params)rrhridge_cvrrDrDrE_test_ridge_cvs     rAzridge, make_datasetr*cCs.|ddd\}}|||t|dr*JdS)NrrMrOr.)r~hasattr)r make_datasetrhrlrDrDrE$test_ridge_gcv_cv_results_not_storeds  rGrcCsL|ddd\}}|jd|d|||t|ds8Jt|jtsHJdS)NrCrrDF)r*r best_score_)r?r~rErrHr)rrFrrhrlrDrDrEtest_ridge_best_scores  rIc stjd}d\}}}|||}t|dddgftd|ft|dddgfdtd|ft|dddgfdtd|f|||dfd d |jD}td d |}t ||j t t |j d |j |j td d d|}|j j|fks$J|jj|fks8J|jj|t|fksTJtdd d d|}|j j|fks~J|jj|fksJ|jj||dfksJtd d d|dddf}t|j sJt|jsJ|jj|tfks Jtd dd|}t ||j t t |j d |j |j ttd d}d}tjt|d||Wdn1s0Ytdd d}tjt|d||Wdn1s0YdS)Nr)rrrrrPg?rGrR)rPrcs g|]}td|jqS)r)rr~r)rrrhrrDrEr=z6test_ridge_cv_individual_penalties..T)ralpha_per_targetr)rrMr*r2)rrMr)rrrMz3cv!=None and alpha_per_target=True are incompatiblerrC)r>r\r]rrrrarr~r,rr+r rrrHr.rZisscalarrrrr)rgrMrNrrlZoptimal_alphasr@msgrDrKrE"test_ridge_cv_individual_penalties*s\   "&&   ,rPcCs>|dur tn|t}tdd}||tt||tdS)NFrr)rr r~rr>roundr)rrhrrDrDrE_test_ridge_diabetesqs  rRcCs|dur tn|t}tttfj}tjd}tdd}||||jjd|fksXJ| |}||t| |}t t||fj|dddS)NrPFrrGrdecimal) rr>r9rrarr r~rrr+)rrhrrNrr<rBrDrDrE_test_multi_ridge_diabetesxs      rUcCsttjd}tjd}|dur&tn|t}ttfD]D}||t|jj||fks^J| |}t t|kdks:Jq:t d}t|d}||t| |}t t|kdksJdS)NrrPgHzG?rr&g?) r>uniquey_irisrX_irisrrr~rrr?r)r n_classesrNrhregrBrrDrDrE_test_ridge_classifierss      r[rZaccuracyrcCsJ|dur tn|t}t|r$t|n|}t||d}||t|dS)N)rr)rXcallablerrr~rWr)rrrrhscoring_clfrDrDrE"test_ridge_classifier_with_scorings r_cCszdd}|durtn|t}tjdddd}t|t||d}||t|jt dks^J|j t |d ksvJdS) Nc[sdS)NzG?rDr@rDrDrE _dummy_scoresz:test_ridge_regression_custom_scoring.._dummy_scorerGr)num)rrrr`r) rXr>Zlogspacerrr~rWrHrrr)rrrarhrr^rDrDrE$test_ridge_regression_custom_scorings rdcCsl|dur tn|t}tddd}||t||t}tddd}||t||t}||kshJdS)Nr7F)rwrqrR)rr r~rr)rrhrrZridge2Zscore2rDrDrE_test_tolerances      recCs t||}t|}t|}|j||d}|j||d} ||||j} |j} tddt ||| } | j} | j dksJ| j |j ksJt t | |d| t|d| j}|j dksJ|j |j ksJt t ||d| t|dWdn1s0YdS)NdeviceTZarray_api_dispatch)rL)xprrD)r.rXr rWrr~rrrrrr r)r%r$)name estimatorarray_namespacerg dtype_nameriZ X_iris_npZ y_iris_np X_iris_xp y_iris_xpZcoef_npZ intercept_npZ estimator_xpZcoef_xpZ intercept_xprDrDrEcheck_array_api_attributess4       rpz#array_namespace, device, dtype_namecheck)ZidsrkrpcCs|jj}||||||ddS)N)rgrm) __class____name__)rkrqrlrgrmrjrDrDrEtest_ridge_array_api_compliancesrurl)Zinclude_numpy_namespacesc Cst|dd}|tdd}|tdd}tjddj}|ddhD]}t||dkd}d |jd |d }tj t |d Ft d d| ||Wdn1s0YWdqL1s0YqLtdd d}d|jd}tj t |d Ht d d| ||Wdn1s20YWdn1sR0Yt}d|jd}tj t|d Ht d d| ||Wdn1s0YWdn1s0YdS)Nrfrrprr2r6lbfgsrprz Array API dispatch to namespace z" only supports solver 'svd'. Got 'z'.rTrhzYThe solvers that support positive fitting do not support Array API dispatch to namespace zc. Please either disable Array API dispatch, or use a numpy-like namespace, or set `positive=False`.z&Using Array API dispatch to namespace z with `solver='auto'` will result in using the solver 'svd'. The results may differ from those when using a Numpy array, because in that case the preferred solver would be cholesky. Set `solver='svd'` to suppress this warning.)r.rrXrWr Z_parameter_constraintsoptionsrtrrrrr~r UserWarning)rlrirnroZavailable_solversrpr expected_msgrDrDrE6test_array_api_error_and_warnings_for_solver_parameters6   J  L  r{c Cst|dd}|tdd}|tdd}t}d}tVtjd|tdt dd| ||Wdn1s0YWdn1s0Yt dd$tddd  ||Wdn1s0YdS) NrfrzkResults might be different than when Array API dispatch is disabled, or when a numpy-like namespace is usederror)messagecategoryTrhr2rw) r.rrXrWr warningscatch_warningsfilterwarningsryrr~)rlrirnrorrzrDrDrE)test_array_api_numpy_namespace_no_warning+s   H r test_funccCs2|d}||}|dur.|dur.t||dddS)NrrS)r+)rrZ ret_denseZ ret_sparserDrDrEtest_dense_sparseCsrcCsZtddgddgddgddgddgg}gd}tdd}|||t|ddggtd gtd d id}|||t|ddggtd gtd d}|||t|ddggtd gtddgddgddgddgg}gd }tdd}|||td d}|||t|jdks:Jt|j |j t|j |j dS)Nr皙rrrrPrPrPrQrQ class_weightrrPrRrQbalanced)rPrPrQrQrG) r>rrr~r,rrZclasses_r+rr)rhrlrZZregarDrDrEtest_class_weightsYs((     "    rrZcCs|}|tjtj|dd}|tjtjt|j|jttjj}|tjdkd9<dddd}|}|tjtj|||d}|tjtjt|j|j|}|tjtj|d||d}|tjtj|t|j|jd S) z5Check class_weights resemble sample_weights behavior.rrrPrrrgY@)rrPrGrGN) r~irisdatarr*rr>rr)rZreg1reg2r{rrDrDrE"test_class_weight_vs_sample_weight|s$    rcCstddgddgddgddgddgg}gd}tdgdd}|||td d igd d}|||t|d d ggtdgdS)Nrrrrrrr)rrrP)rrrPrR)rrrPrTgɿrGrQ)r>rrr~r,r)rhrlrZrDrDrEtest_class_weights_cvs(  rrc Cstjd}d}d}|||}gd}t|}t|r@t|n|}t|dd|d}||} ||| |j j ||fks~Jd} ||| } ||| |j j || |fksJtdd|d}t j t d d ||| Wdn1s0YdS) NrrrrrrrTrrr*rr)rr*rzcv!=None and store_cv_resultsr)r>r\r]rrr\rrr~r.rrrr) rrgrMrNr5rn_alphasr]rrlrrDrDrEtest_ridgecv_store_cv_resultss$      rc Cstddgddgddgddgddgg}tgd}|jd}gd}t|}t|r`t|n|}t|dd|d }d }||||jj|||fksJtgdgd gd g }|jd }||||jj|||fksJdS) NrrrrrrrrTrrP)rPrQrPrQrP)rQrQrPrQrQ) r>rrrr\rrr~r.Z transpose) rr5rlrMrrr]rrrDrDrE)test_ridge_classifier_cv_store_cv_resultss$(    r EstimatorcCstjd}d}d\}}|tur,||}n|dd|}|||}||d}|j|uslJd|jd|||t |jt |dS)NrrrrrGrz`alphas` was mutated in `z .__init__`) r>r\r]rrrandintrrtr~r,r)rrgrrMrNrlrh ridge_estrDrDrEtest_ridgecv_alphas_conversions      rc Cstjd}d}d\}}|tur,||}n|dd|}|||}|||d}|durtjtdd| ||Wdq1s0Yn | ||dS) z1Check alpha=0.0 raises error only when `cv=None`.r)rrrrrrGrrNz"alphas\[0\] == 0.0, must be > 0.0.r) r>r\r]rrrrrrr~) rrrgrrMrNrlrhrrDrDrEtest_ridgecv_alphas_zeros    ,rc Cstjd}d}dD]\}}||}|||}d||}td}t||d}|j|||dd|i} tt | |d } | j|||d|j | j j ksJt |j| j jqdS) Nrr)rCrrrrrrrzrmr&)r>r\r]rrandrrr~rr rZbest_estimator_rmr+r) rgrrMrNrlrhr{rr parametersgsrDrDrEtest_ridgecv_sample_weights     rc sVddg}ddg}tjd}t||D](\}}|||||||dd}d}d}|ddtjf|tjddftdd|||fdd }fd d } d } tj t | d |Wdn1s 0Yd } tj t | d | Wdq&1sF0Yq&dS)NrGrrrPrrg@rcsdSr=r~rD)rhrsample_weights_not_OKrlrDrEfit_ridge_not_ok@szStest_raises_value_error_if_sample_weights_greater_than_1d..fit_ridge_not_okcsdSr=rrD)rhrsample_weights_not_OK_2rlrDrEfit_ridge_not_ok_2CszUtest_raises_value_error_if_sample_weights_greater_than_1d..fit_ridge_not_ok_2z)Sample weights must be 1D array or scalarr) r>r\r]rrrr r~rrr) Z n_samplessZ n_featuressrgrMrNZsample_weights_OKZsample_weights_OK_1Zsample_weights_OK_2rrrrD)rhrrrrlrE9test_raises_value_error_if_sample_weights_greater_than_1d(s.    &rzn_samples,n_featuresrGc Cstjd}tddd}tddd}|||}||}||dd}||} |j| ||d|j|||dt|j|jdd dS) NrrrFrrGrPrzrCrS)r>r\r]r rr~r+r) rMrNrrg sparse_ridge dense_ridgerhrlZsample_weightsrrDrDrE&test_sparse_design_with_sample_weightsOs     rcCsJtddgddgddgddgddgg}gd}tdd}|||dS) Nrrrrrrr)rPrTrr)r>rrr~)rhrlrrDrDrEtest_ridgecv_int_alphases( rzparams, err_type, err_msgr)rPrQiz alphas\[1\] == -1, must be > 0.0)grg$z"alphas\[0\] == -0.1, must be > 0.0)rPrr1z1alphas\[2\] must be an instance of float, not strcCsjd\}}t||}tdd|}tj||d&|fi|||Wdn1s\0YdS)z?Check the `alphas` validation in RidgeCV and RidgeClassifierCV.rrrGrN)rgrrrrr~)rrJZerr_typerrMrNrhrlrDrDrEtest_ridgecv_alphas_validationns  rcCsLd\}}t||}|tur(t|}ntdd|}|dd||dS)zCheck the case when `alphas` is a scalar. This case was supported in the past when `alphas` where converted into array in `__init__`. We add this test to ensure backward compatibility. rrrGrPrN)rgrrrr~)rrMrNrhrlrDrDrEtest_ridgecv_alphas_scalars   rcCs6tddd}|tt|jjdtjdks2JdS)Nr7rP)rprr)r r~rrrr)rZrDrDrEtest_sparse_cg_max_iters  rz-ignore::sklearn.exceptions.ConvergenceWarningcCsd}tt}}t||dfj}tddD]<}dD]2}t||dd}|||t|j t||q2q*dD],}t|ddd}||||j duslJqldS) NrGrPrL)r:r;r9r)rprrw)r7r6r8r) rrr>Ztileraranger r~r,Zn_iter_)rrhrlZy_nrrprZrDrDrE test_n_iters   r)r9r7rvr2with_sample_weightc Cs|dk}td||d\}}d}|rDtj|}d|j|jdd}|dkrPd n|} t| d |d } t|d |d } | j|||d | j||||d t| j | j t| j | j d ddS)aCheck that ridge finds the same coefs and intercept on dense and sparse input in the presence of sample weights. For now only sparse_cg and lbfgs can correctly fit an intercept with sparse X with default tol and max_iter. 'sag' is tested separately in test_ridge_fit_intercept_sparse_sag because it requires more iterations and should raise a warning if default max_iter is used. Other solvers raise an exception, as checked in test_ridge_fit_intercept_sparse_error rvr)rNrOrNrrrrYr2r7r)rprwrrzgƠ>r) rr>r\r]r_rr r~r)rr) rprrdrrrhrlr{rgZ dense_solverrrrDrDrEtest_ridge_fit_intercept_sparses  r)r;r6r8cCsjtddd\}}||}t|d}d|}tjt|d|||Wdn1s\0YdS)Nrr)rNrOrrzsolver='{}' does not supportr)rr formatrrrr~)rprrhrlX_csrrrrDrDrE%test_ridge_fit_intercept_sparse_errors   rc Cs:tdd|dd\}}|rr\r]r_rr|r r~rr simplefilterryr)rrrr) rrdrrhrlrgr{rrJrrrDrDrE#test_ridge_fit_intercept_sparse_sags*     .rreturn_interceptr{rJ container)r2r7r8r9r:r;rvc Cstd}|dd}gd}t||}d}|r4d}||7}||} d\} } trTdnd } |d k} |d vr|rtjtd d (t| || |||| | dWdn1s0YdSt| || ||| || d}|r|\}}t ||d| dt ||d| dnt ||d| ddS)z=check if all combinations of arguments give valid estimationsrrJr)rPrGrrg@)rRư>rRrrv)r:r2zIn Ridge, only 'sag' solverr)rmrpr{rrrwN)rmrpr{rrrwrr r) r"rr>rr0rrrrr))rr{rrprgrhZ true_coefsrlZtrue_interceptZ X_testingrmrwrroutrrrDrDrE.test_ridge_regression_check_arguments_validitysP     $  r)r6r7r8r9r:r;rvcCs tjd}d}|dk}d\}}|||}||}|tj}|tj} dttjj} t||d| |d} | || | j } t||d| |d} | ||| j }| j |j ksJ|j |j ksJ| |j |j ksJ| |j |j ksJt | j | j dd d dS) NrrrrvrrG)rmrprrwrrgMb@?r)r>r\r]rr rZfinfo resolutionr r~rr rr))rprgrmrrMrNX_64y_64X_32y_32rwridge_32coef_32ridge_64coef_64rDrDrEtest_dtype_matchDs0         rc Cstjd}tddg}d\}}}|||}|||}|tj}|tj}t|dd} | ||| j } t|dd} | ||| j } | j |j ksJ| j |j ksJ| |j |j ksJ| |j |j ksJt | j | j dddS) Nrrrr)rCrrGr8rrrS) r>r\r]rrr rr r~rr rr*) rgrmrMrNZn_targetrrrrrrrrrDrDrEtest_dtype_match_choleskyis$          r)r6r8r9r7r:r;rvc Cstj|}d\}}|||}||}t||d||}d}|dk} t} |dkrbdnd} tjtjfD]2} t| | | | |||d| dd d d d | | <qr| tjj tjksJ| tjj tjksJt | tj| tj| d dS) Nrrrrrvr7rRr7rruF) rmrprOr{rrrwZ return_n_iterrr) r>r\r]rrr|rrrr r r)) rprrOrMrNrhrrlrmrresultsrZ current_dtyperDrDrE%test_ridge_regression_dtype_stabilitys4    rcCsRtdd\}}t|}|dddddf}|ddd}tdd||dS)NrrOrGr:rr)r r>Zasfortranarrayr r~)rhrlrDrDrEtest_ridge_sag_with_X_fortrans  rzClassifier, paramscCstddd\}}|dd}tj||gdd}|fi|||}||}|j|jks^Jt|dddf|dddftdd||dS) zRCheck that multilabel classification is supported and give meaningful results.rPr)rYrOrQrxNr:rr) r rr>rr~rrr,r ) ClassifierrJrhrlrr^r<rDrDrEtest_ridgeclassifier_multilabels   "rr2rv)rRrrrrcCstddgddgddgddgg}tdd g}|rHd }|||}n ||}t|d ||d }|||t|jd ksJdS)z:Test that positive Ridge finds true positive coefficients.rPrGrrLrrCrrrSrTrmrrprqrN)r>rrr r~r^r)rprqrmrhrrrlrrDrDrE#test_ridge_positive_regression_tests"  rc Cstjd}|dd}|jdd|jdd}|rDd}|||}n||}||j|jddd 7}g}d D](}t|||d d } || ||j qnt |d dddS)zTest that Ridge w/wo positive converges to the same solution. Ridge with positive=True and positive=False must give the same when the ground truth coefs are all positive. r,rrrrrPrYrr)TFru)rmrrqrwrr N) r>r\r]rr_rr`r r8r~rr)) rqrmrgrhrrrlrrrrDrDrE%test_ridge_ground_truth_positive_tests  r)r6r8r9r7r:r;c Csd}tddgddgg}tddg}||}t|d|dd }tjtd d |||Wd n1sp0Ytjtd d (t|||d|dd\}}Wd n1s0Yd S)z5Test input validation for positive argument in Ridge.rrPrGrrLrQTFrzdoes not support positiverNzonly 'lbfgs' solver can be used)rrpr)r>rr rrrr~r)rprmrhrrlrrkrDrDrEtest_ridge_positive_error_tests* rc stdddd\dd}dfdd }td }td d }||}||}||ksnJt|D]}|||d }||ksvJqvdS)z?Check ridge loss consistency when positive argument is enabled.rrrMrNrOrrNrcsp|j}|dur6tj|}|j|jd||jjd}n|j}dt||ddt|dS)NrrYrrG)rr>r\r]rr_rr})rrOZ noise_scalerrgrrhrmrlrDrE ridge_losss &z,test_positive_ridge_loss..ridge_lossrT)rmrr)Nr)r r r~r) rmZn_checksrrZmodel_positiveZlossZ loss_positiverOZloss_perturbedrDrrEtest_positive_ridge_losss    rcCsjtdddd\}}t|d}t|g}dddd}t|||fi|}t|||}t||d d d d S) zETest that LBGFS gets almost the same coef of svd when positive=False.rrrrPFgؗҜZ expand_dimsrrrr))rmrhrlconfigZ coef_lbfgsrrDrDrEtest_lbfgs_solver_consistency4s   rcCsvtddgddgg}tddg}tddddd dd }tjtd d |||Wd n1sh0Yd S)z1Test that LBFGS solver raises ConvergenceWarning.rPrQg _g _BrrvFrT)rmrprqrwrrzlbfgs solver did not convergerN)r>rr rrr r~)rhrlrrDrDrEtest_lbfgs_solver_errorEsrrtallcCs|dur&|dks|dvr&|r&tdtjd}d}|dkrH|d}n|d}|||}||} |durv||}t|d ||d k|d d } tfi| j|| dd } | j } |r| j } t | }| j|| |d t | j | dd|rt | j | |jdd|jdd}d|dd<| ddd9<| j|| |d | j } |rR| j } | j|ddddf| dd|ddd t | j | dd|rt | j | tfi| jtj| dd}|j|| tj|d |dvr|std|dt |j | dd|rt |j | |dur0|}tj||d|dgdd}t| | d|dg}| }|d|dd9<tj||d|dgdd}|dur||}||}tfi| j|| |d }tfi| j|||d }t |j |j |rt |j |j dS)zTest that the impact of sample_weight is consistent. Note that this test is stricter than the common test check_sample_weight_equivalence alone. Nr6)r8r;zunsupported configurationrrKrrGrrrvr)rqrmrprrOrwrzrrrrrUrJrmrrszSolver z- does fail test for scaling of sample_weight.rx)rrr>r\r]rr|r r~rrrZ ones_liker)r_rr?pirZtoarrayr)rqrrrprdrgrMrNrhrlrJrZrrr{rZX2y2Zsample_weight_1Zsample_weight_2rrDrDrE$test_ridge_sample_weight_consistencyVs            0     rcCstddd\}}tdd}d}tjt|d|||Wdn1sN0Ytddd }d }tjt|d|||Wdn1s0YdS) z-Check `store_cv_values` parameter deprecated.rCrrDT)store_cv_valuesz'store_cv_values' is deprecatedrN)r*rz2Both 'store_cv_values' and 'store_cv_results' were)r rrr FutureWarningr~rrrhrlrrOrDrDrE%test_ridge_store_cv_values_deprecateds * rcCsbtddd\}}tdd}d}tjt|d"||||jWdn1sT0YdS) zCheck `cv_values_` deprecated.rCrrDTrBz$Attribute `cv_values_` is deprecatedrN)r rrrrr~Z cv_values_rrDrDrEtest_ridge_cv_values_deprecateds   rrcCstdd|dd\}}tj|jdfd}|r:d|ddd<d }t|d |d d }|j|||d tjg|jt|Rd}t} t |D]d\} } t | ||D]J\} \} }t | |d}||| || || | |||| d| f<qqt |j|dS)arCheck that the predictions stored in `cv_results_` are on the original scale. The GCV approach works on scaled data: centered by an offset and scaled by the square root of the sample weights. Thus, prior to computing scores, the predictions need to be scaled back to the original scale. These predictions are the ones stored in `cv_results_` in `RidgeCV`. In this test, we check that the internal predictions stored in `cv_results_` are equivalent to a naive LOO-CV grid search with a `Ridge` estimator. Non-regression test for: https://github.com/scikit-learn/scikit-learn/issues/13998 rrTr)rMrNrrOrrNrGrrT)rrrqr*rzr.)r r>rrrr~emptyrr enumerater-r rr)r.)rrqrrhrlr{rr@r0rZ alpha_idxrmidxZ train_idxZtest_idxrrDrDrE!test_ridge_cv_results_predictionss,  rcstd|d\tjjdfdtddd}|jdt}t|jd t fd d | D}t |j t | d S) zCheck that `RidgeCV` works properly with multioutput and sample_weight when `scoring != None`. We check the error reported by the RidgeCV is close to a naive LOO-CV using a Ridge estimator. rGrrOrrrT)rr*rzrcs6g|].\}}j|||d|qS)rzr~rrtraintestrhrr{rlrDrEr sz;test_ridge_cv_multioutput_sample_weight..N)r r>rrrr~rr rsqueezer-r)rHr)rdr@r y_pred_loorDrrE'test_ridge_cv_multioutput_sample_weight s  rcstddd\ddfdd}t|d}|t}t|jd tfd d |D}t |j | d S) zECheck that `RidgeCV` works properly with a custom multioutput scorer.rGrrcSs>||d}tj|dd}|jdkr8tj|ddgd S| S)NrGrrxrP)r)r>r?ndimr)Zy_truerBerrorsZ mean_errorsrDrDrE custom_error s   z=test_ridge_cv_custom_multioutput_scorer..custom_errorcs||| S)zGMultioutput score that give twice more importance to the second target.)r)rkrhrl)rrDrEcustom_multioutput_scorer) szJtest_ridge_cv_custom_multioutput_scorer..custom_multioutput_scorer)rrcs.g|]&\}}|||qSrDrr)rhrrlrDrEr3 rLz;test_ridge_cv_custom_multioutput_scorer..N) r rr~rr rr>rr-r)rH)rr@rrrD)rhrrrlrE'test_ridge_cv_custom_multioutput_scorer s    r metaestimator)Zenable_metadata_routingcCs|dS)zTest that `RidgeCV` or `RidgeClassifierCV` with default `scoring` argument (`None`), don't enter into `RecursionError` when metadata is routed. N)Zget_metadata_routing)rrDrDrE*test_metadata_routing_with_default_scoring= srzmetaestimator, make_datasetcCs2|dddd\}}|j||t|jdddS)zTest that `set_score_request` is set within `RidgeCV.fit()` and `RidgeClassifierCV.fit()` when using the default scoring and no UnsetMetadataPassedError is raised. Regression test for the fix in PR #29634.rrrrrrzN)r~r>rr)rrFrhrlrDrDrE+test_set_score_request_with_default_scoringF s r) rrrrTrPrrrTFFN)r itertoolsrnumpyr>rZscipyrZsklearnrrZ sklearn.baserZsklearn.datasetsrrr r Zsklearn.exceptionsr Zsklearn.linear_modelr r rrrrZsklearn.linear_model._ridgerrrrrrrZsklearn.metricsrrrZsklearn.model_selectionrrrrr Zsklearn.preprocessingr!Z sklearn.utilsr"Zsklearn.utils._array_apir#r$r%r&r'Z-sklearn.utils._test_common.instance_generatorr(Zsklearn.utils._testingr)r*r+r,r-Zsklearn.utils.estimator_checksr.r/Zsklearn.utils.fixesr0r1r2r3r4r5ZSOLVERSrrZ load_diabetesZdiabetesrrrrrrindr\r]rgrZ load_irisrrXrWrFrHZfixtureromarkZ parametrizerrrrrrrrrrrrrrrrrr rr#r$r1r4r=rArGrIrPrRrUr[r_rdrerprur{sortedrrrrrrrrrrrrrr TypeErrorrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrDrDrDrEsf     $         F  ,  &  (  5  5  5 2        - !'  > 9    G  #   *   #      '        &    7 "        % `  +  "