o ?Hh҂ @s dZddlZddlmZddlZddlZddlmZddl m Z ddl m Z ddl mZmZmZmZddlmZmZmZmZmZmZdd lmZmZdd lmZmZmZm Z dd l!m"Z"m#Z#dd l$m%Z%m&Z&m'Z'm(Z(dd l)m*Z*m+Z+m,Z,ddl-m.Z.ddl/m0Z0ddl1m2Z2ddl3m4Z4m5Z5m6Z6ddl7m8Z8m9Z9m:Z:m;Z;ddlm?Z?m@Z@mAZAddlBmCZCmDZDmEZEeZFeFjGeFjHZIZJeZKeKjGeKjHZLZMeddd\ZNZOeddd\ZPZQejRSdde+ddddgejRSddeddgejRSdd dgd!d"ZTd#d$ZUd%d&ZVd'd(ZWejRSdde*ddddgejRSd)difeddifed*difgejRSdd dgd+d,ZXejRSd-eCeDeEd.d/ZYejRSd-eCeDeEd0d1ZZd2d3Z[Gd4d5d5eeZ\Gd6d7d7eeZ]ejRSd8eMd9gie^d:feMd;e&fdfgd?d@e^dAfeMd9d;e&fdBe]fgie_dCfeMd;e&fdBe5d=d>fge]dDe_dCfgdEdFZ`ejRSd8eJd9gie^d:feJd9d;e%fdBe\fgie_dCfeJd;e%fdBe6fge\dDe_dCfgdGdHZaejRjSdIed;e&ddfd             r]c Cstdd\}}tt|||dd\}}}}dtfdtddfg}t|dd }|||||}|jd d ks;Jdtfd t fg}|j |d |||||}|jd d ks_JdS)NT return_X_yr1r;r=rfr9r0r@r5rDr4r>r@) r rr!rrrrKrOrPr#rQ) XyrSrTrU_r@rWrXr[r[r\:test_stacking_classifier_drop_column_binary_classificationqs         rfcCstttttdd\}}}}ddtddfg}tddd}tdtddfg|d d }t||d d }||||||t| || |t| || |t| || |dS) Nr1r;r=rFr>rr9rB n_estimatorsr3r@r8r5) rr!rIrJr#rrrKr*rLrMrO)rSrTrUrer@r`rWclf_dropr[r[r\'test_stacking_classifier_drop_estimators     rmcCsttttdd\}}}}ddtddfg}tddd}tdtddfg|dd }t||dd }||||||t| || |t| || |dS) Nr1r9rgsvrrrBrhrjrk) rr! X_diabetes y_diabetesr$rrrKr*rLrO)rSrTrUrer@r`regreg_dropr[r[r\&test_stacking_regressor_drop_estimators    rszfinal_estimator, predict_params return_stdcCs2ttttdd\}}}}dtfdtfg}t||||d} | ||| j|fi|} |r2dnd} |r>t | | ks>J| |} |rGdnd} | j d| ksRJ|rat || ddd df| j d d | ||| || |} |r{d nd}| j d|ksJ|rt || ddd dfdSdS) Nr1r9r=rnr?r4rD rFrG )rr!rorprr$rrKrLlenrOrPr*rQ)r5r8predict_paramsr:rSrTrUrer@rqresultexpected_result_lengthrXrYrZr[r[r\ test_stacking_regressor_diabetess:          r|sparse_containerc Ct|tttdd\}}}}dtfdtfg}tddd}t||ddd }|||| |}t ||ddd dft |sFJ|j |j ksNJdS) Nr1r9r=rnrBrhrjTr?rv)rr!rorprr$rrrKrOr+rissparseformat r}rSrTrUrer@r`rWrXr[r[r\*test_stacking_regressor_sparse_passthrough   rc Cr~) Nr1r9r=r>rBrhrjTr?rE)rr!rIrJrr#rrrKrOr+rrrrr[r[r\+test_stacking_classifier_sparse_passthroughrrcCshttddtdd}}dtfdtfg}t|d}|||||}|jddks2JdS)Ndr=r`rbrDr4) r!rIrJrrrrKrOrP)X_y_r@rWX_metar[r[r\)test_stacking_classifier_drop_binary_prob s    rc@seZdZddZddZdS)NoWeightRegressorcCst|_|j||SN)rrqrKselfrcrdr[r[r\rKszNoWeightRegressor.fitcCst|jdS)Nr)nponesrP)rrcr[r[r\rL szNoWeightRegressor.predictN)__name__ __module__ __qualname__rKrLr[r[r[r\rs rc@seZdZddZdS)NoWeightClassifiercCstdd|_|j||S)N stratified)strategy)rrWrKrr[r[r\rK%s zNoWeightClassifier.fitN)rrrrKr[r[r[r\r$s rzy, params, type_err, msg_errr@zInvalid 'estimators' attribute,r=svmiPmax_iterrM)r@ stack_methodz+does not implement the method predict_probacorzdoes not support sample weightr@r8cCjtj||d$tdi|ddi}|jtt|ttjddWddS1s.wYdSNmatchr5r0r sample_weightr[) pytestraisesrrKr!rIrrrP)rdparamstype_errmsg_errrWr[r[r\test_stacking_classifier_error*s*""rcCrr) rrrrKr!rorrrP)rdrrrrqr[r[r\test_stacking_regressor_errorYs""rzestimator, X, yrbrrr)idscCst|}|jtdtjdddt|}|jdd|jtdtjdddt||||ddddf||||dS)NTrr7r3r5rFrGrD) r rQrrrandom RandomStater*rKrO) estimatorrcrdestimator_fullestimator_dropr[r[r\test_stacking_randomnessws  rcCs2tdtddfdtddfgd}|ttdS)Nr=i'rrrb)rrr#rKrIrJ)rWr[r[r\)test_stacking_classifier_stratify_defaults   rz stacker, X, yrrkr^c CsBt|d}tdg|dgt||}t|||dd\}}}}} }ttd|||Wdn1s;wY||} ttd|j||t|j dWdn1sawY||} t | | ttd|j||| dWdn1swY||} t | |  dksJdS) Nr4g?g?r1r9)categoryrr) rxrarrayrr,rrKrLrrPr*abssum) stackerrcrdn_half_samplestotal_sample_weightrSrTrUresample_weight_trainy_pred_no_weighty_pred_unit_weight y_pred_biasedr[r[r\ test_stacking_with_sample_weights* !       rcCs>tdtddfgtddd}|jttttjdddS)Nr=T)expected_sample_weightrrr)rr)rKrIrJrrrP)rr[r[r\0test_stacking_classifier_sample_weight_fit_params  rz-ignore::sklearn.exceptions.ConvergenceWarningcCst|}t|}|jdd|jdd||||||t|j|jD] \}}t|j|jq'tjt ddt|j j|j jWddS1sOwYdS)Nr0rrjz Not equalr) r rQrKzip estimators_r*coef_rrAssertionErrorfinal_estimator_)rrcrd stacker_cv_3 stacker_cv_5est_cv_3est_cv_5r[r[r\test_stacking_cv_influences!     "rz7Stacker, Estimator, stack_method, final_estimator, X, yrLcCst||ddd\}}}} d|||fd|||fg} | D]\} } tdd| _t| |} t| d}||_t| ||q || d |d }||| |jd d | DksWJtd d|jDscJ|jD] } t| |}||qfdS)z2Check the behaviour of stacking when `cv='prefit'`r1g?)r3 test_sized0d1rK)name) side_effectprefit)r@r5r8cSsg|]\}}|qSr[r[).0rerr[r[r\ Xsz(test_stacking_prefit..css|] }|jjdkVqdS)rN)rK call_count)rrr[r[r\ Zsz'test_stacking_prefit..N) rrKrgetattrrsetattrrallassert_called_with)Stacker Estimatorrr8rcrdX_train1X_train2y_train1y_train2r@rer stack_funcpredict_method_mockedrstack_func_mockr[r[r\test_stacking_prefit*s,        rrracCs<tt|||WddS1swYdSr)rrrrK)rrcrdr[r[r\test_stacking_prefit_errorbs "rz!make_dataset, Stacking, EstimatorcCsGddd|}|ddd\}}|d|fgd}|jd}tjt|d  |jWdn1s4wY|||d }tjt|d  |jWddS1sWwYdS) Ncs eZdZdZfddZZS)z8test_stacking_without_n_features_in..MyEstimatorz Estimator without n_features_in_cst|||`dSr)superrKn_features_in_r __class__r[r\rKsz.MyEstimator.fit)rrr__doc__rK __classcell__r[r[rr\ MyEstimatorsrrr)r3 n_samplesr=rbz' object has no attribute n_features_in_rz6'MyEstimator' object has no attribute 'n_features_in_')rrrAttributeErrorrrK) make_datasetStackingrrrcrdrmsgr[r[r\#test_stacking_without_n_features_ins   "rrr rc Csttttdd\}}}}d}d|fg}t|tdd||}||}|j|jd|fks0Jtt |j dd d r>J| |} | j|jksKJd S) zCheck the behaviour for the multilabel classification case and the `predict_proba` stacking method. Estimators are not consistent with the output arrays and we need to ensure that we handle all cases. r1r;r0estrMr@r8rrrD)axisg?N) r X_multilabel y_multilabelrrrKrOrPanyriscloserrL) rrSrTrUrV n_outputsr@rrXy_predr[r[r\1test_stacking_classifier_multilabel_predict_probas"   rc Csttttdd\}}}}d}dtfg}t|tdd||}||}|j|jd|fks1J| |}|j|jks>JdS) zCheck the behaviour for the multilabel classification case and the `decision_function` stacking method. Only `RidgeClassifier` supports this case. r1r;r0rrRrrN) rrrrrrrKrOrPrL) rSrTrUrVrr@rrXrr[r[r\5test_stacking_classifier_multilabel_decision_functions    rrautocCsttttdd\}}}}|}d}dtddfdtddfdtfg}t} t|| ||d ||} t ||| |} | j |j ksEJ|d krNgd } nd gt |} | j| ks\J|t |} |rk| |j d 7} | |}|j |j d | fks|Jt | jtd d gg|dS)zCheck the behaviour for the multilabel classification case for stack methods supported for all estimators or automatically picked up. r1r;r0mlpr9r`ridge)r@r8r:rr)rMrMrRrLrDrN)rrrcopyr rrrrrKrrLrPrx stack_method_rOclasses_rr)rr:rSrTrUrVy_train_before_fitrr@r8rWrexpected_stack_methodsn_features_X_transrXr[r[r\0test_stacking_classifier_multilabel_auto_predicts>        rz,stacker, feature_names, X, y, expected_names)stackingclassifier_lr0stackingclassifier_lr1stackingclassifier_lr2stackingclassifier_svm0stackingclassifier_svm1stackingclassifier_svm2)otherrFstackingclassifier_lrstackingclassifier_svmstackingregressor_lrstackingregressor_svm)StackingClassifier_multiclassStackingClassifier_binaryrcCsF|j|d|t|||rt||f}||}t||dS)z/Check get_feature_names_out works for stacking.)r:N)rQrKr!r concatenateget_feature_names_outr)r feature_namesrcrdexpected_namesr: names_outr[r[r\test_get_feature_names_outs B rcCsftttttdd\}}}}tdtfgd}|||||||| ||dks1JdS)zNCheck that a regressor can be used as the first layer in `StackingClassifier`.r1r;rrbrAN) rr!rIrJrrrKrLrMrN)rSrTrUrVrWr[r[r\'test_stacking_classifier_base_regressor\s    rcCstdd\}}dtfdtdddfg}tddd}t||dd}d }d }tjt|d }||||Wd n1s@wYt |j j tsNJ|t |j j vsXJd S) a Check that we raise the proper AttributeError when the final estimator does not implement the `decision_function` method, which is decorated with `available_if`. Non-regression test for: https://github.com/scikit-learn/scikit-learn/issues/28108 r1r9r=r`r4rhr0rkz>This 'StackingClassifier' has no attribute 'decision_function'zD'RandomForestClassifier' object has no attribute 'decision_function'rN) r rrrrrrrKrR isinstancevalue __cause__str)rcrdr@r8rW outer_msg inner_msg exec_infor[r[r\-test_stacking_final_estimator_attribute_errorhs r!zEstimator, ChildcCsTtjtdd|d|gjttgdddWddS1s#wYdS)zTest that the right error message is raised when metadata is passed while not supported when `enable_metadata_routing=False`.z1is only supported if enable_metadata_routing=TruerrW)rDrDrDrDrDarmetadataN)rr ValueErrorrKrIrJ)rChildr[r[r\*test_routing_passed_metadata_not_supporteds  "r')enable_metadata_routingcCs|d|fg}|dS)Nsub_est)get_metadata_routing)rr&rr[r[r\%test_get_metadata_routing_without_fits r+zprop, prop_valuer)r$r"c Cs&|d|tdjd i|difd|tdjd i|difg|tdjd i|did}|jttfi||i|jttfi||i|jtfi||i|jD] }|dj }t |scJ|D]}t d |dd|d||iqeqV|j j }t |sJt d |d d d |d||id S) z5Test that metadata is routed correctly for Stacking*.sub_est1)registryTsub_est2)r8rDrK)objmethodparent split_paramsrLNr[) r'set_fit_requestset_predict_requestrKrIrJ fit_transformrLr@r-rxr(r)rr&prop prop_valuerrr-r)r[r[r\-test_metadata_routing_for_stacking_estimatorssH      r9cCsttjdd}}|d|fg}d|jd}tjtt |d|j tt ||dWddS1s9wYdS) zCTest that the right error is raised when metadata is not requested.rr"r)zb[sample_weight, metadata] are passed but are not explicitly set as requested or not requested for z.fitrr#N) rrrIrPrrrr%reescaperKrJ)rr&rr$r error_messager[r[r\3test_metadata_routing_error_for_stacking_estimatorss "r=)xrr: unittest.mockrnumpyrr numpy.testingrscipyrsklearnr sklearn.baserrrr sklearn.datasetsr r r r rr sklearn.dummyrrsklearn.ensemblerrrrsklearn.exceptionsrrsklearn.linear_modelrrrrsklearn.model_selectionrrrsklearn.neighborsrsklearn.neural_networkr sklearn.preprocessingr! sklearn.svmr"r#r$%sklearn.tests.metadata_routing_commonr%r&r'r(sklearn.utils._mockingr)sklearn.utils._testingr*r+r,sklearn.utils.fixesr-r.r/diabetesdatatargetrorpirisrIrJrrX_binaryy_binarymark parametrizer]rfrmrsr|rrrrrr% TypeErrorrrrrrrfilterwarningsrrrrrrrrrrr!r'r+rrPr9r=r[r[r[r\s          '  $          )                       #     -       4<  !   .