bi7 UdZddlZddlmZddlmZddlmZmZm Z ddl m Z ddl m Z mZmZmZmZmZmZmZmZmZmZmZmZmZmZmZmZddlZddlZddl Z ddlm!Z"dd lm#Z$dd lm%Z&dd l'm(Z(dd lm)Z)dd l*m+Z+ddl*m,Z-ddl.m/Z/m0Z0ddl1m2Z2m3Z3m4Z4ddl5m6Z6ddl/m7Z8ddl/m9Z:ddl/m;Z<ddl=m>Z>m?Z?ddl@mAZAmBZBmCZCmDZDmEZEmFZFmGZGmHZHddlImJZJmKZKddlLmMZMmNZNmOZOmPZPmQZQmRZRmSZSmTZTmUZUmVZVmWZWmXZXddlYmZZZddl[m\Z]ddlIm^Z^m_Z_m`Z`ddlambZbmcZcee"je&j|e&jfZfeegd<ee"je&j|fZheegd <ed!eAed"Zigd#Zjejd$Zled%e+fd&Zmed%enfd'Zoed%enfd(Zpd)eqd*eeeereeereqffd+eeqd%e:fd,Zsd)eqd-eerd.eeereqfd+eeqd%e:f d/ZtGd0d1e/jZud2ed3d%d3fd4ZvGd5d6Zwed7Zxed8Zyd2ed3d9eeyexfd:e d;eerd%exf d<ZzGd=d>ewZ{d?end@e d%eBfdAZ| dd2d3d)eqdBeeere fdCee8d%eereereqfff dDZ}d%eeere ffdEZ~dFewdGeeeewerfd%eerfdHZd;eerd2d3d%dfdIZdJedKeqd:edLeeqdMeerdNeqd%eeBeeeBerfffdOZd2d3dPeere fdBeeere fdQeere fdFewdReqdGeeeewerfdSeeDdTeeqdUeeqenfdVeeAdWeee6dXeeCdCee8d%eeifdYZeG ddddddZdddd[d2d3dQeere fdFewdReqdGeeeewerfdSeeDdTeeqdVeeAdUeeqenfdWeee6dXeeCdCee8d%e fd\Z\d]end^ed%enfd_Zd`e dae dbeeqd]end%e f dcZddededfd`ehdgeefd^eeqdhfdieeqerfd%effdjZdeeAdkeqd]endlend@e d%eeeqdhfeeqerfff dmZd2d3dneeAedffd%dffdoZd2d3dPeere fdneeAedffd`ehdpendqedrendsendtenduendvendwe4dxend%effdyZeGdze jdzdzdzdzdZd{dzd| d2ed3dneeieAdffd`eewehfdpendqedrendsendtenduendvendwe4dxend%e fd}Zd2d3dPeere fdneeAedffd`ehdwe4d~erdqedvendgeefdxend%effdZeGd{de jdZddzdd2ed3dneeieAdffd`ehdwe4d~erdqedvendgeefdxend%e fdZd2ed3deerdeerdeeqd@e d%eeweeeewerfff dZedndd2d3d%efdZGddeOZeXdddngGddeReZeXdddngGddeNeZeXdddngddGddeQeZeXddndgdGddeZeXddndgdGddeZy)aS Dask extensions for distributed training ---------------------------------------- See :doc:`Distributed XGBoost with Dask ` for simple tutorial. Also :doc:`/python/dask-examples/index` for some examples. There are two sets of APIs in this module, one is the functional API including ``train`` and ``predict`` methods. Another is stateful Scikit-Learner wrapper inherited from single-node Scikit-Learn interface. The implementation is heavily influenced by dask_xgboost: https://github.com/dask/dask-xgboost Optional dask configuration =========================== - **coll_cfg**: Specify the scheduler address along with communicator configurations. This can be used as a replacement of the existing global Dask configuration `xgboost.scheduler_address` (see below). See :ref:`tracker-ip` for more info. The `tracker_host_ip` should specify the IP address of the Dask scheduler node. .. versionadded:: 3.0.0 .. code-block:: python from xgboost import dask as dxgb from xgboost.collective import Config coll_cfg = Config( retry=1, timeout=20, tracker_host_ip="10.23.170.98", tracker_port=0 ) clf = dxgb.DaskXGBClassifier(coll_cfg=coll_cfg) # or dxgb.train(client, {}, Xy, num_boost_round=10, coll_cfg=coll_cfg) - **xgboost.scheduler_address**: Specify the scheduler address .. versionadded:: 1.6.0 .. deprecated:: 3.0.0 .. code-block:: python dask.config.set({"xgboost.scheduler_address": "192.0.0.100"}) # We can also specify the port. dask.config.set({"xgboost.scheduler_address": "192.0.0.100:12345"}) N) defaultdict)contextmanager)cachepartialupdate_wrapper)Thread)Any AwaitableCallableDict GeneratorIterableListOptional ParamSpecSequenceSetTuple TypeAlias TypedDict TypeGuardTypeVarUnion)array)bag) dataframe)Delayed)Future)Version)parse) collectiveconfig) FeatureNames FeatureTypesIterationRange)TrainingCallback)Config)_Args)_ArgVals) DataFramelazy_isinstance)BoosterDMatrixMetric Objective XGBoostError_check_distributed_params_deprecate_positional_args_expect) _is_cudf_ser_is_cupy_alike) XGBClassifierXGBClassifierBaseXGBModel XGBRankerXGBRankerMixInXGBRegressorBase _can_use_qdm_check_rf_callback_cls_predict_proba_objective_decorator_wrap_evaluation_matricesxgboost_model_doc) RabitTracker)train)_create_dmatrix_create_quantile_dmatrixno_group_split)get_address_from_user get_n_threads_DaskCollection_DataT TrainReturnTboosterhistory) CommunicatorContext DaskDMatrixDaskQuantileDMatrixDaskXGBRegressorDaskXGBClassifier DaskXGBRankerDaskXGBRFRegressorDaskXGBRFClassifierrDpredictinplace_predictz[xgboost.dask]returnc4ttjSN) parse_versiondask __version__P/home/cdr/jupyterlab/.venv/lib/python3.12/site-packages/xgboost/dask/__init__.py _DASK_VERSIONrds )) **rbc.ttdk\S)Nz 2024.12.1rdr^rarbrc_DASK_2024_12_1rgs ?mK8 88rbc.ttdk\S)Nz2025.3.0rfrarbrc_DASK_2025_3_0ris ?mJ7 77rb n_workersaddrstimeoutc ji} t|dtr%|dd}|dd}t|||d|dn|}n,|d}t|ts|Jt||d|dn|}|j t |j }d|_|j |j|j|S#t$rc} t|dkrtjd t|dt|dt| t||dd|}Yd} ~ |Sd} ~ wwxYw) NrrEtask)rjhost_ipportsortbyrl)rjrorqrl)targetTr!zCFailed to bind address '%s', trying to use '%s' instead. Error: %s) isinstancetuplerCstrstartrwait_fordaemonupdate worker_argsr1lenLOGGERwarning_try_start_tracker) rjrkrlenvrorp rabit_trackeraddrthreades rcr~r~sD C%@ eAh &AhqkG8A;D(#$_' M8DdC(DL 88(#$_' M }556   =,,./ J  @ u:>  R aM aM F  !E!"Iw? J @sCC D2AD--D2addr_from_daskaddr_from_userc$t|||g|}|S)z8Start Rabit tracker, recurse to try different addresses.)r~)rjrrrlrs rc_start_trackerrs Y(H' RC Jrbc,eZdZdZdeddffd ZxZS)rQzNA context controlling collective communicator initialization and finalization.argsr[Nc t|di|tj}d|jd|j |j d<y)Nz[xgboost.dask-z]: DMLC_TASK_IDra)super__init__ distributed get_workernameaddressr)selfrworker __class__s rcrzCommunicatorContext.__init__sG  4 '') '5V[[MFNNCS$T .!rb)__name__ __module__ __qualname____doc__ CollArgsValsr __classcell__rs@rcrQrQs"XU|UUUrbrQclientdistributed.Clientc&t|ttjtdfsDt t ttjtdgt||tj}|S|}|S)z#Simple wrapper around testing None.N)rstyper get_client TypeErrorr4)rrets rc _get_clientrs{ ftK$:$:$<=tDzJ K T+0023T$Z@$v, O  '-n+ "C J;AC Jrbc#heZdZdZe dddddddddddddd deddedeed eed eed eed e d ee dee deedeedeedeedeede ddf dZ de fdZdddddddddddedeedeed eedeedeedeedeeddfdZdedeeeffdZdefdZy)rRaqDMatrix holding on references to Dask DataFrame or Dask Array. Constructing a `DaskDMatrix` forces all lazy computation to be carried out. Wait for the input data explicitly if you want to see actual computation of constructing `DaskDMatrix`. See doc for :py:obj:`xgboost.DMatrix` constructor for other parameters. DaskDMatrix accepts only dask collection. .. note:: `DaskDMatrix` does not repartition or move data between workers. It's the caller's responsibility to balance the data. .. note:: For aligning partitions with ranking query groups, use the :py:class:`DaskXGBRanker` and its ``allow_group_split`` option. .. versionadded:: 1.0.0 Parameters ---------- client : Specify the dask client used for training. Use default client returned from dask if it's set to None. NF) weight base_marginmissingsilent feature_names feature_typesgroupqidlabel_lower_boundlabel_upper_boundfeature_weightsenable_categoricalrrdatalabelrrrrrrrrrrrrr[c t|}||_| |_||ntj|_||_| | td| tdt|jdk7rtd|jt|tjtjfst3D!&[[  #+//!   rbc6|jjSr])r __await__rs rcrzDaskDMatrix.__await__rszz##%%rb)rrrrrrrrc  !Kdttdtdttdtdtf ddtdttffd !d t tdt ttf!fd } !|| |} | |} | |} | |}| |}| | }d i d t ttd tddf fd }|| d|| d|| d||d||d||dg}t tD]5}i} jD] \}}||||<|j|7tttj|}j|}tj |d{|D]"}|j"dk(rJ|j"i|_t'|D]\}}||j$|j(< |Dcic]}|j(|}}j*j-|Dcgc]}|j(c}d{}t/t}|jD].\}}|t1t3|j||0||_| d|_|Sj|j9d{|_|S7Ecc}wcc}w77w)z Obtain references to local data.left left_nameright right_namer[c Hd|d|dt|dt|d }|S)NzPartitions between z and z are not consistent: z != z/. Please try to repartition/rechunk your data.r{)rrrrmsgs rc inconsistentz1DaskDMatrix._map_local_data..inconsistentsA&i[j\B"4ykc%j\:?@  Jrbdcj|}t|jdrIt|jjdkDr'|jjddkDr t dj |S)zBreaking data into partitions.rrEzData should be partitioned by row. To avoid this specify the number of columns for your dask Array explicitly. e.g. chunks=(partition_size, -1]))persisthasattr partitionsr{rr futures_of)rrs rc to_futuresz/DaskDMatrix._map_local_data..to_futuressrq!A g. **+a/LL&&q)A- 4 $$Q' 'rbmetac| |}|Syr]ra)r meta_partsrs rc flatten_metaz1DaskDMatrix._map_local_data..flatten_metas+5d+; !!rbrm_partsrNcd|-tt|k(sJd||||<yy)NXr)rrX_partsrpartss rc append_metaz0DaskDMatrix._map_local_data..append_metasF"7|s7|3\S'463&d #rbrrrrrrfinished)keys)rr rurKrrranger{itemsappendrmapr_delayedcomputerwaitstatuspartition_order enumeratekey schedulerwho_hasrnextiterrrresult)"rrrrrrrrrrry_partsw_parts margin_parts qid_partsll_partslu_partsr packed_partsi part_dictrvalue delayed_parts fut_partspartkey_to_partitionrrworkersrrrrs" ` @@@@rcrzDaskDMatrix._map_local_dataus s) (+ 48I KN    (/ (d6l (  x8 Xd6l=S  T"u%w'#K0  %  12 12*0'): &$v,!7 &s &t & GW%GX&L-0Iu%H12H12 13 s7|$ +A+-I#kkm * U!&q # *    *  +(,C l,K'L #).."? y))) :D;;*, 9dkk 9, : " + /GAt-.D  * /8AAtDHHdNAA4:4D4D4L4L'01t$((15M5 / /:$.? #MMO JLC tDM* + 2 23CC3H I J%  "#'D  *0)H)O)O)Q#QD  9 *B1/ $RsVF!K>(K-)K>A K> K0K>:K5 K>K:BK>!K<" K>0 K><K> worker_addrc |j|j|j|j|j|j j |d|jdS)z\Create a dictionary of objects that can be pickled for function arguments. N)rrrrrrr)rrrrrrgetr)rrs rc_create_fn_argszDaskDMatrix._create_fn_argssV "//!//#33||"&"9"9__((d;++  rbc|jS)zxGet the number of columns (features) in the DMatrix. Returns ------- number of columns )rrs rcnum_colzDaskDMatrix.num_cols||rbr])rrrrr3rrLrKfloatboolr$r%rr rrrur r rrr rarbrcrRrRs6 ,0 9 -115#'0404+/)-7;7;59#(#9 -.9 9 ( 9 ) 9 o.9 %9 9  -9  -9 (9 o &9 $O49 $O49 "/2!9 "!#9 $ %9  9 v&9&,0-115)-597;7;x%x x ( x /* xo.xo &x"/2x$O4x$O4x xt  3  4S>  rbrR_MapRetT_Pfuncrefsrc Kt|}g}|D]}g}|D]D}t|tr!|j|j |4|j|Fdt dt jdt jdttffd }|jtt|||g|d|gdd} |j| d tttdttfd } t!j"|} | j%| | d {} |j'| j)d {} | S7-7w) z.Map a function onto partitions of each worker._addressrkwargsr[ctj}|j|k7rtd|jd|d|i|gS)NzInvalid worker address: z , expecting zz. This is likely caused by one of the workers died and Dask re-scheduled a different one. Resilience is not yet supported.)rrrr)rrrrrs rcfnz!map_worker_partitions..fn s^ ++-F~~) .v~~.>l8*UUU $)&)* *rbFT)purerallow_other_workersresultsc|D]}||cSyr]ra)rvs rc first_validz*map_worker_partitions..first_validFs  A} rbN)rrsrRrrrur rrrr submitrrrrdb from_delayed reductionrr) rrrrfuturesrrrefrfutrrrs ` rcmap_worker_partitionsr" sO FG- !C#{+ C//56 C  ! + +RWW + +d8n +:fmm 72t,b 1  F $   s[-^Xhx&89hx>P //' "C k;7 7C>>#&--/ /F M 8 /s$D&E)E*&EEEEc)(eZdZdZe ddddddddddddddddddeddedeed eed eed eed e d ee dee e e e fdeedeedeedeedeedeedeede deeddf&fdZdedeee fffd ZxZS)rSzmA dask version of :py:class:`QuantileDMatrix`. See :py:class:`DaskDMatrix` for parameter documents. NF)rrrrrrmax_binr rrrrrrmax_quantile_batchesrrrrrrrrrrr$r rrrrrrr%r[ct||||||| | ||||||| || |_||_d|_| t | |_yd|_y)N)rrrrrrrrrrrrrrrT)rrr$r%rid_ref)rrrrrrrrrrr$r rrrrrrr%rs rcrzDaskQuantileDMatrix.__init__Ysr. #//+''1  " $8!.1o2c7 4 rbrct||}|j|d<|j|d<|j|j|d<|S)Nr$r%r )rrr$r%r()rrrrs rcrz#DaskQuantileDMatrix._create_fn_argssLw&{3,,Y'+'@'@ #$ 99 ))DK rbr])rrrrr3rrLrKr r r$rr rrrRrrur rrrs@rcrSrSSs  ,0 *H -115#'049=!%%)+/)-7;7;59#(.2)*H-.*H*H( *H ) *Ho.*H%*H*H -*H c49n 56*H#*Hk "*H(*Ho &*H $O4!*H"$O4#*H$"/2%*H&!'*H('sm)*H* +*H *HX34S>rbrSrrc 2|r tdi|Stdi|SNra)rGrF)rrs rc_dmatrix_from_list_of_partsr,s!'1&11  $V $$rbdconfigcoll_cfgcK| tn|}d}d}t||\}}|||f}nd} tjj |j j }|jd}|jt||||jd{}|j|}|S#t$rd}YKwxYw7(w)zBGet rabit context arguments from data distribution in DaskDMatrix.Nrz/:) CollConfigrIrcommget_address_hostrrstrip Exceptionrun_on_schedulerrtracker_timeoutget_comm_config) rrjr-r.rorp user_addr sched_addrrs rc_get_rabit_argsr:s (/z|XH!GD)'8;C=B>>CcDtjjddS)Nxgboost)default)r_r#rrarbrc_get_dask_configr>s ;;??9d? 33rbdtrainevalsc^t|jj}|r~|D]y}t|dk(sJt |dt rt |dt sJ|d|urCt|djj}|j|}{t|S)Nr!rrE) setrrr{rsrRruunionr)r?r@ X_worker_maprrs rc_get_workers_from_datarEs!!2!2!7!7!9:L  :Aq6Q; ;adK0Z!c5J JJtv~QqT__1134J'--j9L  :  rbcK|jjd{}|dj}t||z }|rt d|y76w)NrzMissing required workers: )ridentityrrB RuntimeError)rrinfocurrent_workersmissing_workerss rc_check_workers_are_aliverLs`!!**, ,D9o**,O'l_4O77HIJJ -sAA7A train_reftrain_idevals_id evals_name n_threadscLtdi|d|i}g}t|D]\}} |||k(r|j|||f%| jdd'| d|k7r t d| d=tdi| ||d} ntdi| d|i} |j| ||f||fS)Nnthreadr zSThe training DMatrix should be used as a reference to evaluation `QuantileDMatrix`.)rSr ra)r,rrrr) rMrNrOrPrQrXyr@rr eval_Xys rc_get_dmatricesrVs % Dy D) DB')ED//3 A;( " LL"jm, -  775$  +5zX% *E 1SCSPRSG1KCKKG gz!}-./ u9rb global_configparamsnum_boost_roundobjearly_stopping_rounds verbose_eval xgb_model callbacks custom_metricc6 Kt||}t||d{t|t||| d{}t |dt dt t tt tffdtdtt dttdtdtd ttf  fd }tj||4d{|A|Dcgc]\}}| }}}|Dcgc]\}}| }}}|Dcgc] }t|}}ng}g}g}t!||||t|||g|g|zd |id{}|cdddd{S7A7$7cc}}wcc}}wcc}w707 #1d{7swYyxYww) N)r-r. parameters coll_argsrNrPrOrMrr[ctj}|j}t||} |j | | di} j d| it di|5t jdi5t||g|||| d\} } t|| | t| dk7r| nd } dddddd jdk7r | d}|Sd}|S#1swY/xYw#1swY3xYw)N)rSn_jobsrS)rOrPrQr) rXr?rY evals_resultr@rZr_r[r\r]r^rNra) rrcopyrJryrQr#config_contextrV worker_trainr{num_row)rarbrNrPrOrMrr local_paramrQ local_historyrTr@rOrr^r_r[rWrYrZr\r]s rcdo_trainz_train_async..do_trains3'') oo' !+v6 yIFG35 i34 -9 - v/D/D/U}/U &"%# IB#" /*"5zQeD+&;)## G  4 ::<1 "(+C  C C    s$&C2<locals) rrXr?rYr@rZr[r]r\r^r_r.s rcrDrDhsHb F 6;; '') "  (  rbis_df output_shapec&|xrt|dkSNr!r)rurvs rc_can_output_dfrys  +S&!++rbr predictioncolumnsct||jrt|dd}t|ddrZddl}|j dk(r"|j i|tjS|j ||tj|}|S|j dk(rt i|tj|St ||tj|}|S)z0Return dataframe for prediction when applicable.indexNzcudf.core.dataframer+r)r{dtype)r{r~r}) ryrgetattrr,cudfsizer+rfloat32)rrzr{rur}rs rc_maybe_dataframerseZ--. gt, 4!6 D !#~~b'~OOG5==(J  !# WEMMQVWW"G5==J rbmapped_predictrOzdistributed.Futurer.rc XKt|j}t|dk\r%t|tj r t dtt|tj |r|+t|tjr|j}n|}t j|||d||tjj|}t|dk(r|jdddf}|S|;t|tjtj fr|j!} n|} t|dk(rdg} g} ndg} t|tj r!t#t%t|dz } n't%t|dz D cgc]} | dz } } t|dk(r1t#|j&} t| t"sJ|df| d<nd} tj(|||d|| | | | t*j, }|Scc} ww) NzGUse `da.Array` or `DaskDMatrix` when output has more than 2 dimensions.T)rrErr!F)chunks drop_axisnew_axisr~)rtrr{rsrr+rryrrto_dask_dataframemap_partitionsutils make_metailocr to_dask_arrayrrr map_blocksrr)rrOrrrvrr{base_margin_df predictionsbase_margin_arrayrrrrs rc_direct_predict_implrsDIIK G <A*T2<<"@  U  jr||4lC  "z+rxx'H--/ )N''      ##D)  |  !%**1a40KX U  "z "))R\\2( 5@4M4M4O  +  |  !01sI.0HI$ -c,&7!&; <=+0\1BQ1F+GHaAEHH |  !-1,=Ffd+ ++%a*F1IFmm      --   3IsF%H*' H%3A7H*featuresinplacec t|tsJtjj d}|j d|}|r)|j }|jddk(rd|d<t|d}|j|fdd i|}t|jdkDr|jdnd} i} t||jrt| D]} d | | < |j| fS) z@Create a dummy test sample to infer output shape for prediction.irE predict_typemarginT output_margin)rvalidate_featuresFf4)rsrrrandom RandomStaterandnrfpopr.rYr{rryr) rOrrurrrng test_samplem test_predt n_columnsrrs rc_infer_predict_outputrs h $$ $ ,, " "4 (C))Ax(K ::n % 1&*F? # 5A FeFvFJ'*:+;+;'CCrr pred_leaf pred_contribsapprox_contribspred_interactionsriteration_range strict_shapec  ()Kt||d{} t|ttjt j fsAttttjt j gt|dtdtdtdttdtdtf    f d }t|tjt j frx|j|j!t"| |j$dt|t j d  d{\}}t'|| |d|| d{S|j|j!t"| |j)d d  d{\}}|j*}|j,(|j.)|j0dtd t2t4tfdt6j8f()   f d }g}g}g}g}t;|j<j?}|D]l}|j<|}|jA||jAtC||gz|jA|Dcgc]}||jDc}ntG||D],\}}|j!d||g}|jI|.t;tG||||}tK|d}|D !cgc] \}} }!}| }}!} }}|D !cgc] \}} }!}|  }}!} }}|D !cgc] \}} }!}| }}!} }}g}"tG||D],\}}|j!|| ||g}#|"jI|#.g}$|jM|d{}tO|D]D\}%}&|$jItjP|"|%|&f|ddzt6jRFtjT|$d}'|'S7777cc}wcc}}!} }wcc}}!} }wcc}}!} }w7w)NrO partitionrur{_r[c  tjdi5t| d}|j|     }t ||||}|cdddS#1swYyxYw)NT)rrr) rrrrrrrrrra)r#rgr.rYr)rOrrur{rrpredtrrWrrrrrrrrs rcrz&_predict_async..mapped_predictZs " " 3] 3 #'A OO+#+ /"3"3 /)$ E%YwFE%   s 8AA"rEF) rrurrrrrrrrrOrrrvr) rOrrurrrrrrrrc  |d}|jdd}tjdi 5t| |d}|j |    }|cdddS#1swYyxYw)NrrT)rrrrr)rrrrrrrrra)rr#rgr.rY)rOrrrrrrrrrWrrrrrrrrs rcdispatched_predictz*_predict_async..dispatched_predictsF|hh}d3  " " 3] 3 '++#' AOO+#+ /"3"3 /)$ E)   s -A%%A.c&|djdS)Nrr)r)rs rcz _predict_async..stF|'9'9!'<rb)rc |dSrxra)ps rcrz _predict_async..s adrb)r)rr~raxis)+rrsrRrrrr+rr4rr-r r rrrrrrrr rrrrr rurndarrayrrrextendr{rziprsortedgatherrrr concatenate)*rrWrrrrrrrrrrr_boosterr _output_shaperrvrrr all_parts all_orders all_shapes all_workersworkers_addressr list_of_partsrwsparts_with_orderrorderrfarraysrrowsrrrs* ` ````````` @@rc_predict_asyncrFs) 'vu55H d["((BLLA Bbhh EtDzRSS%(15@DS NQ 0$2<<01$*NN MM%A r||4+#+ /"3)  %   t*)&    #NN !\\^''+/%  OL! **O&&M&&MllGG4S>emm2IJJK4??//12O&Q  4 '3}- =>O?4884OP Q {I.4 MMQ QQ &AQ>Q?A0Q1Q4Q7Q:QF)rr) rrrrrrrrrc ~t|}|jtfdtjit S)aRun prediction with a trained booster. .. note:: Using ``inplace_predict`` might be faster when some features are not needed. See :py:meth:`xgboost.Booster.predict` for details on various parameters. When output has more than 2 dimensions (shap value, leaf with strict_shape), input should be ``da.Array`` or ``DaskDMatrix``. .. versionadded:: 1.0.0 Parameters ---------- client: Specify the dask client used for training. Use default client returned from dask if it's set to None. model: The trained model. It can be a distributed.Future so user can pre-scatter it onto all workers. data: Input data used for prediction. When input is a dataframe object, prediction output is a series. missing: Used when input data is not DaskDMatrix. Specify the value considered as missing. Returns ------- prediction: dask.array.Array/dask.dataframe.Series When input data is ``dask.array.Array`` or ``DaskDMatrix``, the return value is an array, when input data is ``dask.dataframe.DataFrame``, return value can be ``dask.dataframe.Series``, ``dask.dataframe.DataFrame``, depending on the output shape. rW)rrrr#rsrt) rrrrrrrrrrrrs rcrYrYs5f F 6;;~ UV5F5F5H UFH UUrbrc  Kt|}t||d{} t|tjt j fs.mapped_predict-sm " " 3] 3  00 /)'"3)1J &iWeL   s AArET)rrurrrrr)rrrsrrrr+rr4rrr-r r rrrrrrr)rrWrrrrrrrrrOrrrs ` ```` ` rc_inplace_predict_asyncrs F%fe44G dRXXr||4 52<< 8$t*EFFz rxxryy1(2<< CT+EVWXXc    * ! ZZ]T2<<0%+%    KE4&%    S5:  s4GF?E:GGG:G;GGGr)rrrrrrc~t|}|jtfdtjit S)aInplace prediction. See doc in :py:meth:`xgboost.Booster.inplace_predict` for details. .. versionadded:: 1.1.0 Parameters ---------- client: Specify the dask client used for training. Use default client returned from dask if it's set to None. model: See :py:func:`xgboost.dask.predict` for details. data : dask collection. iteration_range: See :py:meth:`xgboost.Booster.predict` for details. predict_type: See :py:meth:`xgboost.Booster.inplace_predict` for details. missing: Value in the input data which needs to be present as a missing value. If None, defaults to np.nan. base_margin: See :py:obj:`xgboost.DMatrix` for details. .. versionadded:: 1.4.0 strict_shape: See :py:meth:`xgboost.Booster.predict` for details. .. versionadded:: 1.4.0 Returns ------- prediction : When input data is ``dask.array.Array``, the return value is an array, when input data is ``dask.dataframe.DataFrame``, return value can be ``dask.dataframe.Series``, ``dask.dataframe.DataFrame``, depending on the output shape. rW)rrrr#rsrt) rrrrrrrrrs rcrZrZXsBj F 6;; .4.?.?.A EKX rbdevice tree_methodr$c,Kdttdtdtffd }td d|i|\}}|d{}|||fSg}|D]<} | d|ur|j | |j | dd{| df>||fS7Q7w) z(A switch function for async environment.r rr[cTtrtd|d|Stddi|S)N)rr r$rra)r=rSrR)r rrrr$rs rc _dispatchz2_async_wrap_evaluation_matrices.._dispatchsC  V ,&3;A 3&3F33rbcreate_dmatrixNrrEra)rrRr rAr) rrrr$rr train_dmatrixr@awaitedrs ```` rc_async_wrap_evaluation_matricesrs4x ,44 445XIXQWXM5''M }e##G + Q4= NN1  ad AaD)* + ' !!(#s!8BB=B<B =BBDaskScikitLearnBasec#JK ||_|d|_y#d|_wxYww)z-Temporarily set the client for sklearn model.N)r)rrs rc_set_worker_clientrs%    t s# # #cTeZdZdZdZdddeededdffdZde d e d e d ee d ee def d Z edddddde d e d e d ee d ee def dZ dde d ee defdZ dde d ee defdZdeefdZdefdZeddZej0ddZdededefdZxZS)rz._!s Ks)r r _client_syncr)rrs` rcrzDaskScikitLearnBase.__await__s/ 3   #--//rbcH|jj}d|vr|d=|S)N_client)__dict__rf)rthiss rc __getstate__z DaskScikitLearnBase.__getstate__&s(}}!!#  Y rbc0t|j}|S)zThe dask client used in this model. The `Client` object can not be serialized for transmission, so if task is launched from a worker instead of directly from the client process, this attribute needs to be set at that worker. )rr)rrs rcrzDaskScikitLearnBase.client,sT\\* rbc<| |jnd|_||_y)NF) asynchronous _asynchronousr)rclts rcrzDaskScikitLearnBase.client7s25S--e rbrc |j}t|dd} tjd}|rWtj 5}t ||5}|jj|fi|d|i}|cdddcdddS|jj|fi|d|jjiS#t$rd}YwxYw#1swYnxYwcdddS#1swYfxYw)zGet the correct client, when method is invoked inside a worker we should use `worker_client' instead of default client. NrFTr) rrrrr worker_clientrrrr)rrrr in_workerrrrs rcrz DaskScikitLearnBase._client_sync>s << "4%@L "&&(  ..0F+D&9#T.dkk.. $*9E # ## t{{VVT[[=U=UVV "!  "##  s5B? C'#C7 C'? C  C C C''C0r])r[r)rrr[N)rrrrrrr0r rrLr rKr&rr3rYrrr rr rpropertyrsetterr rrrs@rcrrsFG;?!HZ$8!3!SW! %% %  % o. %".1% %N $"&1548      o.  ".1     *59 ".1 059W W".1W W09S>0d  ]] WWSWSWrbz3Implementation of the Scikit-Learn API for XGBoost. estimatorscReZdZdZdededeedeedeeeeefdeeedeeed e e e fd ee e e fd eed efd Zeddddddddddededeedeedeeeeefd ee e e fd ee e ee fdeeedeeed eed dfdZy)rTzAdummy doc string to workaround pylint, replaced by the decorator.ry sample_weightreval_setsample_weight_eval_setbase_margin_eval_setverboser]rr[c0K|j} |j| | | \} } } } tdid|jd|jd|j d|j d|d|ddddd |d |d | d |d |d|ddddd|jd|jd|jd{\}}t|jrt|j}nd}|jjtd|jtj t#| ||j%||| ||j&|j(|j*| d{}|d|_|j/|d|S77&w)Nrrrr$rr rrr rrr r r eval_groupeval_qidrrrTrrrWr-rXr?rYr@rZr_r\r[r^r.r]rOrPra)get_xgb_params_configure_fitrrrrr$rrrcallable objectiver@rrqr#rsr>get_num_boosting_roundsr[r^r._Booster_set_evaluation_resultrrr r rr r rrr]rrXrmetricr?r@rZrs rc _fit_asynczDaskXGBRegressor._fit_async]s$$&151D1D v2 .vv> ;; ;; (( LL        ( $ ,  $: "6  ! "LL# $ $66% &,,'  , DNN #&:4>>&JCC (( ;; ++-$& 88:  "&"<"<nn]]!)  $ *  ##GI$67 ] 4 s%B4F6F7B6F-F.%FFNTr rr rr]r rrctjD cic] \} } | dvs | | } } } |j|jfi| Scc} } wNrrrtrrrrrr r rr rr]r rrkrrs rcfitzDaskXGBRegressor.fitsR"(!1TAQ>S5S1TT t  9D99U AA)rrrrrLrKrrrrrr r-r9rr3rur%rarbrcrTrTWsLA A A  0 A o. A8E/?*J$KLMA!)/)B CA'x'@AAsDy!AE'8"345A"/2A AH 4815PT.2=AFJDH59: : :  0 : o. :8E/?*J$KLM:%T *+:E'3"89::!)/)B C:'x'@A:"/2: : :rbrTzBImplementation of the scikit-learn API for XGBoost classification.ceZdZdededeedeedeeeeefdeeedeeedee e fd eee e fd eed dfd Z d d d dd d d d ddededeedeedeeeeefdeee e fd eee ee fdeeedeeed eed dfdZdede deedeed ef fd Z ddede deedeed ef dZej&j*e_dede de deedeed ef fd ZxZS)rUrr r rr r rrr]rr[c.K|j} |j| | | \} } } } t|jfid|jd|j d|j d|d|ddddd|d |d | d |d |d |ddddd|jd|jd|jd{\}}t|tjr<|jjtj|d{|_n6|jj|j!d{|_t#|jr|jj%|_t'|jr|jj)|_t+j,|j|_t/|j|_|j0dkDrd| d<|j0| d<nd| d<t3|j4rt7|j4}nd}|jj9t:d|jt=j>tA| ||jC||| ||jD|jF|jH| d{}|d|_%t3|j4s | d|_|jM|d|S7V777Lw)Nrrr$rr rrr rrr r rrrrrrr!zmulti:softprobr num_classzbinary:logisticTrrOrP)'rrrrrrr$rrrrsrrruniqueclasses_drop_duplicatesr5to_cupyr6rrrr{ n_classes_rrr@rrqr#rsr>rr[r^r.rrrs rcrzDaskXGBClassifier._fit_asyncs$$&151D1D v2 .vv> KK ;; (( LL        ( $ ,  $: "6  ! "LL# $ $66% &,,'  . a ""&++"5"5biil"CCDM"&++"5"5a6G6G6I"JJDM  & MM113DM $-- ( MM--/DM DMM2 dmm, ??Q "2F; "&//F; "3F;  DNN #&:4>>&JCC (( ;; ++-$& 88:  "&"<"<nn]]!)  $ * '#K0DN ##GI$67 E 0DJ$ sKB2L4L 5ALL 6L>L?FLLAL LLLNTrctjD cic] \} } | dvs | | } } } |j|jfi| Scc} } wr r"r#s rcr%zDaskXGBClassifier.fitsR"(!1TAQ>S5S1TT t  9D99Ur&rrcK|jdk(r tdt| |d|||d{}t t t jdt j}tt|dd||S7Ow) Nz multi:softmaxzSmulti:softmax doesn't support `predict_proba`. Switch to `multi:softproba` insteadF)rrrrrT)allow_unknown_chunksizesr.r) rrrrrrrvstackr?r)rrrrrrr2rs rc_predict_proba_asyncz&DaskXGBClassifier._predict_proba_async#s >>_ ,6 w-/#+ .   BII =ryy "'$ a"@&&QQ s2BBABcB|j|j||||S)N)rrrr)rr3)rrrrrs rc predict_probazDaskXGBClassifier.predict_proba<s1   % %/#+ !  rbrrctKt ||||||d{}|r|St|jdk(r|dkDj t }|St|jdk(sJt |tjsJdtdtfd}tj||d}|S7w) NrrEg?r!xr[c&|jdS)NrEr)argmax)r7s rc_argmaxz1DaskXGBClassifier._predict_async.._argmaxhsxxQx''rb)r) rrr{rastyperrsrrr r) rrrrrr pred_probspredsr:rs rcrz DaskXGBClassifier._predict_asyncMs!71 '/#+ 2     z A %#%--c2E z''(A- --j"((3 33 (3 (3 (MM':CE - sB8B6BB8)TNN)rrrrLrKrrrrrr r-r9rrur%r&r3r r5r7rrrrs@rcrUrUs U U U  0 U o. U8E/?*J$KLMU!)/)B CU'x'@AUsDy!UE'8"345U"/2U Uz4815PT.2=AFJDH59: : :  0 : o. :8E/?*J$KLM:%T *+:E'3"89::!)/)B C:'x'@A:"/2: :"R R Ro. R ".1 R  R8#'1548       o.  ".1   *77??M    o. ".1 rbrUzZImplementation of the Scikit-Learn API for XGBoost Ranking. .. versionadded:: 1.4.0 aq allow_group_split : .. versionadded:: 3.0.0 Whether a query group can be split among multiple workers. When set to `False`, inputs must be Dask dataframes or series. If you have many small query groups, this can significantly increase the fragmentation of the data, and the internal DMatrix construction can take longer. zf .. note:: For the dask implementation, group is not supported, use qid instead. )extra_parametersend_notec 2eZdZedddddededeeded df fd Z d e effd Z d e d e dee dee dee deeee e fdeee deee deee deeefdeeeefdee d dfdZeddddddddddddd d e d e dee dee dee dee deeee e fdeee deee deeeefdeeeeefdeee deee dee d dfdZej*j.e_xZS)rVz rank:pairwiseFN)rallow_group_splitr.rrAr.rr[c dt|r td||_t|d||d|y)Nz5Custom objective function not supported by XGBRanker.)rr.ra)rrrArr)rrrAr.rrs rcrzDaskXGBRanker.__init__s8 I TU U!2 J9xJ6JrbcFt|}|jd|S)NrA)r_wrapper_paramsaddrrXrs rcrDzDaskXGBRanker._wrapper_paramss"(* &' rbrr rr rr r rrrr]rc K|j} |j| | | \}}} } t|jfid|jd|j d|j d|d|ddd|d|d |d | d |d |d |ddd| d|jd|jd|jd{\}}|jjtd|jtjt| ||j|d|| |j |j"||j$d{}|d|_|d|_|S77w)Nrrr$rr rrr rrr r rrrrrrT)rrrWr-rXr?rYr@rZr_r\r[r^r]r.rOrP)rrrrrrr$rrrrrqr#rsr>rr[r^r.r evals_result_)rrr rr rr r rrrr]rrXrrr?r@rs rcrzDaskXGBRanker._fit_asyncs $$&151D1D v2 .vv> KK ;; (( LL        ( $ ,  $: "6  ! "LL# $ $66% &,,'  * (( ;; ++-$& 88:  "&"<"<nn]]!)  $ * $Y/ S * s%B2E4E5B E>E?EE) rrr rr rrrr]r rrrrc 2d}|| t|| tddtdttjfd}dt tdt dtt tjfd}|jso||r$||dr||d r||d r ||d sJ||Jt|}t|j||||| \}}}}}| g}g}g}g}| sJt|D]\}\}}| r| |nd}| r| |nd}||sJ| sJ| |}| r$||dr||d r||d r ||d sJ||Jt||k7r!t|j|||||\}}}}}n |||||f\}}}}}|j||f|j|||j|||j||}|} |r|nd} |r|nd} |j|j||||||| | | | | | S)Nz=Use the `qid` instead of the `group` with the dask interface.z`qid` is required for ranking.rr[cNt|tjs tdy)NzJWhen `allow_group_split` is set to False, X is required to be a dataframe.T)rsrr+r)rs rccheck_dfz#DaskXGBRanker.fit..check_dfs'a.$rbrrcZt|tjs|td|dy)Nz*When `allow_group_split` is set to False, z is required to be a series.T)rsrrr)rrs rc check_serz$DaskXGBRanker.fit..check_sers:c299-#/@G!!rbr r r)r r r) rr rr rr rrr]r rr)rrKrrr+rrurrAr'rHrrrrr)rrr rrr rr rrrr]r rrrrKrMX_id new_eval_set new_eval_qidnew_sample_weight_eval_setnew_base_margin_eval_setrXeyewebeqes rcr%zDaskXGBRanker.fits&N *"4S/ ! ;=> >  Ibll,C  /* 25  x * + %% c5)a%m_=k=9  : ?q} 44a5D4B +' 5 1AsA}k#! ! -/*+-(x#,X#6bn<<"v~-; KKRR.*BB./Q {-R*BB ''R1 ''+~299"=~077;7<:('2L.RV'1I,d%  OO'##9!5+!  rb)rrrr3rur rr0r rrrDrLrKrrrrr9r-rr%r:rrrs@rcrVrVos2 )"')- K K K :& K  K  K  KS = = = o & =  0 =o.=8E/?*J$KLM=!)/)B C='x'@A=8O45=sDy!=E(G"345="/2= =@ ,0)-3715PT:>8<.3=AFJDH59!w w  w ( w o & w  0w o.w 8E/?*J$KLMw Xo67w 8O45w %T *+w E(C"89:w !)/)B Cw 'x'@Aw "/2!w " #w  w v--''CKKrbrVzjImplementation of the Scikit-Learn API for XGBoost Random Forest Regressor. .. versionadded:: 1.4.0 rzI n_estimators : int Number of trees in random forest to fit. )r>cFeZdZedddddddeedeedeed eed eed ed dffd Zd e e efffd Z d e fdZ ddddddddddededeedeedeeeeefdeee efdeeee efdeeedeeedeed dffdZxZS)rWrE皙?h㈵>N learning_rate subsamplecolsample_bynode reg_lambdar.r\r]r^r_r.rr[c 0t|d|||||d|yNr[rarrrr\r]r^r_r.rrs rcrzDaskXGBRFRegressor.__init__g0  '-!    rbcBt|}|j|d<|SNnum_parallel_treerr n_estimatorsrFs rcrz!DaskXGBRFRegressor.get_xgb_params{&')&*&7&7"# rbcyNrErars rcrz*DaskXGBRFRegressor.get_num_boosting_roundsrbTrrr r rr rr]r rrctjD cic] \} } | dvs | | } } } t|j|jt |di| |Scc} } wNr!rartrr>r[r^rr%rrr r rr rr]r rrr$rrrs rcr%zDaskXGBRFRegressor.fit_"(!1TAQ>S5S1TT455t~~F  d U A#A#rrrr3rr r0r rr rurrrrLrKrrrr r-r9r%rrs@rcrWrWZs *+%(,/&*)-   E?  #5/  UO  :&      &S#X 4815PT.2=AFJDH59    0  o. 8E/?*J$KLM%T *+E'3"89:!)/)B C'x'@A"/2 rbrWzkImplementation of the Scikit-Learn API for XGBoost Random Forest Classifier. .. versionadded:: 1.4.0 cFeZdZedddddddeedeedeed eed eed ed dffd Zd e e efffd Z d e fdZ ddddddddddededeedeedeeeeefdeee efdeeee efdeeedeeedeed dffdZxZS)rXrErYrZNr[r\r]r^r_r.rr[c 0t|d|||||d|yrarbrcs rcrzDaskXGBRFClassifier.__init__rdrbcBt|}|j|d<|SrfrhrFs rcrz"DaskXGBRFClassifier.get_xgb_paramsrjrbcyrlrars rcrz+DaskXGBRFClassifier.get_num_boosting_roundsrmrbTrrr r rr rr]r rrctjD cic] \} } | dvs | | } } } t|j|jt |di| |Scc} } wrorprqs rcr%zDaskXGBRFClassifier.fitrrrsrtrs@rcrXrXs *+%(,/&*)-   E?  #5/  UO  :&      &S#X 4815PT.2=AFJDH59    0  o. 8E/?*J$KLM%T *+E'3"89:!)/)B C'x'@A"/2 rbrX)NN) )rlogging collectionsr contextlibr functoolsrrr threadingrtypingr r r r r rrrrrrrrrrrrr_rrrrrrrr dask.delayedrrpackaging.versionrr r^r"r#_typingr$r%r&callbackr'r(r0r)CollArgsr*rcompatr+r,corer-r.r/r0r1r2r3r4rr5r6sklearnr7r8r9r:r;r<r=r>r?r@rArBtrackerrCtrainingrDrhrFrGrHrrIrJrrrK__annotations__rLrM__all__ getLoggerr|rdr rgrirrur~rrQrrRr r r"rSr,r:r>rErLrmrVrqryrrrrr rrrYrrZrrrrTrUrVrWrXrarbrcrs 2f#%44(   %4!@@'-*1/   0    #,KK7"288R\\299#DEE"((BLL01 1  F   + ,+w++999888-- hsmXeCHo%>>? @-c]- -`SMU38_-c]    U*88 U!56;O"ll^ : t_D ) *D 2x< D Dc] D  DN9+9x%T%S%W%)-%) % %%d38n %%z" %  #uS#X  %P4(4S>24    (% S8H2I)J K  #Y K #YK 4K K sm      7Dw|,- -.>e eS>ed38n % e cN e  ee HU;#345 6e ) e$C=eT "e e!123eF#ez"e l!eP  6 :>#+/#'%)6:&*%)6 6 cN6 6 6 HU;#345 6 6 ) 6$C=6 6T "6!1236F#6z"6 66r,$,e,, )-c;?:RR"R  R /* R S/ R sCx.RRj" " #",0";?"KN" 5c?DcN *+", ).w>R/R)S0W WS>W $ 44 5W  W  WWWWWWW$WWWt  YY!#"&,3V ) *3V w(<< =3V  V# $3V  3V  3V3V3V3V3V3V$3V3V 3V3Vl> >S>> $ 44 5 >  > $ >>>>/*>>>B '-YY"-1: ) *: w(<< =: : $ :  :::/*:: ::z" ) *" SM"#"c] "  "  ;eK,<&=!>? ?@ ": *>UW(UWp= g?VW:')<W:W:tH7t)+>t tn 7 %0P(N$7P(10P(f  k /)/ /d  k /+/ /rb