t:iSSrSSKrSSKrSSKrSSKJr SSKJrJrJ r J r J r J r J r JrJrJrJr SSKrSSKJr SSKJrJr SSKJrJrJrJrJr /S Qr\\\ \\44r \r!"S S \5r"S \ \#S \ \ \#\\44Sjr$\"S5r%S\%S \%4Sjr&"SS5r'"SS\"5r("SS\"5r)"SS\"5r*"SS\"5r+g)z}Callback library containing training routines. See :doc:`Callback Functions ` for a quick introduction. N)ABC) AnyCallableDictListOptionalSequenceTuple TypeAliasTypeVarUnioncast) collective)EvalsLog _ScoreList)BoosterDMatrix XGBoostError_deprecate_positional_args_parse_eval_str)TrainingCallbackLearningRateScheduler EarlyStoppingEvaluationMonitorTrainingCheckPointCallbackContainerc\rSrSr%Sr\r\\S'SSjrS\ S\ 4Sjr S\ S\ 4S jr S\ S \ S \S\ 4S jrS\ S \ S \S\ 4S jrSrg)r3z;Interface for training callback. .. versionadded:: 1.3.0 rreturnNcgN)selfs S/srv/projetos/modelo_ml_acdoc/venv/lib/python3.13/site-packages/xgboost/callback.py__init__TrainingCallback.__init__=s modelcU$)zRun before training starts.r#r$r)s r%before_training TrainingCallback.before_training@ r(cU$)zRun after training is finished.r#r+s r%after_trainingTrainingCallback.after_trainingDr.r(epoch evals_logcg)zqRun before each iteration. Returns True when training should stop. See :py:meth:`after_iteration` for details. Fr#r$r)r2r3s r%before_iteration!TrainingCallback.before_iterationHs r(cg)aRun after each iteration. Returns `True` when training should stop. Parameters ---------- model : Eeither a :py:class:`~xgboost.Booster` object or a CVPack if the cv function in xgboost is being used. epoch : The current training iteration. evals_log : A dictionary containing the evaluation history: .. code-block:: python {"data_name": {"metric_name": [0.5, ...]}} Fr#r5s r%after_iteration TrainingCallback.after_iterationOs&r(r#)r N)__name__ __module__ __qualname____firstlineno____doc__rr __annotations__r&_Modelr,r0intboolr6r9__static_attributes__r#r(r%rr3s#Hi" VFvfSXRVVCHQUr(rrlistr c0nUSR5SnUHnUR5nX$S:Xde[USS5HhupV[U[5(dUR 5nURS5upxXW4U;a/XU4'XU4R [ U55 Mj M Un /n [UR5SS9Huupn [R"U 5n[U [5(dU R 5n [R"U5[R"U5nnU RXU4/5 M U $)z#Aggregate cross-validation results.rrN:cUSS$Nrr#)xs r%_aggcv..vs AaDGr()key)split enumerate isinstancestrdecodeappendfloatsorteditemsnumpyarraymeanstdextend)rEcvmapidxlinearr metric_idxitkvmsgresults_namesas_arrrYrZs r%_aggcvrjes413E (.. 1 C!f}}'AB0NJb#&&YY[88C=DAe+)+1o& q/ " ) )%( 3 1 CGu{{}2CD 1Q#s##**,CJJv& &(9cS)*+ E Nr(_ARTscorec([R"5nUS:wdeUS:XaU$[U[5(a [ S5e[ R "U/5n[R"U[RR5U- nUS$)zHelper function for computing customized metric in distributed environment. Not strictly correct as many functions don't use mean value as final result. rrzBxgboost.cv function should not be used in distributed environment.) rget_world_sizerPtuple ValueErrorrWrX allreduceOpSUM)rlworldr_s r%_allreduce_metricrus  % % 'E A:: z % P   ++ug C   sJMM$5$5 6 >C q6Mr(c \rSrSrSrSS\\S\\S\ S\ SS4 S jjr S \ S\ 4S jr S \ S\ 4S jr S \ S \S\S\\\\\4S\ 4 SjrS\\\\\4\\\\\44S \SS4SjrS \ S \S\S\\\\\4S\ 4 SjrSrg)rz^A special internal callback for invoking a list of other callbacks. .. versionadded:: 1.3.0 N callbacksmetric output_marginis_cvr cf[[RU55UlUH#n[ U[ 5(aM[ S5e SnUb[U5(d [ U5eX l[R"5Ul X0l X@l UR(aSUlgg)Nz3callback must be an instance of `TrainingCallback`.zmetric must be callable object for monitoring. For builtin metrics, passing them in training parameter invokes monitor automatically.)listdictfromkeysrxrPr TypeErrorcallablery collections OrderedDicthistory_output_marginr{ aggregated_cv)r$rxryrzr{cbrds r%r&CallbackContainer.__init__sdmmI67Bb"233 UVV  R   hv&6&6C.  !,!8!8!: + ::KOD  r(r)cURHinURUS9nSnUR(a([UR[ 5(dU5eMM[U[ 5(aMdU5e U$)z Function called before training.r)z'before_training should return the model)rxr,r{rPcvfoldsr}rr$r)crds r%r,!CallbackContainer.before_trainingsjA%%E%2E;Czz!%--66;;6!%116361   r(cURHinURUS9nSnUR(a([UR[ 5(dU5eMM[U[ 5(aMdU5e U$)zFunction called after training.rz&after_training should return the model)rxr0r{rPrr}rrs r%r0 CallbackContainer.after_trainingsjA$$5$1E:Czz!%--66;;6!%116361   r(r2dtrainevalscJ^^^[UUU4SjTR55$)z*Function called before training iteration.c3\># UH!oRTTTR5v M# g7fr")r6r.0rr2r)r$s r% 5CallbackContainer.before_iteration..s( DRq  ueT\\ : :N),)anyrx)r$r)r2rrs``` r%r6"CallbackContainer.before_iterations# DHNN   r(rlc UGHynUSnUSnUR(a2[[[[[[4U5S5nXV4nOUnUR S5nUSn SR USS5n [U5nXR;a"[R"5URU 'URU n X;a[[/5X'Xn UR(aQ[[[[[4U 5R[[[[4U55 GME[[[U 5R[[U55 GM| g)Nrr-)r{rTrr rQrNjoinrurrrrrrS) r$rlr2drgrhrZrJ splited_names data_name metric_name data_historymetric_historys r%_update_history!CallbackContainer._update_historys5 A!DtAzzDsE5'8!91=a@AH JJsOM%a(I((=#45K!!$A ,*5*A*A*C Y'<< 2L.,0R,@ ))6NzzT%u -.?FFue|,a0T%[.188eQH/r(c^^^TR(aLTRTTRTR5n[ U5nUTlTR UT5 OsUc/OUnUH!upgURS5S:XaMS5e TRUTTRTR5n[U5n TR U T5 [UUU4SjTR55n U $)z)Function called after training iteration.rz#Dataset name should not contain `-`c3\># UH!oRTTTR5v M# g7fr")r9rrs r%r4CallbackContainer.after_iteration.. s%XA##E5$,,??r) r{evalryrrjrrfindeval_setrrrx) r$r)r2rrscoresrfrgrl metric_scorerets ``` r%r9!CallbackContainer.after_iterations ::ZZt{{D4G4GHFF^F!'D    /-BUE yy~+R-RR+!udkk4CVCVWE*51L  u 5XXX r()rrrxrr{ry)NTF)r;r<r=r>r?r rrrrCr&rAr,r0rBrrr rQr6r rTrr9rDr#r(r%rrs\&*" P,-P"P P  P  P6 V   F v         U7C<012    IT%U +,d5eU9J3K.LLMII  I<  U7C<012   r(rcn^\rSrSrSrS\\\/\4\ \4SS4U4Sjjr S\ S\S \ S\ 4S jrS rU=r$) ria[Callback function for scheduling learning rate. .. versionadded:: 1.3.0 Parameters ---------- learning_rates : If it's a callable object, then it should accept an integer parameter `epoch` and returns the corresponding learning rate. Otherwise it should be a sequence like list or tuple with the same size of boosting rounds. learning_ratesr Nc >^[T5(d@[T[RR5(d[ S[ T535e[T5(aTUlO U4SjUl[TU]%5 g)Nz=Invalid learning rates, expecting callable or sequence, got: c*>[[T5U$r")rr )r2rs r%rK0LearningRateScheduler.__init__../sX~0Nu0Ur() rrPrabcr rtypersuperr&)r$r __class__s `r%r&LearningRateScheduler.__init__!su'' KOO441 1 O'(*  N # #"0D "UD  r(r)r2r3cFURSURU55 g)N learning_rateF) set_paramrr5s r%r9%LearningRateScheduler.after_iteration2s )<)r?r rrBrTr r&rArrCr9rD __classcell__rs@r%rrs] #HcUE\$:HUO$KL "VCHQUr(rc^\rSrSrSr\SSSSSS.S\S\\S \\S \\ S \\ S \ S S4U4Sjjj5r S\ S \ 4Sjr S\S\S\S\ S\S \ 4 SjrS\ S\S\S \ 4SjrS\ S \ 4SjrSrU=r$)ri8aCallback function for early stopping .. versionadded:: 1.3.0 Parameters ---------- rounds : Early stopping rounds. metric_name : Name of metric that is used for early stopping. data_name : Name of dataset that is used for early stopping. maximize : Whether to maximize evaluation metric. None means auto (discouraged). save_best : Whether training should return the best model or the last model. If set to `True`, it will only keep the boosting rounds up to the detected best iteration, discarding the ones that come after. This is only supported with tree methods (not `gblinear`). Also, the `cv` function doesn't return a model, the parameter is not applicable. min_delta : .. versionadded:: 1.5.0 Minimum absolute change in score to be qualified as an improvement. Examples -------- .. code-block:: python es = xgboost.callback.EarlyStopping( rounds=2, min_delta=1e-3, save_best=True, maximize=False, data_name="validation_0", metric_name="mlogloss", ) clf = xgboost.XGBClassifier(tree_method="hist", device="cuda", callbacks=[es]) X, y = load_digits(return_X_y=True) clf.fit(X, y, eval_set=[(X, y)]) NFg)rrmaximize save_best min_deltaroundsrrrrrr c>X0lX lXlXPlX@l0UlX`lUR S:a [S5eSUl0Ul SUl [TU]15 g)Nrz(min_delta must be greater or equal to 0.) datarrrrstopping_history _min_deltarpcurrent_rounds best_scoresstarting_roundrr&)r$rrrrrrrs r%r&EarlyStopping.__init__gsg & " *,# ??Q GH H#$!##$ r(r)cUR5Ul[U[5(dUR(a [ S5eU$)NzP`save_best` is not applicable to the `cv` function as it doesn't return a model.)num_boosted_roundsrrPrrrpr+s r%r,EarlyStopping.before_trainings?#668%))dnn#  r(rlrgryr2cX^^^ S[S[4Sjm S[S[S[4U U4SjjnS[S[S[4U U4SjjnTRc1SnTS :wa"[ U4S jU55(aS TlOS TlTR(aUn OUn TR (dS Tl0TR U'[[U/5TR UT'0TRU'U/TRUT'UR[T "U55[U5S9 OU "UTRUTS5(d7TR UTRU5 T=R S- slOTR UTRU5 TRUTRU5 TR UTSn UR[T "U 55[U5S9 S TlTR TR:ag g )Nvaluer c:[U[5(aUS$U$)z+get score if it's cross validation history.r)rPro)rs r%get_s+EarlyStopping._update_rounds..get_ss)%7758 BU Br(newbestcd>[R"T"U5TR- T"U55$)z-New score should be greater than the old one.rWgreaterrrrrr$s r%r.EarlyStopping._update_rounds..maximizes%==sdoo!=uT{K Kr(cd>[R"T"U5TR- T"U55$)z,New score should be lesser than the old one.rrs r%minimize.EarlyStopping._update_rounds..minimizes%==tt!>c K Kr() aucaucprprezpre@mapndcgzauc@zaucpr@zmap@zndcg@mapec3F># UHnTRU5v M g7fr") startswith)rrJrys r%r/EarlyStopping._update_rounds..s!'WFV(9(9!(<( >)S))CDOCTU '   **Kx}}/3K  &0 >? ?%b)"" PU#  r(cUR(dU$URnURnUbUceUSUS-nX!lX1lU$![an[S5UeSnAff=f)Nrz4`save_best` is not applicable to the current booster)rrrr)r$r)rres r%r0EarlyStopping.after_trainings~~L "11N))J!-*2H HH.NQ./E#1 )    F  s4A A& A!!A&) rrrrrrrrrr)r;r<r=r>r?rrBrrQrCrTr&rAr,rrrr9r0rDrrs@r%rr8s+\ &*#'#'$)c]  C=  4. D>  2V==&)=36=?E=NQ= =~" V" C" H" QU" HFvr(rc ^\rSrSrSrSSS\R 4S\S\S\S \ \ /S 44U4S jjjr S \ S \ S\ S\ \ S\ 4 SjrS\S\S\S\4SjrS\S\4SjrSrU=r$)riabPrint the evaluation result at each iteration. .. versionadded:: 1.3.0 Parameters ---------- rank : Which worker should be used for printing the result. period : How many epoches between printing. show_stdv : Used in cv to show standard deviation. Users should not specify it. logger : A callable used for logging evaluation result. rrFrankperiod show_stdvloggerNcp>XlX0lX lX@lUS:deSUl[ TU]5 grI) printer_rankrr_logger_latestrr&)r$rrrrrs r%r&EvaluationMonitor.__init__s8!"  zz&*  r(rryrlrZr cvUb&UR(aSUS-U-SUSSUS3nU$SUS-U-SUS3nU$)N rrGz.5f+)r)r$rryrlrZrds r% _fmt_metricEvaluationMonitor._fmt_metric#sa ?t~~tczF*+1U3KqS BC tczF*+1U3K8C r(r)r2r3c U(dgSUS3n[R"5UR:XaUR5HbupVUR5HIupxSn [ US[ 5(aUSSn USSn OUSn X@R XWX5- nMK Md US- nX R-S:XdURS:XaURU5 SUl gX@l g)NF[]rrr ) rget_rankrrVrPror rrr) r$r)r2r3rdrryrlogstdvrls r%r9!EvaluationMonitor.after_iteration,sugQ<    D$5$5 5 ) 1 (. $K,0D!#b'511 #B "2wqz #B++DuKKC)7!2 4KC #)T[[A-= S!#  # r(c[R"5UR:Xa(URbUR UR5 U$r")rrrrrr+s r%r0 EvaluationMonitor.after_trainingEs7    D$5$5 5$,,:R LL & r()rrrrr)r;r<r=r>r?rcommunicator_printrBrCrrQr&rTrr rArr9r0rDrrs@r%rrs((2(E(E   #%  !$-29A% VCHQU2Fvr(rc ^\rSrSrSrSrSS\\\R4S\S\ S\ S S 4 U4S jjjr S\ S \ 4S jrS\ S \ S\S \ 4SjrSrU=r$)riKaCheckpointing operation. Users are encouraged to create their own callbacks for checkpoint as XGBoost doesn't handle distributed file systems. When checkpointing on distributed systems, be sure to know the rank of the worker to avoid multiple workers checkpointing to the same place. .. versionadded:: 1.3.0 Since XGBoost 2.1.0, the default format is changed to UBJSON. Parameters ---------- directory : Output model directory. name : pattern of output model file. Models will be saved as name_0.ubj, name_1.ubj, name_2.ubj .... as_pickle : When set to True, all training parameters will be saved in pickle format, instead of saving only the model. interval : Interval of checkpointing. Checkpointing is slow so setting a larger number can reduce performance hit. ubjr) directoryrg as_pickleintervalr Nc>[R"U5UlX lX0lX@lSUlSUl[TU]%5 grI) osfspath_path_name _as_pickle _iterations_epoch_startrr&)r$rrgrrrs r%r&TrainingCheckPoint.__init__hs>YYy)  ##   r(c0UR5UlU$r")rr#r+s r%r,"TrainingCheckPoint.before_trainingws..0  r(r2r3cDURUR:Xa[RR UR UR S-[X R-5-UR(aSOSUR3-5nSUl[R"5S:XaNUR(a,[US5n[R"X5 SSS5 OUR!U5 U=RS- slg!,(df  N$=f)Nrfz.pkl.rwbrF)r"r!rpathrrrrQr#r default_formatrropenpickledump save_model)r$r)r2r3r*fds r%r9"TrainingCheckPoint.after_iteration{s ;;$** *77<<  u{{*+-"__6Ad6I6I5J2KMDDK""$)??dD)R E.*)$$T* q  *)s D D)r r"r!rrr#)r)Fd)r;r<r=r>r?r+r rQrPathLikerCrBr&rAr,rr9rDrrs@r%rrKs4N  bkk)*        VVCHQUr(r),r?rrr-rrtypingrrrrrr r r r r rrWr_typingrrcorerrrrr__all__rTrrArrQrjrkrurrrrrr#r(r%r9s       )  ueE5L)) * /s/d$s)U3u+<%= >6vTd&xxv#,#NE$EPH(HVB)Br(