-i0SSKrSSKrSSKrSSKJrJr SSKJr SSKJ r J r J r J r SSK Jr SSKJrJr SSKJrJrJrJrJrJrJrJrJrJrJrJr SSKJ r SS K!J"r"J#r# \$"\RJ"55r&\&Vs/sH o"5PM snr'\'\"S S 9\"S S 9\"S S9\"SS9\"SS9\"SS9\"SS9\"SS9\"SS9\"SS9\"SS9/ - r'Sr(SSjr)Sr*\RVRYS\'\(S9S5r-\"5/SQ\R\*\R\/4\"5/SQ\R\*\R\/4\"5/SQ\R\*\R\/4\"5/SQ\R\*\R\/4\"5SS/\R\*SS\R\/4\"5SS/\R\*SSS\R\/4\"SS9SS/\R\*\R\/4\"SS9SS/\R\*\R\/4\"SS9SS/\R\*SS\R\/4\"SS9SS/\R\*SSS\R\/4\"SS9SS/\R\*SSS\R\/4\"SS9SS/\R\*\R\/4\"SS9/SQ\R\*\R\/4\"SS9SS/\R\*SS\R\/4\"SS9SS/\R\*SSS\R\/4\"SS9SS/\R\*SSS\R\/4\"5/S Q\R\*S!S\R\/4\"5/\R\*S!S"\R\/4/r/\"5S//4\"5S//4\"SS9/S#Q/4\"SS9S$S//4\"SS9S//4\"SS9/S#Q/4\"SS9S$S//4\"SS9S//4\"5SS//4\"5/S%Q/4/ r0\"5/S/4\"SS9//S&Q4\"SS9//S&Q4\"SS9/S/4\"SS9//S&Q4\"SS9/S&Q/4\"SS9/S/4\"5/SS/4\"5SS'/SS/4/ r1\RVRYS(\/\0-5S)5r2\RVRYS*\/\1-5S+5r3\RVRYS,/\"5S-S.S/S0S4P\"5S-S.S1S-S4P\"S'S 9S-S.SS'S4P\"S S 9S-S.SS S4P\"S S 9S.S-S-S2S4P\"S'SS39S-S.S4SS4P\"S'SS39S-SS5SS4P\"5S5\Rh"S05S0S\Rh"S05-- SS04P\"5S5\Rh"S05\Rh"S05S'-S'S'4P\"SS9S5\Rh"S05S6SS4P\"SS9S5S1SS\Rh"S5-- SS4P\"SS9S5S1\Rh"S5S'- SS4P\"SS9S5S1S7SS4P\"5S \Rh"S05\Rj"S05S \Rh"S05-- SS4P\"5S8S9SSS4P\"5S-S9S:S!S4P\"5S8S;SSS4P\"5S-S;SS!S4P\"5S-S?S@SASB4P\"5S8S?/SB/S-Q7P\"5S-SCSDSASE4P\"5S8SC/SE/S-Q7P\"5S8SFS@SGSB4P\"5S-SF/SH/S-Q7P\"5S8S>S>SSI4P\"5S8SJSJSSK4P\"5S-S:SSS4P\"5S8S:S:SS4P\"5S-SJSSLSK4P\"5S-SMSSS4P\"5S-S\RVRYS\'\(S9\RVRYSZSS_/5Sg55r?\RVRYSh/SiQ5\#Sj55r@\RVRYS\'\(S9\RVRYSZSS_/5Sk55rA\RVRYSl\"5\RSm4\"5\RSm4\"S S 9SnSm4\"5\RSo4\"5\RSp4\"5\RSp4\"5\RSq4/5Sr5rDSsrEStrFSurG\RVRYSv\R"/SwQ5\R"/SxQ545\RVRYSy\R"/SzQ5\R"/S{Q545S|55rI\RVRYS\'\(S9S}5rJ\RVRYS\&5\RVRYSZSS_/5\RVRYS~\Rn\Rp45\RVRYSS5S5555rK\RVRYS\&5\RVRYSS~\R0S\RS34/5S55rM\RVRYS\SS0\NS4\SS0\OS4\SS"0\OS4\SS0\NS4\SS0\OS4\SS"0\OS4/5S5rP\RVRYS\'\(S9S5rQ\RVRYS/SQ5S5rRgs snf)N)assert_allcloseassert_array_equal)approx)LinearConstraintminimizeminimize_scalarnewton) logsumexp) IdentityLink_inclusive_low_high) _LOSSES AbsoluteErrorBaseLossHalfBinomialLoss HalfGammaLossHalfMultinomialLossHalfPoissonLossHalfSquaredErrorHalfTweedieLossHalfTweedieLossIdentity HuberLoss PinballLoss)assert_all_finite)create_memmap_backed_data skip_if_32bitg?)quantileg?power@c[U[5(aUnURRn[U[5(aUSUR R S3- nU$[U[5(aUSUR 3- nU$[US5(a8[UR S5(aUSUR RS3- nU$[U5$)Nz (quantile=)clossrz(power=) isinstancer __class____name__rr%rrhasattrrstr)paramlossnames P/var/www/html/venv/lib/python3.13/site-packages/sklearn/_loss/tests/test_loss.pyloss_instance_namer/4s%""~~&& dK ( ( j!4!4 5Q7 7D  i ( ( j0 0D T7 # # G(D(D gdjj../q1 1D 5zc [RRU5nUR(a[R"XR 45nUR USUSXR -S9URSS&[R"U5R[5UR -nXv4$[UR[5(aN[UR5up[R "XS/5n[R""XS/5n X4nUR USUSUS9n[UR$5up['XS5n[)XS5n UR XUS9nUR$R*S:Xa&UR$R,(a SUSSUS-2'UR$R.S:Xa&UR$R0(a SUSSUS-2'Xv4$)z9Random generate y_true and raw_prediction in valid range.rr )lowhighsizeNr4)nprandom RandomState is_multiclassempty n_classesuniformflatarangeastypefloatr&linkr r interval_y_predamaxamininterval_y_truemaxminr2 low_inclusiver3high_inclusive) r, n_samplesy_bound raw_boundseedrngraw_predictiony_truer2r3s r.random_y_true_raw_predictionrRCs ))   %C 9nn"=>!$! 1^^+"-" A 9%,,U3dnnD,  !!' dii . .+D,@,@AIC''3! -.C77DA,/0D I! 9QrcU"[R"U/5[R"U/5S9nUR[R"U/5[R"U/5S9nUR[R"U/5[R"U/5S9upUR [R"U/5[R"U/5S9upU[ USSS9:XdeU[ USSS9:XdeUb3U[ USSS9:XdeU [ USSS9:XdeU [ USSS9:XdeUbU [ USSS9:Xdegg)z7Test losses, gradients and hessians at specific values.rgV瞯<)relabsN)r7r|gradient loss_gradientgradient_hessianr) r,rQrP loss_true gradient_true hessian_trueloss1grad1loss2grad2grad3hesss r.test_loss_on_specific_valuesrs^x &*288^DT;U VE MMxx!"((N;K2L  E%%xx!"((N;K2L&LE''xx!"((N;K2L(KE F9%U; ;; ; F9%U; ;; ; }%UCCCC}%UCCCC}%UCCCCvl5AAAA r0readonly_memmapFTdtype_in dtype_out sample_weightout1out2 n_threadsc U"5nSn[UUSSSS9upU RU5n U RU5n Ub[R"S/U-US9nUb[R"XS9nUb[R"XS9nU(a$[ U 5n [ U 5n Ub [ U5nUR U U UUUS 9n UbXLdeURU U UUUS 9n UbXLdeURU U UUUUS 9upUbXLdeUbXLdeUb%UR(a[R"XS9nURU U UUUUS 9upUbXLdeUbXLdeU"XUS 9 URXS9 URXS9 [US5(aURU S9 [US5(a+URU U UUUUS9upUbXLdeUbXLdeggg)a Test acceptance of dtypes, readonly and writeable arrays in loss functions. Check that loss accepts if all input arrays are either all float32 or all float64, and all output arrays are either all float32 or all float64. Also check that input arrays can be readonly, e.g. memory mapped. rqrsre*r,rKrLrMrNNrdtype)rQrPrloss_outr)rQrPr gradient_outr)rQrPrrrr)rQrPrr hessian_outrrQrPrrQr predict_proba)rPgradient_proba)rQrPrr proba_outr)rRr@r7r| empty_likerr,rrr:rfit_intercept_onlyconstant_to_optimal_zeror)rr)r,rrrrrrrrKrQrPlgr[ps r.test_loss_dtyperxs$ 6DI9   F]]8 $F#**84N #!2(C  }}V5 }}^=*622>B  $5mDM %#  A(192 2d %#  A(192 2d   %#  DA(192 2d(192 2d D..}}^=  %# ! DA(192 2d(192 2d]S6G!!!Mt_%% .9t%&&"")' # !,qy66$ ,qy66$y'r0rangec[USSSSS9up#US:Xa0[R"SURSURSS 9n[R"U5n[R"U5n[R"U5n[R"U5n[R"U5n[R"U5n UR UUUUS 9 UR R UUUUS 9 [XE5 URUUUUS 9 UR RUUUUS 9 [Xg5 UR RUUUUUS 9 UR RUUUUUS 9 [XE5 [Xg5 URUUUUUS 9 UR RUUUUU S 9 [Xg5 [X5 g)z:Test that Python and Cython functions return same results.rrrrrr rrcrQrPrrrQrPrrrQrPrrrrQrPrrrN) rRr7rirlrr,r%rrrr) r,rrQrPout_l1out_l2out_g1out_g2out_h1out_h2s r.test_loss_same_as_C_functionsrs:   F Av||AFLLOL ]]6 "F ]]6 "F ]]> *F ]]> *F ]]> *F ]]> *FII%#   JJOO%#  F#MM%#   JJ%#  F#JJ%#  JJ%# F#F#%#  JJ%#  F#F#r0cX[USSSUS9up4US:Xa0[R"SURSURSS9n[R"U5n[R"U5n[R"U5n[R"U5n[R"U5n [R"U5n UR UUUUS 9n UR UUUUS 9n URUUUUUS 9upURUUUU U S 9unn[X5 [X5 [R"X5(de[X5 [R"X5(de[X5 [X5 [X5 [R"X5(de[X5 [R"X5(de[X5 [R"X5(de[US 5(aUR(de[R"U5n[R"U5nURUUUUUS9unn[U U5 [U U5 [UU5 [[R"USS9SSS9 gg)zTest that loss and gradient are the same across different functions. Also test that output arguments contain correct results. rrrrrr rrcrrrrrrQrPrrraxisdy=)rtolN)rRr7rirlrr,rrrrr shares_memoryr)r:rsum)r,rglobal_random_seedrQrPrrrrout_g3out_h3l1g1l2g2g3h3out_g4 out_probag4probas r. test_loss_gradients_are_the_samer,s:   F Av||AFLLOL ]]6 "F ]]6 "F ]]> *F ]]> *F ]]> *F ]]> *F %#  B %#  B   %#  FB " "%# #FBBr"  B ' '' 'r"  B ' '' 'BBr"  B ' '' 'r"  B ' '' 'r"  B ' '' 't%&&!!!!~.MM.1 '')' ( E F#By)u1-qu='r0onesr8c 8Sn[UUSSUS9upEUS:Xa$[R"U[RS9nOK[RR U5nUR US9R[R5n[URUUUS9UURUUS S9-5 URUUS S9upxURUUUS9up[Xq-U 5 UR(d[X-U 5 O[XS S 2S 4-U 5 URUUS S9upURUUUS9upUR(d[X-U 5 [X-U 5 g [XS S 2S 4-U 5 [XS S 2S 4-U 5 g ) zTest sample weights in loss, gradients and hessians. Make sure that passing sample weights to loss, gradient and hessian computation methods is equivalent to multiplying by the weights. rsrrrr)rlrr5rN) rRr7rfloat64r8r9normalr@rr,rr:r) r,rrrKrQrPrOlossesr losses_sw gradient_swhessian hessian_sws r.test_sample_weight_multipliesr}sI9   FirzzB ii##$67 299"**E  )'   )))   ))%*F "//%#0I F*I6   0+>D#99;G--%.H #33%#4K   0+>/<D#99;G4"88*Er0c[USSSSS9upURS:XaUSS2S4n[URXS9URXS95 [UR XS9UR XS95 [UR XS9UR XS95 [UR XS9UR XS95 gg) z5Test that reshaped raw_prediction gives same results.rrrrrr Nrg)rRndimrr,rrr)r,rQrPraw_prediction_2ds r.test_graceful_squeezingrs:   Fa*1d73 IIVI F IIVI C     f  O   f  L   MMM J MMM G    ! ! ! R  ! ! ! O  r0cUR(d[R"/SQ5n[UR[ 5(aSnUR RnUR R(dXC-nUR RnUR R(dXS- n[R"X$U5nURRU5nO[R"UR5R[ 5n[R""URUR4[R$"S5*[ S9n[R$"S5UR&SSURS-2'US:Xa0[R("SUR*SUR*SS 9nUR-UUUS 9nUR/XaS 9n[1Xx*S S S9 g)zrTest value of perfect predictions. Loss of y_pred = y_true plus constant_to_optimal_zero should sums up to zero. )rrurrrr6re绽|=re)rlrTrNr rrrcrr+=r)atolr)r:r7r|r&rBr rCr2rIr3rJclipinverser?r<r@rAfullexpr>rirlr,rr) r,rrPrZr2r3rQ loss_value constant_terms r.test_loss_of_perfect_predictionrs   "<= dii . .C&&**C''55i'',,D''66zWW^$?N"">24>>*11%8>>4>>2r {  68VVBZ1t~~112 Av||AFLLOL %#J 112M JUGr0c^^^ ^ ^ Sn[TUSSUS9um m TS:Xa0[R"ST RST RSS9mTR T T TS 9upEURT R:XdeURT R:XdeTR (dOUUU 4S jn[ UT S S 9n[XGS SS9 UUU 4Sjn[ UT S S 9n TR(ag[XYS SS9 g[TR5Htm U UU UU 4Sjn[ UT SS2T 4SS 9n[USS2T 4US SS9 U UU UU 4Sjn[ UT SS2T 4S S 9n TR(aMa[USS2T 4U S SS9 Mv g)zTest gradients and hessians with numerical derivatives. Gradient should equal the numerical derivatives of the loss function. Hessians should equal the numerical derivatives of gradients. rrrrrr rrcrc&>TRTUTS9$Nrr,rYr,rrQs r. loss_func6test_gradients_hessians_numerically..loss_func4s"99 + r0gư>)rZh㈵>rrrc&>TRTUTS9$rrrs r. grad_func6test_gradients_hessians_numerically..grad_func>s"== +! r0cV>TR5nXSS2T4'TRTUTS9$r)copyr,rYrawkr,rPrrQs r.rrSs;$))+AqD yy!#&"/!r0Ngh㈵>cd>TR5nXSS2T4'TRTUTS9SS2T4$r)r rr s r.rr _sI$))+AqD }}!#&"/%Q$ r0) rRr7rirlrr:r`rapprox_hessianrr<) r,rrrKrr[r g_numericr h_numericrrPrQs `` @@@r.#test_gradients_hessians_numericallyrsI9   FN Av||AFLLOL  %# ! DA 77n** ** * 77n** ** *    )NM 4e< )NM     At% @t~~&A  -Yq!t8LRVWI AadGYT F  -Yq!t8LRVWI""!Q$EJ9'r0zloss, x0, y_true) ) squared_errorgr)rg@]@g?)rryry) binomial_lossrrr)rir)rrw) poisson_lossrVrz)rryr)rg6g$@c ^^[T"SS9m[R"T/[RS9m[R"U/[RS9nS[RS[R4UU4SjjnS[RS[R4UU4SjjnS[RS[R4UU4Sjjn[ UUUUS S S 9nTR 5mUR 5n[TRRU5T5 [U"U5S S S9 [TRTUS9S SS9 g)aOTest that gradients are zero at the minimum of the loss. We check this on a single value/sample using Halley's method with the first and second order derivatives computed by the Loss instance. Note that methods of Loss instances operate on arrays while the newton root finder expects a scalar or a one-element array for this purpose. NrrrYreturncD>TRTUS9TRTS9-$)zCompute loss plus constant term. The constant term is such that the minimum function value is zero, which is required by the Newton method. rgrQ)r,rrYr,rQs r.rXtest_derivatives..funcs6 yy!  ) ) ) 89 9r0c$>TRTUS9$)Nrgrrs r.fprime test_derivatives..fprimes}}F1}==r0c*>TRTUS9S$)Nrgr )rrs r.fprime2!test_derivatives..fprime2s$$F1$EaHHr0rsgHj>)x0r r#maxitertolrrrrggƠ>) r r7r|rndarrayr ravelrrBrr)r,r%rQrXr r#optimums` ` r.test_derivativesr,ps4 4=t ,D XXvhbjj 1F 2$bjj )B9 9rzz99>"**>>>I2::I"**II   G\\^FmmoGDII%%g.7DM151DMMMH!RVWr0c ^^^^SmTR(d0TRR[R"SSTS95mOH[R "T5R [R5TR-mSTSSS2'TS:Xa[R"S S TS9mTRTTS 9nUUUU4S jnTR(d[US SS0S9nTRT[R"TU5TS9nUR[5:XdeURTR:Xde[!U5 U[#UR$S S9:H UR'5[#SSS9:H g[)U[R*"TR5SSS0S[-[R."STR45SS5S9nTRT[R0"UTS45TS9nURTR:Xde[!U5 [3X$R$SSS9 [3UR'SS9SSS9 g)zrTest that fit_intercept_only returns the argmin of the loss. Also test that the gradient is zero at the minimum. 2rrcrNrrrrr!rc>TR(d[R"TUS9nO5[R"[R"UTTR 4S95nT"TUTS9$)N)rlrTrlr)r:r7rascontiguousarray broadcast_tor<)rYrPr,rKrrQs r.fun%test_loss_intercept_only..funs[!!WWI1EN11)T^^)DEN)'  r0gHz>r&rs)r'optionsrrg-q=)rgvIh%<=SLSQPr )r'r6method constraintsrrrr()r:rBrr7rir?r@rr<rrrrWrltuplerrrrYrrzerosrrrhr)r,rar4optgradrKrQs`` @@r.test_loss_intercept_onlyr@sI   ""2;;r1)#DE9%,,RZZ84>>Iss  C :  v]KA      cti5EF}}<<2'  ww%'!!!ww&,,&&&! VCEEt $$  fQE**  HHdnn &$(!T^^1D)Eq!L  }}771y!n5'  ww&,,&&&!55t%8a(!%8r0zloss, func, random_distrc,[R"USS9$)N)q)r7 percentile)rYs r.rEsr}}Q"/Er0poisson exponentialbinomialcb[RRU5nUS:XaURSSSS9nO[ XB5"SS9nUR US9n[ U5 U[URRU"U555:XdeURRU5[U"U55:Xde[U[5(a%[URRU5U5 URR(a?URURR 5 UR US9n[ U5 URR"(a@URURR$5 UR US9n[ U5 gg)zTest that fit_intercept_only returns the correct functional. We test the functional for specific, meaningful distributions, e.g. squared error estimates the expectation of a probability distribution. rHr rfrsr5rN)r7r8r9rHgetattrrrrrBrr&r rrFrIfillr2rJr3)r,rX random_distrrOy_trainbaseline_predictions r. test_specific_fit_intercept_onlyrOse$ ))   2 3Cj ,,q#C,0#+5111A)* &W )F"G GG G 99  0 1VDM5J JJ J$ %% ))*=>@ST )) T))--."55W5E-. ** T))../"55W5E-.+r0c [RRS5nSn[US9nUR SUS-SS9R [R 5nURUS9nURU4:Xde[R"XRS9n[U5HnX6:HR5XV'M [U[R"U5[R"[R"U55- 5 [US S S 24URRUS S S 2455 [R"S S 9[R "S S 94HXnUR [R 5nURUS9nURUR:Xde[#U5 MZ g ) zATest that fit_intercept_only returns the mean functional for CCE.rrrr rsr5rrNrer1)r7r8r9rrandintr@rrrlr<rrmeanrlogrBrr)rOr<r,rMrNrrs r.(test_multinomial_loss_fit_intercept_onlyrTse ))   "CI  3Dkk!Y]k5<.alt_lossus; EAIwwrvva"&&!h"77899r0cPSU-S- nU*S[R"X!-5-- $r`)r7rras r. alt_gradient>test_binomial_vs_alternative_formulation..alt_gradientys, EAIrQ --..r0N)rr@rrr)rQro global_dtyperdrgbin_lossdatums r.(test_binomial_vs_alternative_formulationrldsq":/  !H ]]< (F ]]< (F  E U vh&67 77 7H%%u-|U/CDr0c Sn[UUSSUS9up4[US5(aQURU5nURX R4:Xde[ R "USS9[SSS 9:Xde[US 5(aS S [ R"U54[ R"U5S 4[ R"U5[ R"U544HvupeURUUS UUS 9upeURX R4:Xde[ R "USS9[SSS 9:Xde[UURUUS S S95 Mx g g )z2{{y..9999vve!$qe(<<<<t%&&  2==0 1 ]]> *D 1 ]]> *BMM.,I J  KD ---"! .KD;;9nn"== ==66%a(F1%,@@ @@  !#1"&!%   'r0rorder)CFcSnUS:Xa[R"U5nU"US9nURUUUS9upVUR(a&URU4:XdeURS:XdeOrUR (a;URX@R 4:XdeURX@R 4:XdeO&URU4:XdeURU4:XdeURU:XdeURU:XdeUS:Xa;URR(deURR(degURR(deURR(deg)zTest that init_gradient_and_hessian works as expected. passing sample_weight to a loss correctly influences the constant_hessian attribute, and consequently the shape of the hessian array. rrr)rKrro)r rpN) r7rinit_gradient_and_hessianconstant_hessianrlr:r<rflags c_contiguous f_contiguous)r,rrrorKrrs r.test_init_gradient_and_hessiansrxsUI * m ,D667H  ~~)---}}$$$  ~~)^^!<<<<}}NN ;;;;}} ,,,}} ,,, >>U "" " ==E !! ! |~~****}}))))~~****}}))))r0zparams, err_msgz+Valid options for 'dtype' are .* Got dtype=z instead.cU"5n[R"[[4US9 UR"SSS0UD6up4SSS5 g!,(df  g=f)zDTest that init_gradient_and_hessian raises errors for invalid input.matchrKrN)pytestraises ValueError TypeErrorrs)r,paramserr_msgrrs r.%test_init_gradient_and_hessian_raisesrsE 6D  I.g > ::QQQ&Q ? > >s A Azloss, params, err_type, err_msgrz4quantile must be an instance of float, not NoneType.zquantile == 0, must be > 0.zquantile == 1.1, must be < 1.cp[R"X#S9 U"S0UD6 SSS5 g!,(df  g=f)z/Test that loss raises errors for invalid input.rzNr|)r}r~)r,rerr_typers r.#test_loss_init_parameter_validationrs&B x / v 0 / /s ' 5cSn[UUSSSS9up#[R"U5n[R"U5nU"X#S9[ U"X#S95:Xdeg)z Test that losses can be pickled.rrrrrrgN)rRpickledumpsloadsr)r,rKrQrP pickled_lossunpickled_losss r.test_loss_picklersjI9   F<<%L\\,/N v =fDB  r0r)rrr rvr!r6c[US9n[US9nSn[XSS9upEURR U5nUR XES9UR U5-nUR XFS9UR U5-n[Xx5 URXES9upURXFS9up[XU -5 [XU -US-U --5 g)zCTest for identical losses when only the link function is different.rrerrVrgr!N) rrrRrBrr,rrr) rhalf_tweedie_loghalf_tweedie_identityrKrQrProloss_log loss_identity gradient_log hessian_loggradient_identityhessian_identitys r.%test_tweedie_log_identity_consistencyr$s'Q/3!<I9 F " " * *> :F $$%11&9:H*../66v>?MH,!1 A A!B!L+@*P*P+Q+'L+<"<=//&!)>N2NNr0)rrr)Srnumpyr7r} numpy.testingrrrscipy.optimizerrrr scipy.specialr sklearn._loss.linkr r sklearn._loss.lossr rrrrrrrrrrr sklearn.utilsrsklearn.utils._testingrrlistvalues ALL_LOSSESLOSS_INSTANCESr/rRr`mark parametrizerpinfY_COMMON_PARAMS Y_TRUE_PARAMS Y_PRED_PARAMSrrrSlog1prfloat32rrrrrrrrr,r@rRmedianrOrTrXr]r|rlrnrxint64rrrrrrrs0r.rs  = $@    ,K '.." # %/0ZT$&Z0 t$!!!#!$!$!$#&    CE""J U5GH <I 1SzRVVGRVV+<=3#srvvgr4-HI1SzRVVGRq"&&+IJ1SzRVVGRq"&&+IJ2&c bffWbff4EF1%'>"&&"&&@QR3'#srvvgr45PQ1%SzRVVGRq"&&3QR1%SzRVVGRq"&&3QRBFF7B266*BCB"&&"c266 :;'2R [1#r241ay"-3!b)2&<1%ay"53'!b1!Q$M2.  QC 2M21r=13aS)2&M:1%}b93'aS1aV$S#JA/  'm#7 7'm#7 7Js  S#q!Q/s #sCd3s c "Cad;s d #S#~xN s d #S#xE s Cq )3_dDQ s Cq )3ZtLs  CABFF1I ,=uaHs #rvvay"&&)e*;YNs q !3q 3DdDQs !q )3QRVVAY5FdSs !q )3RVVAY5FdSs $! ,   )    s,    FF1I HHQK$* *    +sr  S%Aq1sst  S%r15usv  S$1a0wsx  S$R3ysz  S%r15{s|  S%Y8MN}s~  S%F+@*AA*EFs@  S%y:OPAsB  S%F+@*AA*EFCsD  S$I7LMEsF  S$E*?)@1)DEGsH  S$a1FGIsJ  S%3JKKsL  S$1a0MsN  S$a3OsR     $ #  Qs`  S%Aq1asb  S#q!Q/csd  S#sAq1esh ! ,   o & ,    gsx ! ,   o & ,    wsH ! ,   o & ,    GsX ! ,   n % .    Wsh mwpBqwpB6,*UDM:bjj"**%=>rzz2::&>?4)4$+$+q!f-Y7.,,5@?;-Y7x5GH4/:K$;IK$\5GH4/:L>;IL>^5GH68*<=BF>IBFJ5GH I <5GH4/:-H;I-H`5GH4/:VK;IVKr "+X#$+X\5GH4/:>9;I>9B  RWWh/ "))X. d #%ExP  BGGY/ "''=1  BGG]3  RWWj1 / />/."<"BHH[$9288K;P#QRBHH^$S#TUEVSED5GH(I(V,4/:2::rzz":;*-!*.<;-!*H,bhh 9"((9 M R-R%     B    O  )  z3'5TU     B    O  )  Z%z3RS5>?> 5GHI"56)7)E(1s:u