From 224015ec7b1a9e022e3c5f6218d6ea9ad52f7066 Mon Sep 17 00:00:00 2001
From: Simon Johansson <52204459+SeemonJ@users.noreply.github.com>
Date: Thu, 24 Oct 2019 10:11:17 +0200
Subject: [PATCH] added the latentgan model (#59)

* Added LatentGAN

* attempt to remove submodule in LatentGAN

* attempt to remove submodule from LatentGAN

* removed ddc_pub submodule

* Added refactored LatentGAN

* attempt to remove submodule

* removed data directory from git during development

* untracking pointer files from git-lfs

* re-added data files from origin

* merged with changes from main MOSES repo

* added the latentgan model

* refactored latentgan code to comply with travis ci

* removed temporary files and too heavy pretrained models
---
 .gitignore                  |   1 +
 images/LatentGAN.png        | Bin 0 -> 34333 bytes
 moses/latentgan/README.md   |  26 ++++
 moses/latentgan/__init__.py |   5 +
 moses/latentgan/config.py   |  92 ++++++++++++
 moses/latentgan/model.py    | 226 ++++++++++++++++++++++++++++++
 moses/latentgan/trainer.py  | 271 ++++++++++++++++++++++++++++++++++++
 moses/models_storage.py     |   3 +
 scripts/table_config.csv    |   4 +-
 9 files changed, 627 insertions(+), 1 deletion(-)
 create mode 100644 images/LatentGAN.png
 create mode 100644 moses/latentgan/README.md
 create mode 100644 moses/latentgan/__init__.py
 create mode 100644 moses/latentgan/config.py
 create mode 100644 moses/latentgan/model.py
 create mode 100644 moses/latentgan/trainer.py

diff --git a/.gitignore b/.gitignore
index 99287f0..74317c2 100644
--- a/.gitignore
+++ b/.gitignore
@@ -4,3 +4,4 @@
 build/
 dist/
 moses.egg-info/
+
diff --git a/images/LatentGAN.png b/images/LatentGAN.png
new file mode 100644
index 0000000000000000000000000000000000000000..0d541da90bf067f681c6cbb1263275773c1c1df3
GIT binary patch
literal 34333
zcmZ^L1yq%5*Da}pbm<1^MghT1cPrgp0-|(xDBU3;h=72EfV2XFbcYCnpeP-RDBbn1
z?K$87-Er@@XN+?^gnPg5`>dF2&bfB9hMFP)9s&;y4UOQAlAIPA8ipAf8oC<}27ITi
zTNWMuKzG+tltHWbK=Tv6z_OKAl}1CWPQ*X8#D=eNU6c&n(a;DxQU9X%I+s|Zp@rAn
zk(1W(G5ei;*<f6WE^=$YTYh7A{rCENeI0UteH|>U>c=GpTWaK@c=r_LZ%j#PR6oLF
zk9yA#LmzXSgg9xYY<J-xG}>R;7#-dF>BY%X0AI?N^%<wWEOWkor2tl(5PF;tA}uBc
zy^PGv;b+KSLz(<0ne;gMr%4Fh>W<0DdavJKr|X?p2e71xmm|lOqA1$i+c7U)@!4CW
z(1*XyF8NYHrn~`u>&IvT6B84Ud5eRiV|Gr?hnvLMCMG7_CJh!24i5M3ZEwzZ4h^X|
zwXcro>FFsbD467((ntn<`}{ebD}cb;`yli((SqpG0G8$sJ(BcZcfxhi7d`H)AFfLT
zG#J+2FILN5pp3#GV^w+ZrKYf^X8Yxx6u<rT=!@jZO7mLzot5E~)Ku#oiWk8b=Un;~
zOL*p(eonJ3@1H)U(&y%9Q_l&TY{Za`;)AOV=o=o^)zJ9yF)y^WRWgQ2A<m%oK7lba
zX&jBP+hF?5jQ+tvd;$XbYUQWgw{G2X_$e6T>E^B*QChkdf{owspI3=WNO1a86rY#J
zkmWpCk^LY%L$g5Pp}oC(_(-l~z@bKymm?-7<`C0mIx!z&e%|Nt@dQr_pFjVW%%q5i
z2qMJ6kz&`;(&}iHkdQDmG!zjL5p?=yP-ER+V>385HnzY1#mdn!KE2ZH{nwVjllT4y
zRZ3=N=lv;cuU@@s^4`c1c4wWz>7{(IxVHAr^VetfT#5PcL$P}2*`C2s1AYCc-U$b1
zCkMYKEA@4BboBLmSLFyK%%<NuZe@?aF^?50m)yB~_iLj&TTC;8n4J9Ot5*-5ofEyk
zy>+B6W5%%zg`*}?Q&UrYcH1&}baZrU(mZ%?ch@>r4x=QwP=Qu9k}zA)nf5;AO{3d>
zyAOuGHM*~Ev|dE$%<+qhi;qhQe1GSel9nd!yUlFk`Qsyl!E4)*oR4c=(PRu&OG``W
zn3o*BHIT$Re}A{Pm?*iwF-w;dT5VMCB<?W9i$o$1Ys}vJt-GyEl<6loTUl9YYiSW1
zQ{R421%U?7Lcnf3>NPVjugkXv*QAA?pNf0qt}$_NWM*Xe9Q-jIQo3_z{$tmdFV`xw
zc`Y$!aJp5q1m@oyw)wnnYC5?%KWJSU&Kk+M6&e^6LM)>cA9jewWic679_kwyc=j0c
za*Uqf^98%-w;ztdUEQkn+gpQou)4HwH&IH_rClYKJDepb8+Ps5wel!=V%^rDvln;t
zFe8ZQ2&wK}awRYd>ETUuIXc;BwR&^oMrlcj)cuUd5_xqE4MOFm&a*9>jGX8*B@o#-
z2?kvWl4yhy=ZA|#VM;@l2Gtl0X%kGW9mD0c>r>SQ^0Aar+d)A=aJ(!Wq%wrBl8Nc6
zp6cuA-QS6lb8~ac%gdwqp|<^=y`qPNNnvwb$nop#p@7pvlH-k;_b1g>y`-e18ln0(
zO&WMx_ox;Y7cUdhIy`*XpLR6vv)DsGb8EB0{CscH+<$-Fa_8c7he$8{S`LClkOa%o
z!NCB6Hx!bvX7Umi7JXw!i&bwzI7TMN(;J2%9>yQ{Ya=2qVX$mqtj1FDh<L6o{r(-F
z?!qs6S(#V{A1;)Pn7CS2(q#^fD@P*mo%hB}mVo1e=QepzObN+-yQ!-A&##Sdwc1T^
z7Ac?!-%{@H>!YVnpp!IHRgJ%Sj~8jy5|GKY^J)tY>M*^dql1^o8A2g(0<#(0IwF(N
zuCA)ef`GVdt0?M;Qjk)`-Me>tdU|;6LoX2o1O|?x*_kx^+_EpPs7SanpuJe8U-{1C
z=k}lPqD2a3&zN!M!yveE8yg!nFfQIdA|WA}YIVofv{>p(7WG=U*nw>1(kp-DnIW%N
zVbZV7gjw_ObNRLNiBF$?O_Wh>1k<d=#K(ste8tVBY+jM{^W41od#OJq`22{?UvY|X
zG$l|;S64UZrRxKX%E-vbTE_F%uXPVYGh+8O3KA2E%GmzCKXC_!meuNqY>0I#ZWDGE
zmI)|mwYDQQ@>PU8>4k;M5d_q6adG4n6qgAQVFizg=p;_h4rcl94?U`08On5rl#=WJ
z^Svo6GqZX!L?ZCGiuN{C!i|r4vPw!gD|tOeZ|oQhsBZUf{0ybI2M;I0vl9}&W3|y`
z9>c!MeN`>s<@mS`8(Z=5-mlKj5E)&osYyu|rJ<tpZr878M-tI3gtk}m{(^F~`Pg!6
zu{W^*!rnRP<hLM+$e`-Dic&ll!wniyuPYI<BF`ZLzP_<P8WBNIQ*TVy6Y(d~ia(PK
zmT1P|-#VUs_wF6!0XqkWbeRtY=_(v*uDE|sdWhrnThsS`6#Pn=P^Tj6d5MtR<}J^?
zIkrxZcFM_q!Z$}R{Uk#tkqJCM6}Gp4`m|STrsn14ReOKfsLBEztNiurv#mu%9hv^a
zDSq1z*1zPe$H&I*4S$pfrof)OJ25d~=)i)L7e1~;4E-w{8Vmd(#{icuO^l_9qs0V&
zNRx-4YF*ZEjEUjE@tXmS{oiamhaYhUmk3A`-@A;8Wj^=qqTH<j>f?SM)_B#N`P`!H
zEv1oxU_1o8+7||BR~X|Bn@`l$EQ#`2<KfNZg#Y<wiQgz2az~QnzV&Lp!j<tK8yn7c
zc7B_4?U4SupFb&@w*>eubm2m(+1uOi-}vmz%*50=Gc)t?<0iDzL^{d)i8t7ANZ|UT
za2HAy`ijh3gFY3j>q9+n&9KE3)XV7V>Z0J%yK<G+0y+kyBCV)bd1~tTYvcNcZi<fK
zVaI_~j?M@IZC%}L$>8%B(xKKiHb1W|g#D1QWA4JL)>(w}g_E;=_%Jg&JN}-YUV3=A
z{K22^Tw-$kHP^?nso?BtY=b*fp+UX(+bi94pMtAB+5L&hv7M8#_Ul(!NeR@Qx}_y+
zJz3&>Y}DiGQjdoM6zb;mt)V@w)xzFh)MwKICz|VNBy>~;o=e!+uFE_Q-{0xpxf6}s
z(EQ-R1NujlzK-|rv%0-(X^{X}k&HSL+#|c@C2hIT5)xA+v^xbevmCBXbX?VJ5enr-
zw@hgjlN#$g(VaN{etwL1w`m8AyzW_9^(N8*0tl=aij0o#hqTnz)^6mJjv|D<uhKVs
zQXbG@Rea=+sT?fuyYWX;+ot1PKf+0w@u_;}+@c~@efGF#&k)qT)h2DPq9PzY7Y72u
zA|oM~k7wTdn=}p`9v`b^@aA=GQPD{R2n8H&&dpg=+(>OEyklswayI+EYC_C3sP)J8
z$zfegm}mZ7R2azUsm+A!Yy@;lg!jCx{V6L~TAC?%!J#+WF*VgtH!c)->|xDUVRC;;
zgPNsCfryCc;X`P}&hQ;5*(8?A%JBxfKR>a-+dL1vwl(4zJOghdE??aL%*Rh&ql!`l
zI#@=VTFG{<cz3x{uEyT>_S!_*BxGyjHr4v?-zWROsS${Qw874Be5$Gsq?JgSdXv28
zQuQ%0F(&i)KY#vAVN;)J^5)T})cepVhY)+!CunF`-mtcM$BtYA_2TsD;yMHb1pd1}
zT=&*=y}G%Dg^wU;o;-PSUDPXVh?I=Xu$FIozBBS@XSLqxJHqlqcQ*?g+l7w1fx*=I
z*@>vf4_=Elw3(7_o7K?*#RA1dr_ucGXGnV=Z+odV$c1o()xCQL@gLo0ShOxQ&)YuC
z#@eK}q$0wJ0h+Q#6(Z0*SZ?0D$;4FaGB3lfdiU<Y(9qEEFad*eM;I<Zq=m161kI&@
zfu22k*3!}fW&A3ibyP@5NK6b~mw?Rx`Q^)(yFN*WUb<=gwmeQkSy>s75cQDby?YBk
ze@5%hyMM?MB+}}NG^w(9tQ4njatpa)6SyBp=RLO89ic5aNmKn<pt?0+`X*l;V+0ct
zj|~-0#QMmPsi`TztpG=UK&)lhre$Vk8ozb;IsWRC{)VfHy86XxZg3L*@bGYBX$aeT
zxzU?a|ARkJ%ZK{<EYhT<rJ-oii1}0%6fl)>#J9DzMU%0~6vBPjAfY{d`0$}5tXo`6
z%%E^n8y)G_8f$ppVl-&FX!Z)F((}|bp_@#W%m<U`do#w~>$l(2A+%bIh3~|7;QmcQ
zgLt&+E3}+J@x+;=jgEV36NTdI!`<E3c`m1iTh24h^QA2ofH3F!C|WiF-^5c3%4DKo
zx$dnBlSV!`E>F+BA3q*ctl{@1L-LzQle1SMDyRKDXI|WNy=%76V;!BP&ElkX!l#;C
z^^DO~>EZZ&+4}lqlV5yNcI0Y*gYw-x`REHJ@6P$l5x-_zgQLZj0s{kWVa|ZeQB0(D
z)isxTvHqoY6%g#*08WA{k#T($LX`8+*R-^>1_lP?t>gB}4Qq?^E6ob`62tWboo9w9
z<Q}`d8r}bMqL)?2%&neQmBnc(Uhu=3-*n-#4({SLS%F{v{%L!RT(2~RQ0W(P_<gc3
z5e*p<7k5SI72sAF2CR_)?4cO_{`%(E8~aHERo52^hK65ACvaqBWo?HtM(dr;b6k@}
zd*P|rjkBX4_(WABFfqye9i|vuT}#7$*X3(1<ys7r)$n^?Ee_`V6rpbtpPrtsU96Ut
zoZMa+YDG0ff3WebF~wL#2mf!dw4NZ29wRPaxw;8Bq{4)Be}5lfeqR&>Ll?I$Annah
zj?n$C9HkX0Bt3oVeYiOf@p@g@4U6`=fWs7|8#d9WYRk?`TRK)VtHW&06Cy>ax%jaS
zRfpDsg&!*9pZ=b4tE5nJMKadF5N0AP3u#76KrLYM#%|&Xi!$fwxJmv!nCfml&em9#
z^pE-S%ta<G_r;4BYS|)pl$1(JN?ddcF_Cym@nRL`t>-%<l2+MJO;=0rDw<4OB{_ao
z`G+w2R|<<mOmx+uhx@y(##aN!MVUX%XZNw5YkYrWPj%5fmzqj`N=;2Yn3JEBk<s@{
zGA%7FT$Xab*(WJ0EA}JLsrLsQ{=bi7!%4_S)5^!^FouGQRXwM?un-Aj1VGNZy4y~M
zcT51VmDzQjsO3tS-nnzf)HD!+wmVl+1ItxB&u_0nG4%c5&dFEuYK!QUVUOrv+D(6^
z)|Sb--?w;RmAFE3N9{SxeEDLirA68boyPMiw`mL_%!i+!|Lu5BPZ1#N@`G)dYl~Gf
zQK3%5JYP@TOh8;H{kJRX7tE2Be*XL!Qd>_~H$Ey#Y6}K0IIgMc^g|ErOXKWwVVrAw
zdwb9aYyA(L!br!T3%Q1Pu90kZxwz^s=IjvJUt_=fEtSKreXNL@uk{yq%gK16f8ftl
z&3HI!IFYr{g2?c2?Ay`i#Zdp6&AI>uN|h@n$Z4&AZ#pQ|EyKXXtR|>?_-O@zKskXX
z_uo3aZ$~5K@&Zt#{tA<*sQ+wBAT;`j&W2Kqg@uLj@o^YcbOQ{ydX@*DD*wtuT)QR=
ziC$saL>V3J<m3dCZpQ70S`UwOO(q)7cCSaLU<*DZ*R*(;@?9tTyZic|G(VG$f;hW?
z^HX0y(mvkF3l`Xs958y{lc!mhipPmsSy_2)^d>~Kv!f$FKcAeO+}hgOfI8nK2@!pI
z9zR79^5e&ksVM^ib9pMxO{x~R<$+RzMxU~+-_!N*f}9dg66E3EXZw<wuG7#oa7t`%
zZzs?QV~EP=chl_zo*nNk93U}DJf^BFH2`nYHF8Q+h0jr~1Fnr`@_AryZx#6gvY)S1
zb7NgR+RANNDgLeYMZNakK9l*My+UmTB-E*MK*`OLG(`$CZPr2qj}dU>Ft@Iazv6$1
z#=-FlCK^WR8UeXRz=B8O3rWJ8k(1?y{4|+w9lq2+Sae1bt(s0$%EiUUZx#RgWxU4S
z`ThI%{{CpL)XSd<ilZ)d7f~m?e;)`<)i>|4D-;kYRy@XUuN*xp<HUX7xzH62m%f_l
zcj!uR<;u&#!XqMWVku7#4?xJ)4<4`|PS)4dticiI%f}v?6P^G+;?n<wWlGUqS4REX
zgwMdh&!_M42TK*Qev<0(4ZnTud;TuWDwZ+TH2$!2N!*{vZvFFX>J78&(8EeddK-?9
zkH!6VH+OboqoQz#h|~e%I}dy%UIqZQG>ds~2<SMG))6QZ(<U#7Xd+rMpQCM-`Ub6A
zf>2DG8-W4Q=kDo=r}k|Mbz`VFK>GRkl$4b%D%s4nJw{O@Fi0iPLg;&~PoeYVK$P3F
zC544yJ-8ygIFQCAEG$g?8!9w0L9$$)%6dpLHYNrJ%qO9Bn)lFdCII_l2F)1w2>tb7
zfU^?5+t$n!<<bfYmmaeU>SGb!9Vk(#^d7ywD{yy8h1;&3w3zx63%6=!RWC1tL^dRh
zj!OKY)7H*#wkTEWeRzIBQMtkA!YnM<k<!6udsHg@&yEfc1K#NKYcjr6SdawB`S=V`
zsVKyE7sjCS=+@gbg)Kf?AqTwn0$5nTNG^W_;4m~kPTr)iryT}?m6ee(H*(Sujt_C&
zp?Q-tS^3Ym#!#B*sexzW{`(Gqo*^t{S$G(}w**pmL-WS{a4X-(+)C=OI59aygjgcr
zka&K5Hu%ER)m56XO}*;=Fye<TppwIUjis+?@z7-AX~q2@{-Bt^FK+^OQ>Fjr&6~8R
zPZa}bkapuh#TfGB@P(+_0POae&$wk(<sIQ28WIA(?Q*>P@XnonsN$(U47Fp?>L0f8
zZp@}A`$8RQ%*?gj;7{XCI~zDVcc{raPI<h|{fV@v>P7+@((d8I`M%I$N#E_IM9K3u
zIs{h;Ou+(9->#ZVybZgP!iof*g^%w7-P6y{58!({zKxB|7MFp6LDl}>+_-pbN`0Q5
zu~{`Kt0m1Xov=P#&+~-E>($Qci0#eftIW*Io2xOtCHSsb!7(50CM!(y1V2YyE{}d0
zrF07p0dBbtfZ*j;A1WG>#hbwRRC#<VQWYZZ;ihn&gxG)hsmPKxc<pd!<&E?Ua$>nD
zT;229;tYyyveNtYl4@gqL0<)Uxj9<#ACN$$nX6i0BHEa3ZH2s}pcbc#WSxqnyf#di
zNi@qFWL{ySMr`N5x7NKsDCUwBreq|3vOHH(XE<$Um({$3qqI;upmJ>jhsF#GIhCJG
zM?-=a+mICePFQteY`n$TD<bu`@FK?ekDovLPwYEp2LU+8ev5vL$x7oYA}Y$p%sj`=
zFP@G6s&A{LclNq`51y<Y^LeOwJYYRuU?zt*ZTA753*n~U6VyS+A}ks0@5c=;i$XTD
z-G$+34lVsci|lz!TAN*8!2JBYfh|-*-ca!qAes!6sj0JARi9M_iW}V%G~FCK-NGJ?
z_%3W=V>40rP#ayM#6|G=`t#=19M{hy9gew*Iy-d}Q~ocG3)%;1)u(qC;v6jk-13w?
zqOg^$n$NB$!ZZj7a^{_9EDT$49{`P?ac$1e%{@9OZR6EC^ufud;xStq2msKf2-DYD
zfEZtmTW(AI>u=7}<L8TEZxWs;(<Pejs7!ZF^H<S?+-O3$mb_W)I?MIuPHHfGcYVJ5
zmo~z|Byi$wxm6;L2GiDcz*rZW%*X^7_=8jT7wiH~_HX{uZt?u37uF2$mYtK+Yi*o^
zn>zuC2^9dEk-$bo^WamS&eZiVP8SakF^JLho5ns=^tTdCZyFC+I991EW-XZB_KnI%
z)IaanU;Id+zVyZ%@db-qv!JErT=L?ylLD`xwAAQv!V4%sE@`h`ap<pLY5{$_1X<SL
zmf2o;N&RVtT9T?_+*J~U(Ss{zr5cNdCy8W}!h|7+i@>*EKc2k~o+#PcS*=qS&Px1p
zQKPEb;db|&Y>{3j3qx_i+&J^}*$$toV7aj45nts{F{jdBTz#0hU?&FCg)!Fsp^WN)
zBR2t}dg*vtTG~GD){Aq11D_b=BPuGmH|$uXb{~%BV{ZsNf)en?;S1?}qI@ocY{VQ(
zhBQl)TI0h{tY&s{3(tX|v=6;jhkNCUb!-e%44V$mH-em+`R*^qbW96(e^gpNcu-PQ
z)I)*C59KH-g)8-%+yyp1W$IzGU_do=E1r#RWwf37RW4UK8+-d-u8Te0y}j<soJ==Z
z16R_Y8nKDFMx~?}w>V8KYcchyjfh&#kNam;bV=AmyVsCT7AYC1KdDMh=1o7jqi$24
zGe3M1lH+bnqgCJn48eIp1ESjdek|H)<aMqrZ!fP1#P@ICW}1C5t#BJ&&abT<157Mn
z;p%u3k3h!6#fkatvYYk2qp)ZlQjSM)Iym@WSIl)}sVPOPbX@GnyPbqKLbWkEVnKoE
zY;VtlQN5eac5!(bJ03U4!{;HOD2!ICt&}<#y=C7&Qf|b6{^J@=;x(EtUsyOZz#<|1
zD@|QSO-=p}lCu(V?J;_^ZDkWX;wsVj_X%TT@4gT3V-rp5S{AQUYyT;(Nk)W|X)$fl
zMR}C>60GDYJbd^Q@PD<ul%nEgH^1$rerVMYA?Qen;Bzq^9%qmQta{>H-P}Unf5;IZ
zY$C`^2ExA5ycJcg^qn3)`~oaD<f~NtgNSS8c%S7i*?oM0{%Hmv0Cc&i!w_p<fbB)9
zSqs-0ELd0yfy2YoC^-DyjIP-(0xZm8D^?U{I8@&=U|kEv7ZrG2t_BHRjlCeH59$R%
zRJL>d$4is0?(X@W)9pc?DIWlKo70xYoLgJsGV<NyeSK<!%Bp?m<WEzcWuzF(NS5+R
z{~fiuf(;A}<FAXRn{tF9NSJ)o^D(xzw))&RKZoIx5z>l91@wPtJvutdYk^>tjkwaD
z4~J8`FJ{&nL=Y&G^x!R!^FZ-gnG@N-3_?@c!3Y%Qy4jbYBX~kd=xbY)KD|IQZGKhB
z5j*0VM0Ed24-UYD0qyV}pkN^J-vGl41?@w&s5cxPV2$rEL4DasXv^z?-U^g5jbc0^
zobCV;L_k2mezM~CXhAQajxG1ywF!LWH~fJ%C<)c2!tN`8aM4;fpex?Iaf92eISO%V
zXC6q|$ZN4e<J(pJEjy7jIZZDpXa1!^9*7jDiudo|2b90DFFjRum|4HMx!KhvH)O$a
z5h}pQ$S5Ntb6Cl@9g1Z7Q&s4LKNBt59OHScuBy8HdwKNAM#x_6y=cc6?Rh658N!9^
z#ztZMVd!v?J_hwO#p<~-$|b9N4@n=re}6XD9y+h%B~4}i>hWXr1WjOcF_F-qlaTzk
zZb^H6tFgg9iN=1EOWVs0utDyMjU?CD#s2SUs0jQpU&sZctIJSvRy4`;2<&r?RTh5*
z6iE`(`(3gt)uk}=umv@jNt;ac`JC)K0;-P{(_|14@q?((7V(t%wptqVI5Zlb>Ko7x
zd^V^7`&lbiY+?gf^wKL#nTLm`qV%eei!^uFD#uKLCq$ZYz0)#YKH95fc9OpgN<G5-
zy`S#aK59UN@_MrOWNB7QThP)t+1Zuu-rdd->`Dv|?*{EJ*86>QboA)Qk8RUwKJ^PA
zwY+@!a%p*4xtX`obx|Qqt%4r!Bd%_~+~09z#?FI8D0K^B>EA*5Jc}Iw?IVkN7(~W$
z&Mxf?vj;Z!fF8H|=ez%WM|gQJpVD9M#TYsh{UW2__73r+=6M<6S_ISczq6@l=llWv
zefjs)go%}eHqW0t!NA}DG1~dz14wb2>gtv75JDoWk%EF$FBH(oprHf-;`W*S_)$Y%
zUcT!SObza_$|2(hX8+V{Vlug{mYy0WA>=k_l!$yYKq@xY)-7MfSI1bS3r_j@`1m|L
zJ%OP5Sg1@@a!_`#F^Pp_T@CcwPQ2B)^rA92@W3fnHnN<1#$_)`<>oGX0iW!@>;`>B
zsMak5yVFucrAZ@3B&alwLm9U)E_GYKs4g!rw?N*vwRQel_X>UhG#O}cJVC;>I18_P
znn!Xs8pGF4-PeDh9c8Kd`Hdg*Yb>xm&80ST6n44z=<Dxg*ns`?<Y?uk!arKYdYD~u
z7`qyqPBH@fxkh|^JUoV*Ux9pLW@UAP>;qg1)XlerCx>=_`|mel%ML8iRDqqu#tzSu
zjl>K3z&Xxt^xb8kqDS2~mncJoW}Rd4qLVI{g*isOGWcPg<I=qyzPR-fqW{n)Q{J7@
zb~k>{^AQpf0{#cudDWYZh6Bo4Q4u9Md6zDQ!seSE=<vJiUw~yzc}iYeQv<bT3hL$e
zChy>5o$a09T}b5pfb-^^p#rkuOUo4ej2Efw4ApPyJHof8wv5NGUcCwhQ}%p1HhSPM
z<-71&L0^9oDhXw94(C*Ba4=L8s4&U_5}nV5-QRirwsUqK(cN-g9-t5wZd|~WQ<Rr~
z1rpoiw{PE?P|(3_?Qmn6RDH_nYu%l?KlV2Y5S0=^!F8P(jEpy2@1`eB2?UI0&0+mX
zGcS-s0$I3FB46}~L%RqXSCp~FRVJna;CXjw6Qp%uoB)c?8QBtah7fIMw7Y_f`?yns
zXYQvHFmsm)X?}nfgkhd+OJUmVQ+>GE%{vPMw=%VD4CqPK4@Mb*wE#{Z*t1`sUr#r<
zGU*dz59IeGy;RBe6^SN&6Jt<S*MHXQ*W_g~aAW$@2W6dt{QP@BTp(}X)`MBZx}TJa
zipojaQbk1t+DQ77Cmm_g11MD%S7WhbP+m4|ZTTR-qNlE|4obB8v)ivMJ0k#L%W5d8
z4$r21{qlu?R_yADssLZKMUJ??z`ZV<=I6K@@Vmfz(202bfRW`j;0k`O-7m%JbX^}8
z(CdVt3xvA@*EXyb3yh@KG7u-fMhh-u87}bZ$|xC*`|Q1%UG52tIaJ^DkkT5xS*Oy&
zsTmLATi>k7<HV{e-t#7%OvIOi+_bdroMIdov9Q-6@nljuS5P&`8gw7=Mc-xDs~a39
z`nh4Rp6?!}GIWURRC{ZK8b4iM3D^z*4Rn(DBj^JFQH!BlM}?X+@wKU?$V37UA;S%p
z0kLmhP$e5)gCIFY0-|Ytopxp6{GUG(Zc9r0KCV|?q0satuQP}q`Yiit9UUx`*$o|^
z8k%q9pH#-j_Ni}Lf2un;vbRB;*4#!?;?4pyQqt9~0g<*P&i@;z-=!r;E1>z&d+(I&
zZf<7q*$g~F$AkgC&UfcNjS)rpBX+vK(IJV1=7XY+o5apvJhDbJ6tMri4n5FB*Y}Z0
z@bS-L5V*D^KU~5k(n@{uWFcP2)C)LsP+&l89bp`C4fn404g$Cg#CuC=PgnDZXzCj>
zyE7fRpb)nD-1D*7z;NSmsk=Ih;5%>`vb47L*!?jIWPL<f7_db8%F2VYv$GQuZ+v%F
z9O&&!T%r6V<<`{jZHw@+2qZ!zR~o(H<mKH2@(N}=lGCZn`M3WT?jSZD60(!MUy*?{
z`sX3JxwBCuOq}<#nG_TH1_!A&()@movB(h!6e;>jU;Pdq0sT45TOLQ-%O=tWN2Q2@
z9f@E4<sA1~D%C7WNk!j`xR((tMce6UYs;=^>*zd3+EWrH`b-8cE-uQ-%DUEpw;*}}
zv<H{d$CvsN%Js>uzm)27+1lDZ;N#@PQ`CeSjLpL`@K<W<s-*q^LITxIqb|VsFJIO<
z&q`Ei+;?+#ucp2!fpr-N2j?pEU1-uE0pSu7_T;?l;)(@kqj_F3JiW5?*7=UD_|e`q
zn?IqQtgS|^8{aiMI={qf_mGw5<m3nywL9N_@bN|K`LVLGF?E{I3uyxf5TiloH>?FJ
z?ei;bTP1Yho}pmNg;>W%x&0*sx3m|bB;CG!yFtd1c-+Ioqd_K9+CT?vELC#@0rat8
z9I2_x!oq?I5DD|wz24!pqXzd?!q3r%9ckfD;Htzob}E7sj?Z+q9nT9@57xYdT(>Ff
z@gAp^^n9tctu^WE8yJugc`o4i75IVtf`a9xCCF?IPzz%gpFn@@On^u$`SQkIAGD~`
z(^G&p9i5#iPoDg!=#BaJ8kxL&HSm8W4yfO|$2cb2-Sg2Lv>3oqjTm&$Cl&hGm2D>(
zUltduXN&AXa+zcjfm36B0b9&b@#<kIV#k7>>uHncG*xo$+)tf)+{EvL9x57?$fB1o
zm6VjAQd`hQB9vmN1svMMTD%-^Grm9}>Bw(tYO1dn<SQ&Jv^gJ?2;Se?I$G<*jwb!9
ze6=9D?U&*+GOn&48c5990%?}Qs(QR8b;`1^v^4%oyBJ6bw`Gi~)oh7Czwix3F-T$X
zQ^bf3(=`1A+Y8=RD?=r?w)U)#OX6dkjoqjEU&&dvb_CK)w^)d#byG<vq8{}HXA1D}
zs4FTK1Ahksxs(DBTPS@>GZ?S|WI?T{W7QpxKi`D8xe1>+d@35As<r~UgN20!AST{t
z=~R>fiZyDvkD8mCyXpgo&vd}4!C69EMG;Q}uAZLDMya`wd0sz0N*f!WfENK|CK#n@
zL_A93;#6d04C&^yu-2bd)AdRm!;ST^!%PUG`ubQQKKX|?#P<gB6&$1#KS<P6ekzK>
zP|k|`dZl%%ynG~vOEaFwtl7ogeG$*x=WsKxx_S$c@cZUw1GXFcFN8jS5fTwO2BvMj
zh|qzSm6a9ZfczTQmA?{?=MusC<u3$|FtoI{;^E=VF33lK%}6U86ks4(t3L!^oKcDD
zIT?oc_4f;d8ADG9`oK>NVwPCR5H%c*ScJ5J+vjH#ir0hfbh<*i*>+mr{kq<w`?f1X
zJ=%`BMHE%#EZB5nndPdl`7+Vdqv6)nT@KfO1V%17JcE7e!ah{6)g&}E`9CD0qmltO
zZ;m3HQ1t7hQ;Mldq{A^YgX`O$k8WOe>Pg?a=8B#4cOZjjZ7xWYk6>>?NI!5(Z;uxB
zzsx0hdbZI<_lsAt=&sV?Y2|tRyKFBNBS*jQjm;xD4F5`QK-yBgFtM{13XbA`>(k=i
z-JYF-^orM=y=Ce?+rv+_K6@ON=1*hw9|20rj+wJf_^DDT+j5HywspWc&C?xWkyZD5
ze^#9vpS@ow&1HL~d5H3xVZ2H%L9;biICcL%%ftHGBqi9vPbxq`Lx#1eElg$Tr1vG&
zL+v6C5!KIKDOv@@f7SEP1d+3$85CJgLb}pUPBW3b$v=6t<^R!wUpvj81Qrc$P?-kN
z<th-Bll)a_>C+F8dK?n%{iXgUv)H3ox3DzH|Hj_0z<6Cbvl-javh(+p5<-^O+jL2=
zwy;q@t0F6vCuPjmxEX)Iz@hzF@~P3e_2>Pk)psba#m3NgT}AzlhT!v>l&q{`tK*9L
zwb1#poPD#yw!VlErhm*0xG#HZFp(|039UzB_p*~CzsL5mIo%R&WZcvc3@H&nwFD`<
z8$8zxg?n2~9*~FkW|PY8{jDlmR=&}3%93__kGY7k+W%H)rTAqQF@fx-2;7i4lp*jh
z9@%K@pwi)KWkaSbxafTCr+#emm>%b^cAS)*#peI*Y@mR5dX>cC-7&MueBX5Hm3NiZ
z)*+1A>a)y9QaN0J{#)&%Y%|A(LhA7XBaCK#JvKJlL@7_Uh|WAdSBzbH#Ejz#0Ff9;
zo;zn<E870=%VTNEZs92Lwd%Lz;o;^-z;Q<*2BwLksWws&xcOM%^U##_{hj5qtSPUs
z#e!1}*Gr5{)tA6cPwWIEA8`P@ael<Y()Hlq4~6vqu|i6ZYveTceidW-_rmfcwhvX~
zzQ^KJND*T}-w!Ft`xZ>$IW7F;-$z1k$L@Ux-U;uYpZSp-&3WGb?~6J7_h(aL_`%x(
z>HqzE0E2}6x?HhC=utr^N3gsdn)GxAUWXkWyeZ=v+~4rMwxU<@u%tKPIN;unn%X~X
zpl%D}@0(?O4fY7kZKJ>mnFDADw|3CZaVt9WKTqfXGP(SJJuu<W^#Q#5Z7pEyvqn9G
zV}^SI8<eG`Wp#CRW@fn4JPsjYdQ#HIv9X1J9h!0kEUc^w($aZBXU8y=E|=n=OrU@&
zfB*UOAq_Ww1$DvC2D@XaU>3bjrT>Fr@$1*G;NP6X!_s7e?jSd817=G*J3EuSN6pR6
zfcTzE1U};;lZyv?7BTWEbO%}>^ne!O<m8Mbygtz1e@9gnAgR5D1^r|Qy&d!4N5IU+
zcCa>K514NxM_d#ra{$4h*tom9$L<a=L?Lkg-i&&V7`)yG;5cpp=X8d05&G^#++>VG
zpj>q<V3dO~$_BhS%L8fPp=4Ze!H?Rm1)c(^o#MhmOEVZfL8s&3;BX+39(RF^>9er4
zorGo$bdW{lOx`0H`&|Z0!$C&XEz`rs!g8@8QG`b*FCRWU2>%QW)cA(b!{2)_H#dj!
zhGJu5OG>ux?(PCN4rVK`J0f{_674nj<mBW|w|eOSm6{l)-ceCWO-)5Hjw_1`3&#f=
z&eqmllq6X2aOFb5;YWzL&CbOY%SCl*Z0z&r9Hf1EX({JM$QN)~sH&<C4-L_egBBPW
z5m9se{5cI785vB;OD};0s=gcL1+y@_+H)b4$qu+|_zL`8Iq++z=;*<Ro9MrefTh~H
z18NQ#C1r#@Fs)!cwbVr*5N2j(4!UsBJy|s)WH1VHh7;J@u7cz2eZUdv&f~|AfqGWN
zs;{bosJTyolg|o_xg1(>@P$G=qMDGY8z_dy6`=ZFiIg^i1NFZ+7hkjF!^Oq@p=xg4
z3IhO)DI*^~L@WUw8V3#n#!2Vtx3A&Beq=_-=1ZP#Vw&NP05H8=h7fX@gBH&a2S(JZ
z2m~uLvjy!dlq?69H@IAQEyuB!{eXg!>*1uJGqVM#mD?Tf{;W!<tJ?*y?Gzh{Opau5
zMSgzB1;@~$L3>*pS~tu{w$|2Fm#{QbAW|9I;CjH;;kz*-K4j(G0<J|I6>uYEV~T;X
z3FbTBlU?AoA%ZCQ!NNy->mC;VGZ;v56rvQPTw(kV+?<n^o*tnVbOPMPc0Y*<0xJ+)
zN_rO%Uf}kq_HDji7n2EYi-0TpSjeushgTb`Ui8~G`-5xpbGc!tZ$K3&DiG?m@>tSQ
zG|-l@TtIJ<XbL#uTLEt@%p8$Amr*Xt7x#fQ0<#Bv^GQldC?8BAY!C_KTwGkvQ&kKX
zd%2RLA|h{{XF(mm){gVj1(d12RG&K_kbiya7<&muz%&?DMpOVhc03y`hapSl@kdTl
zW>MU0G&FHUZK8Dels60>QQW%}Z)u`;hd7KKdZ8Q|mXLJ|<huJZ2Jr{9`(QZ+&!-b$
z6_ZEd;o-TtbZdD%V6>pD8*Ysx!V5z%G{ZOr;3*0zASmeV;v$7P1f*7N6BS6pkP~Z@
zl@JDI5U)O|do$K{Va2x+(eR9ojg0Q)JWlK##}SA6F7p!3Cchn?Jg8BJ+F!weV`%s7
zy}zJxkQlgQH+jbU7MLC(rcqR#%tI7C4?x;b+#=8<uLp+Z>e?3&ARs6$zxV2np#Wgo
zp&}(ERiF?Lxfp%~R{?DOrbGO_Ca>QHO7b6<3Ux#1<xq4Q0%YCc9C6fvld-9dfBcvs
z>g@!oPDUnRa!WpI39vE1>kfm%*z_(6(?giynPtzS?^U**ceG}hzw>z5n{XXO!1%a0
zNWROp`J_8wu7g3kLu@EU;JS#1R0H@JJMazd;t~^kspIJ+X*ZrEC84QbKxHV?do2Py
zJ0}=J$**0Dz7I9Ejjij7G%+I{s+NK4oj{uL#=ziUu81d;&X@@%{0hTba3+_67B0Ik
zy9~&7BwMr>0cN%eu$TAtUfuySb3{Z0h*?@XIxq~enZ@2UaEHiVALT`(9G{rz#?y?J
zLRl5{5_g3Ik9PrSclY$<Kd`lpOsk5GZO)P<q!F@ka{90mo=%-p%|;`mE)=oDnqq6&
zGWzX8;%~{s(p)@D&14E(8jdaw2ft+V!Tz+29~G;#4p>4$N?f~&@AJGf*^L-%YS|L~
z6R-q*q88OB;+Yt38ZYOoCo8U-dMA!HDMMokK=r^y=$T;_|L6aORV}WrdV>=ZWT0nR
zS;40UxF>%=@D=s`1H}Qnk?Y`eO5^kDJvrD=RZ*FWT=LYVFoIWLK*6YS8i@d(wnykS
zpwO?a!Bh%|c-B*$()|8?v1%6U&6}@FOW#?vhqQ-Y9(gXD^1W5_mc?hx%Y<{G^RnEA
zHP#>&!=t$+Ah1G-&jPnZj6&&V;BXiN1`bl*#@4nVs?PiT^a#@47^K!q%Ph>y)*oL;
z^WO9E5h~l-+Y289U8&ta!^d}YauT&x@a7HjBneK(_QUh`$Y$RiR(txeu40=qGtRHF
z6Q4z$P`w%to`XybxI*#`T-(7p^RO$5q&e$sFg!h-l8=#mFby9_*DN7d#@j<_gP45m
z63o=3VRZcbwa_JcySrU&7?{LG_5VaopDPDlN<VkttJ}@B`cZVj8)h#}jHHGq^YSV+
zb!~AZ6$1;44Os3^NA|dxpoqgj1VfsD{>mco`jB#k@hLaZU<m}8Ohucbj7$d%A>jf-
ztc9yX{5ymRck$!z*lHW79CJignI5pi(MmuiQ$o#=)aO+~RiiLu!(@E~rUfuV&$I*%
z!WixOi!#J2AV3VZH~>GeNZb))W4ml7>G6YDpOTMsTfJmOl0Olc(`t{`$@EuIug7r{
z$;!so=>iRHalnR^gX1Bv^<cL$$xH-)7MRb<`l3JAxIkY6HOCEGB1t1$7^x*l^Urx{
z!D)k*3|iV?TB{4d0XS`ahbw_y8gL2%))a|Q7(isZF6NWO_4G2daHu~st--lm!R_Fu
zHZn4LP>Fm3#fcX##pf0;FW7f3V=z+R!PFf3_N^KGyC9LE%$ClnaeNDNbG$*PhoF8|
z7`+*ooz2w+HJ;n34kr!9SPLuir`gc_fkT4I=>&0!6SckrrH+1M!AgJ5?!g-X0XXe|
zVDnaBWCi9d7i{35wxB#spzgm^5OM$?0t7RdaTmYhGV}9mgYNYX;#%gaE^sw3vmh;i
zRRxAWo!b;o8jKIM&@aIrw`8`1G;8)jAN}xLn5<IV`wxx85?GNo8Q^DeM|iVlTH^4w
ze{XRL!T=Zy^~Iiez+bUI2m&cVd=5;l-`TOJo;CvLP~Kl{k1=z_d^_x*y3<d>(*qCh
zL}zEHZU{8m&V<s^QXj>NQt~8x__EdMyNH<^6h@$1<U(%q^II)4;PZQ!GfCq_z_fu@
zQBeUL8XO2M?>C5FOGQV=-(!9szcu^1_a!j^FSPo7aRjq+Y+_;=25LYSH9sK=N$?s=
zJLD7<?H@eY{WVE6pH_A(BPG=aYB`rzrwvl};GaqohBF5Srq<vKaOJt!NT5I&sM8?W
z0q5(giw$n4s+1BOcN1Mpu;f*N#0_EY<+TqMR}wm~*7VW3bS?tkLdmkf<tl&|dGer8
zZ23)T-pRj@3yb)LLQ-mK9nibyDP^$R0kk^Z?FUJZa*!H}5C;u1x7OF!p?6l4mV%Xj
z&<^m~XrAolsk*wbf$^*4%ov!<p924_R~WoV*v)WRe!?p#2untrfP)+z9ig){dHft(
zVADJr9UCJBnWTR>SJE6TH_sq-3xFB`_M*{-g~7<~rIQdA0~+Kv298?1*z@dc*s!5z
zV8|;j?$u2r95-0NonQn&YZ0!LWPaT)24ML;OeksSKxRPTL-s(EfhQVACwacTJoqg{
z%b+#ry#;b*c^PD@yGBOGFb=@~t-%km%@twN3!8!wO%AQZ6`~nKtpo`8{a=&LfYtz5
zvl|&0ylC+FsfC^>PZ1z9F*_R!g&ZCwU{MAXs}5&N2ky5=wyn{}Z+eVdw%$RFWq*VU
z3YB;kLZQ=tT&#d?TKayp13v@h2_}e_Wo2CYx>wO(Z9$y!mX((~xVZ4@>$brx0J2XN
za6DiPt%xBa(X!G~NZ1zMD0a4T%n6^CNVG`VXPL#^XWX2}%}3B9^~eAML%7T5%ouW4
ziS{5GWROn)E>loY9ECSDH0<trI2n4*FD&o@!?^I>4xZ<HD`*Q1O-=5)0rb)x!O$Iq
zJ=e(fyDvAF6clt)v?Qgay@9a;iYP4*NqyPWaljctvF76BlzDG5K}CF<P3<{*!$x8f
zJU3LXw^FF9kAThpSy21nV63r51`I3ro*FICHpkJ|0L9hVb>^8?1E38=Xp|EUlmt5v
z6#!fejQJP3Rlj|!z#WFu=!OnVlIc4SX>0P+k}b%(gN<2pRn?m!BG1Z>Plb_4&-M^(
z=z*u+_rk)$hM-7*Cl5%(mD0NymoEbV;XF^(F(6aBxe;p6itizlMvo<HcdD0u!%z!;
z2zrCpug?VjC}+1TJ|1kS#T0@lv@)=43pE++8?|UiT8V(GZ|CsRE_ZwcbPs&n5|~tO
zurM&Vt_)psGMrePpGSX80495s4NWn@q$S`8diHz=SfM~mJShuwyb9UV=WfssY}pnz
z7}GGM-C?l;#O+L2ILget#tHUYCKeWC@cC(_1WSUYg~cOt6LobWx9&*T4%!07o0)}W
zivjeOfPk2o;!=)pg)3Yd|FnTE<4kbJ0a*WpYMorFrb+ieums;jXWHNhbSW@A-IR)l
zBCVmTyNsucIfIeF@jK^Wd;1pOFwZWTK6^AwM;*b?^Sd9*gUJ=90^-RC!s{YIr#_%t
zxyienion*@HJ}+Apd2P#$1jGh2$&jSkp}OsEX%pL+Y-*?JBMZfg?f6F<~Vf(y0t{n
zm^i0+RprISY?Bp6u*A4Sa3@-w`;$6MSASz84Kfns*CZ}-h${dak!J<X&0=1?4pOjJ
zu+Liy<J|(wjchP^!77d=kv~j9r&Ls|w{E$@%nbBlAA80t{YqMGCYk7>qN4nKMlJGR
z2YHp1vw&Y=V$LHyIfK_*FR1o_nFQvcdC8{+mc*3vE6&1fCSju7n%aV5k^#Rvd;3l+
zOK>zbHF3P4!z>M$PrY?C2Hu*}>&qUv_kIGvOco_i=0JHSoHL+{nw!r;yi(HrA}Q*s
zB%*U)wzHgNF<-vUElAV(iLb@&P8O7&5C&ur<OL^6<T^(ccCQ+o1Xv}7Wz>(KK804C
zpPx9xj=*hMvoJ*7Em&-VKtnaW{jVS+0`pvb0K&Sc5#^UonQ8<6+mD(XK&5+t>!?9T
zNh$E)BmGh_)C<O6^tgb@V2YsnrD$-AkFUdj-W2vs8)1rrtb+d%j8fWLmA$^|^9hDo
z+;`IqyJTJif~~th*$2Z+8^>eVLZ6C9&=VORM|P8w_Cu*9*@&6O&c$LNT37`%!o-BB
z$bg!QG7};0%gV}DS61AX`C!)*wi~pYJMCOtHj@}+H@RGF9(OU{pYglq+gbhNKzrt4
zcV7J&l_a8KOUJXaEKSzO&hJ%-wH++FR&B);CJW+(NJ9p`D7i#{xN=>D{N0oE^dR7K
zykeVG3SPe^@U*;SiQjS?%nGDNtwG{;V2qfPpLuiVK@F74Q%jt(<5^gY-2JIN)L8(&
zj=f)#Bfz)Zy}KYj^R=_H0OqCjZ{MK~5R(~-pSCw`%>mFb;NLKOL4m1=26fHSn;=~Y
zd%WS`$wpP0dc+NXKASW5_wF09B_x+Af#qMp3zv?QK%q8v{zDlU#dQ{h9hVM4E!kFR
z<`fkwG6a?gWG2ZJaNIy&Z>*8xRF3B&m?1Pc#Zjcjn>I7^BW9F}kN){(ho?C<W&|ht
zd;@o}htRTV1F#4qoz`M~jYT`gI#%^`_!_wQ4F{wnj2qnwo0|im=zw+@gMg))t_FB(
zsU+;mlK_}*B+n1603uC|hC&DT{_`#KVrUiiq5OdM*M0^*j}WG&GfW1GvF_vK+$yzl
znl#v$f#S>kAS*F8HWv1UzJhx~W_mWi2TJJ0{<QPqQ!ahd1JeSwhvJj1>j&pnxINt$
zLCO3RanRMT@!wDNIXC2m#DGZScCF*U$H&L=@LFp!X`6+O!#TVFdc}qJA95u}(#pb3
zsb)r|*0DHPUj{5+k-n>tFlNZJ<Iixo?YpS;iq6;GC)R0(9yJHe4nRIxDpqKqI$9Re
z{;ka*C2Y*fy@%HGf^>^8XzbIcPR7rbzz<Qnx`W5`!KEr{d8wrh8><HoqUu0p?u}82
z&yco0g_YIn`^|O$7Y!Jitg!4*`&B?&0Ca}^*@=%IrKnF&ZlWeqMYB`bYiV#>CLvE=
zyQZ;L(hAtGb6#d^`Z8>^3EPdA#KfP4EvCWt1mJ5JA7c~08&H~<nu1CcD;e9?-j1%w
zY##W-etCf+gND|)s4VM9(3xB`&o3o{k2Q)1`dpeXwo5n$=+6-}{VVm3(*jUAhoF-(
zVKDywJ=;oWN#X}j&mO!FAb8m0ZsQ%q`oM|+PwQ#NpOnGj;qW8afr|8HPJ8P(UE`GB
z7Yo;N-wpz}2o}8Vquu)f^?#}v3_tmH*Ea63fB1(*sCohj7rX#J07^}KIv~9m#Azoc
zx=`Kag)L5t!1j9DbR{qwwdB(U9Rl1>E#sU6EeQ(E4nQ2qK+<v3w{oyNkuZxhnk)I|
z<nZvyH9o9aYJEzH5^6#efIWQON7+V!2RJDZR@M)GPZz&_{k=q!si*|{jZ_%qV)#WB
z+wYHgm<#lzD$q*-XIfZW--adzoLViCm%bJH1;}?h0I8!wVV$YVZp+O8UBm>S>Wx^T
zLpYZjvw}YH2ag29Vc&jeE)k?m`r?K6uu!38iG<tS2NDAM*L4ZbF;2a~rhLN*`YKKi
z4)80$OtT4eE7WYc1hAzuoYMjI5yKO2uYV~C2^-e@_Mgi^k+)<ylt4LNX5rAOdkIRq
z^u<A4USN208^yYi)siG8&uwv8^+3Ft`#{?SW-c{7y$4AFdH`zBUaWU}iB_S3mU`(S
zY}(6mg79v2rbgneVQo7T5*i9$#$rYI3SGIGjFZe0JIh8(O^uJuhyg2rz+sHpOd>$u
zi`Hj&4XhpQ$vv-&OD3`PkBWc){;hdiKtMss2<{XxH~^!Bq14u%?NC#(&qLy@+<YK5
zmx|=-KTS*XobSL6z!XA2_kIJ5#9xlVsinbsWp|=FAB`C=67D8?S%&5gEHk{gr1cG&
zH$V@~{+2Zu1<jkhUeC#gv?*_j!;J7ed~$u=;WUBRI+pS#@weCI<uT9Eh(S>Khfu%}
z5*0<Q;P)-wk(H5oq4LtsD@l_vNd)5_bi<efh7MPSEq_gm26%gWJG=BlG*__12__km
zz!Pv~R>RUx_KDzx0Uy8>v)rTAkz67gp(-C|DlCuv+b%#dEsXr|-CF2|`3Az)WFgd%
zgi$%*>WvwDC~9yP<+D7JKyg()grtY{Jy@rJuoi@YGiHyd^!~pQ3FtW_l25@73<^)u
z(w32cD<&o<6<>S<&h=rvb-8KN+d8(ZV2A)qIkdm{i-*-Pf@!eI4bnc;p~3NiMRmqo
z>=fUdrBNN4lo3t4;89*2kN|LwqSk%HKn%lkEL46f;{YFUFh}_`46wzc$dgA<NC8xM
z0y+ha0mYWa!ZRdw$<_t>zy6dAMeKy(Y))vNUjtKD<j^C?TkHOb^bV*ASY;%0*I~>Z
z0|p%A2Sa0H_cw*~GIrLJ2+*N^NV{USqnt8m=Bn{=FhJCb>ZYcpwffZ#fRhGHIPf_Q
z((zGo@(Q*$5^&<T)xW#o{GDa7Kxvi?I&E;7?|^=nXQQJtzR(?;%!j=y2Ed}-?zW&{
zZx*CCZTg#kb~r$m^(A8CB-COWXXpG5s=&X~6i9a6{v~`)5T}KJSU^TGnOGeGL*bvv
zO1dQ~I=T@sPe}>V^y!+KUgud=kAad4mHRVz|L;=ohJy4HMS$XXec*5*&toswz$!S@
zV9%dp-CbQT4^8*M%K+<LuyMA(xj9#Q2<W^M3S9#R9h9$Yuu%BqiIyuFI4t0Wo8T`3
z{8=nL+}zr_2KE`}HZn`ri<Nj~`AX!>cKjp@5HaX~-wyVEfJ2BW8J(RqS5tEWVk+sH
zzHS_>dqDVD4=ju{jE;}r*e-GHo9OE+0bK_WC8f)tqztFnd$?9c%H8d41Wa5?CAT=3
z6px-Ce5$V&80qk}GJ0sfEpDMM=uG^V@+rBatPJ393e2epq#ZLCmp_1h0oGbrJqB?G
z_Hne+N#w!%iBpT(2hTQ<zZA}sjD+>m>Yy_}fe`3hB_MM!u7EUvXU2?k8!CZB9JsAw
z@A>&1Ljx~@&;YJ1li!xHgEIL1k;Z?Rl6jP7WoAZ;2bFFPIKruvCfK`b0G<!9d8&P1
z52|V)kawk}=fTSXl1>=*>({U6^)t$W7%8FQzYpi2(kL<t%r%Z%cPmte^wiX@D|Mx1
zW%xfAWc4g)6753OO8&BvVM!r#NGMgn9snN(?13x6d^uzgOM=pk3|N=8u~~-k_<f(_
zRp8uTYZnt@JizW5sH(5;wFmddVW<5+{Dg9I5(o3YV@Wcx7R*gi%?8b#tMKTPE?WE&
zt)gLF=jnrTu;&Ks(e{#pdl{hL%Eq-pltPfOyR)-6Aam*8A`(U}E87X&tt++!@IrvO
zy@1_-MPo)LretXbe%P7@8z_h*TrPMX5IP?@_2-zx#jpSTJTw%?g@5fooTe7o{7Ulj
z+9~juczDbgrl>-22?)XwrlzKid~}fIZTWa{vabN#zIxS9armee%ofmwdeXw<VEK4v
zGX^3K3lbEE1K>OG?VLepL=n_506(G@xXZ-P@7@!4?KJe^=Eu=dz)JnY!^;CUE8yn<
zn*&Esv7&{=3!toOYx#iyd@|;~lIeR{@$A51t14qx8{pud<iBtbNXxFC9zvKP!47M3
zgb%Zryd7-ht`02E1MLHo7kpI6b$#wRfVzNRt7~f)2bLKiJdBLq^))2=`uYMS%MkZ}
z`*EftIVYRTyUX#S@Iv9`|4WWGSGAuduO^YehrYN#qQOcC0pfOFZ}E@*HolN6rFnUy
zW%@iqW7E^qot;(?-@u5V*eRn_y33a%feD7EBp!G?4py$2nIM?SV1y<4Dj`HKaS^uW
z4=!J0#6Sjp3lIQd2z7OJ@Lxh3g8qTfdd_c~pnpep|LJu(5m?Fx4n<B2OX3t?)oqY7
zCMzo|Bm`qa7VK%2O1u;T|1I}XabR;v6DpAh>qfxJpFcUzo(<>*@*gHauO>~iEkUuD
zXmHdeuK$vJpnx;1f7&ayJh$vP5vngmNzJHQk4$y<4iWv(Xz$g;I`HV-JTa<f%Mo<1
zd$}}y@mBe5pMT}Y3c+7mwIpwR>3_?*96=##Su!oT;=iwgs8CpokR`?d8XF0A2#D3Y
z@D}45Do%k847Sb=`W|9^&%n$4w-4H}kBHryyyh$Be&OtK{tWY#W*B&<VB4#YYulbg
z7btgnI~o?R&T5#7P7s3$t*sZJOU2_?RaZyH#x7Fo2msP@fKgsW<%&LQDOfC|$Aw5g
z)!3-{W62kW2R8a*ri?BUJ5EbgU~eIzsO}H)w89e0hm2bVARWWBSyflJl&;H3Too23
zXKl^=zc1ppEm2u}{=&U~a&z$JMTW=7(Kq+_6(eGED~LddUCW*R{(hjmcnecf$jV-!
zfFdDM0ZM~Sz7gCymUL<sbbnY-Oy?puA%U&O-171pP3(st*FbL=f^~Ut;5~k9LOC)#
zjO63f($LV*(*yYo1~utIu&4tGJgNn3#f=+rZvgoLZb<}P9h!As-aL`yX`6Nd?8)S>
zfy)~{plHO-XHFO3UC7-n(q=CZ;#~xc0Rp(qwV^j7&HO>>90#-w?n*{x=8Gf6FJHa@
z$W2U3(}jTmY9UZd!J7RsNLa6ckHBbrjPQMxj9%~g`zwXQV|<|02ymstKpTTP`2e_k
zM7c?04IoF2)VFU1)~}M0x&It1M6F4LCw+lo7kC&@`bOPg4-%l)m)79chzJEBlwrY7
z-|IDr<{RiVkXY02Jh>3y5fHhEZ%s;09&HW2fB_EkgKQQMr`)iDxd$I*0F+brG0}Aq
zEbO;V-;aS2Ok(U{mq0^EEoDG!^J}-0;4!GO7<+jK*}O;uo*#(IWWgxs;syllXj;h0
z36`iC89mbl9ob`T9QP8iBBkc7Eug}bse?dy_io6{$Bc<KT~HB3!K(&VK*(n3RDGaW
znwyvD-MkHMX{Yg$Q9$d4;xiQh=sz-WVGMN!1Ib;EbGftNPz<=XO5=KNo@TWy0aZ0M
zLVi#sZ2olhp8Tq81*!oTUG_KZ%L?-Ib3HJ2|C?sd0tO|wev|7j!ZQPBJ$9rjm{UdI
z)1#g|eF|{)3{C?yUO=D3*UnB)8$M$f-)49TOrp#3NG>2lV3F3~Oy-xH(_#4{$^#xm
z&6_v!jB;1tW3bu=@+`}ff^Rte@PZG(0h!5C-raqN+w>jy5I_$*48Ay_lep4`Kab+=
z^U!a{33d#@dNK@kTSh(yP&`X^w>Z8&ZY+WxYsZfCT_RYR=BVg_SrlrN2pgL%7`(NM
zRD}Ka%K_&BeF{cF9Ubx~hQL!FP5f9}GlH@MTM4eqAz}w3k{2+{qBf$>&(F)B6e}=)
zfLs1EU7z^<mK%U}MaAB4-%LJV;YGr3b2Vc?+@4fC>Z5D$zm&BB=AW0m4MsXy2fzi~
zDnIAusz9s*?WCiCz2?^o<TWiOP1(G0r2u^JFiMFm;pIxohDo07euA60)#~tt@0?bI
zo`+{-din&eBW%?`6+HED!VrmhB6?(`q+|K=_$B!K@QN#*F(|%bJc<|35V#{WLaV(i
z=UgEw@c>nA<x5D8cvx9`W@czz$?Prm9%T~!PS%F~7wlf!`}cEz;nSiMaLCBXp|#gT
z*$YrLZ|kN5Yzn|8K8`P#Gvv*}>TV;EnQl;~KEcBV=72`r4>ky>3^#!_D#;b~=GOWO
zJS1@42}gZ>O3>AT*C;n{ooQ>c0LBzjPfkQ`5Ii+iPXANccZXx$hi|u2N(gbwE`)}%
zXJjU2kCZ|sBT}+MHra&Ctjw$iGAesDtZd2(QAQaN@A-Y6=RJ<!@gB$f9>?21&vW3u
z@9+Knem<Y;y3Xr7&r1a8i_ak8+v6qOaa_M>TdO9hudQnO`rtEEteKxvRh>avdXk-2
z_MV>O>C>BNM9kXKfZ_fatYG|^`Q*u_u;)U{hh~orQV`S*Y&nhqr3gziRkUL|)Rcm1
ziru+|?v9_kX$>j?=sW~WOh@x&Z0x1W$~=PMQNwb2b?V1^9;wl6VQ5K9Nx1`ZvE8v>
z0FW{_&*}h(dY7|cfL^*b{LNrrgsRd2A@uu5EDUMX{OW+Y=+VBY?>#kpOl<(oGd;MG
zQN2<Ss+BOz152HYOBlkq?lA2wAa(>D*3zR>kFn~z3Pm~z?QmAVHX^D-#!AfZ__sj$
zpa1`VF$4NPcv5nEy%h#od5EqKtPSpZTfeQZH^q@Ikh`%2XPyqVi72-ek<!wnYLoU-
zS<%P*CUt~QT$S*b;o&wE+#xJx&1;GOTu@NCOk~4ma_OTS7j%cVy*sml{!Oky@p*N1
z7^`$*%|p%&&Za$zB6UE52@w&`?Pf0GQ_*9PdkcX-bRRu^eb2kZL`BWa&5JJiMPVC7
zL=&c9l{O<g=z<tdJ^h!KAjeq#sj0~v`Np(2v|tWsG$k1m#zNR&dNb(P_Us?QUkbv2
z1c)>ti_7?AG=>_pUJ-R*?YVUxzoLJ(bR?;kdFWH*&b-zb_K|hQpV4zaOq%}cwuTHC
zug4A6NVqWVPY6wR>nRTDt+t8%_(HY(N;JC?y#>&if@mDmC~1)x0^#q#6(Ds<1>->&
zC-tX3LdgxwGA1S_!qO!`xWNY|bl7b@r|MAyw1ca|2Tlso7>~F(#`s=z_;MFfav>J2
zrCtGC{_<w4N6@q6w6tH<`=kSrwrgwC{sy*+AkzBz^B}sCh#H9YfkK&<o!Olze%1Fq
zqRaK<-k+wB;v;KL<GgrfaVGuGlfTntIuu{gZxM>0H$P@NbZh<oV8^||zpn!9rw!#2
zUsH)#jzy?%i2S;;rewYn;YsJ(m?h6r%tRsuxa*BJu~YYfVASJs_U!JC%)Go00pNXo
zV!Uu3ZV3QiARLkDN4Jozian=3XZCo@?&kUPnX-?G7Vkplbni=E##{$*0&dnZizk9&
zhs#cMqM}ST;H$DAM9n<@1PR@oJ58bR+pM-*iR!(OoH#dj-QU~4n|@`y;5Zq(vL^ia
zkEP(^;vOZbJMY~-wbcWIjv5%YZKxXJJ#mDq_~sp$<#^vit(AtXov}(8X*G=2-dAh;
zSs)sj)`HKQIR_Xax1occi3!k@>aJb>pl$1Xgh1Epqu$eP4jlMWLLc)L$`imU$g#Jv
z$%07;Ub5gjtENUTX3KrYR8KFaxY!<GL)i7>XhQ(=A??5Z;NMfiIOIXP^d9>@(E#4)
zAf|6FoVoWToc>n(&T66Y3sVB5^c_XlLCQCv;eX>WXZ(uHoM37r5b9Hz<M8_hOjv&o
zn4Z!-H4a0)6cPg3%*aS}07wDQAYEGs3ky>lK)mPQCzxkY9NhT^RJ6&v0NGYjml+vx
zVp8awzwq(r1)75g%kf%(Qs``6yGA2fcrFs<QD*X|21et3n!p!zk8Le2b)@)FSjo1+
zCDh!2N{}$UV;~FcIwL40#cV$MyR7lU2S{@MAPdr`I_sGeIt2<q$P3@;hA9xQLKR_e
z_8E`|5Tb&!FjE6}t6&8mD)Ru?M~p8CfOPXz^11Swh6X>{hX89m?SK9LJ=Rpy^-l^+
zay&Dg<xXlQJBW<UpC&USqYHWokax7Kx(kv~PXeKHfqE7D2{)v^%#RG_6&l4>m*q~Q
zPCrDr8?LUbjE#-K`L6)<0iai1ot-eI0iN%15<r8`9&=;kE&A+o>nq=qLVqC1MK8nK
zF^U9Cr3j+tS9c0flOpnhH#Cl737tTPXh_;n?+)L``7VV30puPmi@VSK^NHeGR-(tP
zF+Zf&)8axkfeS+eg>(U@{}Eo^B;a&DXAY+@x*P#l^UJ%fUT;vWqCWR|xs!mWo+8b)
z`~#!~0*s<KVb(X;G$5qrWM_ke5H?_IWM(Fc^cgUQOd&TEtEko{DHD-H(=0E~nP~mf
zIsHqIds?k18(U-Q3_*=KjNy{??6gmwI1wty(FAfgE7{yONA-o6dh&dcymOS-5xgk9
z9Kz$Gqdhma4VGS@JPaBe+)({wiGx~ILC5OG*CV`T6PGRhm@<zS9_|_@zith-+X+B%
z=G-h*QZsKx%hsOso%Jj(H|vop4&tgXrao1wustp9hPC+W`O6!sVNr4C9cvHMnm-PV
z;*>WyCg0p{Xw&F@k$M|j3-KXMwa#qeQy~g@c3+Zwp1FwKiqO1yF1hdkwPsPIeaRmE
z#Wbbn$+(1oz_!rPa6yBv>kad=M}6~Lc*<&zq;#7|?DC)lbqPG&lgCD0TBj(UrrsD`
zXKNJa7Cn{oY2{;le>n)_*3yw3MYE6F)}r1W(bGyBD+`e~pzb_<+F|NcNcZqZIxfF_
z;)}~l@R$EFnrPw@3V!6)EX}rv(k~@;cCRvj5yfo>m5)PNoD>(A_iDLmxpeRKSmq(h
zL(U(s<$kL>m(1`9r)*L0)Ee*{{+D4bkKI8*PX5r}KdV1%H*<-?zKehKx&j&1A7q;2
zg1wRR$f?8I3~Ouv-YYwI?##<R=}s!l7_<rQ&{M^|AuJaBs8MKH0;n$H(9PSU6x4~*
zE)%R45AFv5W@hU`p7G&_=Q<Hn{UrbSG18s4!9Z}}*+Vc!(Ks#s!ic{8DS`PN>PA54
z)FWY#fV2sa&n!AVfPO+lL!lV{5_Ny8eSE4@X8(}!^!rEtJU&FgCrZhH;U7QNVNi-i
zMz#>j-Z3#gUPpJI{GGbl(?u4RH~z^Lf-+>H#HTj_Z362hHZsy{1??SUDNauCi@|+`
z7J-4e%%%Iz=%D8T*D!v61_pt^GqfkI2PF{Ih4O)AWM(#u`<t^$a_l!)uJ~4I5Zr(7
z_TEwRT`5j%f&qI>pT!%f3NpN$t}7YSrTgBU@WNr`ORe+9)iee%aoxwG<KuRdZSv@B
zFM!3119J!<uYsHvt)ru>OzTq8YW$1a7t>6{t$H4{He0`l;up1E2(UO!*-4(Ie(eNm
zHo-3Qz^CmEogdSp23RufQis-f+t*(jzFhsxJi@ilaCtnV@!Z*nAs`9{0Bhjx%g@5Z
zG=OLfLL5#aun3EG$fdtk@h$r(p<mc+^@sG`b%F0tr_~+<R~7@2R||%B^RJZtj<vYV
zc5`%ZN>wOH-e-%TAxrUMZMWdZIy15Hk2j@QDyJHs9Jb%dS1zhCz$2|X)tgY%av_3+
zTUCb!y<AX#<>lnAGVhwlaReer@$=^>$4vJ61$f#nQ%%lF9%l-^XfCKW!}-d{vavg{
z-pn}1u4SmIt8zkcsaH6=kTbCSf$A?5B{nv;o@Wn3!^^PT*&r-RH+}Y?;>P!eB+iU)
z8OmjA0+XYR@@5($nU6WF*P<zcl5$7ezS0_s)z~%n`K$;}pH)xo4k?E%fF%+QW8*u>
zav%e`{Zj5a@tn`9n<Qs>36w^*r4^`oDhxJU@a5-YI`)}1l0qepM?T1S;|W><WmvzU
zkUBazSPoT6f^%VT+65HIuw*{-(YUk95f=0Lsy_r>=RY-uv|SVA804_NJvr<BleyB?
z@7ej=VtlvQ1*5im$pi3g!NZ@Lm~e1%Dsmd`BNoe;7Op@k_Mnp4yfbU?zJTV14SLTR
z-}LIju?xQGst|(QM*7`NbqXl}LMP0ea}$^Cl!Dl+i!;Nlc}!w+y%c={1kChHWeO)<
z`OP*|Tg_~;disZVvoV{)wXTNlPEvaHfL*)bv%<dxwjwEOlLasQJM`$FNCR|Xhx84_
zM|V$;&40IqM>)v*DRukk@|bSaL~lY~kJ+8BijboEp=^WY`KR&+2H&4pZ29YBTJB#&
za-5tUGrnAQyo=)28%d@pA?*NCt?OtIvJ;4cQ6oA!If13x_OkA*+*+sKGvB!VHrcX4
zE}HMcDsLnU^te@SkLFZO{QN>7>)t|j#6Ewx8<MvSUBCA2j`uvrMRPf^K)xeP6wf+r
z^?J3AnA5C=34<&viQfO>w6nZ?E%XRu<2o*@hnA^K2^FXWA`L(l2x4hHO|R+{lE_s(
z+DFf1ENvc6<P`4N-tWxvk#@IIbCY((#KKW?8t*XS%Cs}hYqrZ{y~)oG9cLQ1rqv#<
zV%JR1RG(HHgs<^4wBO*$47oJGY<_fN0{1vo%8apoFo$h#VQ_G8NXSQ&m+(^InPODE
zmyIX#wf9ilXyBTFS@8vz_G{Xi>LQARSGy=!L{<uIMN*yGjm-Xe`*${OU?3yW*d|Cp
zf8w-)f=`E@zTa~a5&~;@B0UC?5aAV{7}oiadx&@@8ttPLW$n89v4ON90{XP(G@SEm
zo~5=Whc<e@j;S~2zEQzStJ*ms`fchZgkUnEq#Ozx9l|dZnZOCSZ6Dep8Rsp7+aD(2
zX#J&%@$dLphH{XHz{<qowrk-NPnG91Io!nLhLe)bs-r2GDs}OGq>L4px3YbIzg@cY
zmpY+*r2AJDuUkN%%D46CN=o{$sR@Q=i^VI=4*W~ovI?gtO$A}zBn!$@b2DLCjfCgJ
zhxMn4OG8X-nkLCdq*ce4)a7Db0!QDeL`;<nFt8h6P~3k&@Xq8cP1{$BD2j!j;xvi2
z+*hj<1gDZrIdsJXKRGmUPHJE9$#b|fW5c~9y|jI+%7%;z%Uo<>RJD#o;T_7p%B(#a
zCY@{_%v+k9@7maux~vpEdj_k>)<GY3-xE1u*25PfC^piSRV{)fue4puB};1zvfCWY
zvfz7iT=Dn(YRf?*T@wWb9WBc++nZrY6w$j(#Q5&D*&eK%f6(;dK-@<MYmv!D&S(1*
zzXwVjlnh|L(IY{A<!(ztS76tCX-15+ZCr=mJv=*Pk*}P7o7*vK>*_86w#zk?qoNX|
zi*+#5xY*H__T$-UCKYL@s>&PZmAgImGas3(a{M9V)o$d!U+Q*#Y1gryiVyGK|GeHL
zP5ZOD#wmp^siwr5$CG$-M`6OUx8(I}09`L`_31#-k+z{Dr;ew&%dE()P~h<q0ka~W
z4!u$iYgWhSwk2JoW-?+g=o)SHnrb?3nxSU~4N9(9O*!NFe;M<%jEsGdrvmB^t7Ors
z>D0T)lgOz^o2LA7=2i6c;O&j-p-0&w$rt@OtYi8r)%#pE_p}+A?bqZOC`V8DpZiU?
z$7q_sG#=Umuyb)nS7xcjPNzDl+~|}yD*2nr^ZKjS=LSYc$5Kt5m?h1!qdWdPe3XK;
zcOsfmYU1u7mCDY%H?#2V=f)}RFcR;WjE$qC=x&RA3WG7V(Z2l|#A)^43o>_^AACK`
zpEZ|r+x*f>T0=3{`Tbdie8f`MNNqT6o(mhxA$#Ju#TCM8g#j6dvbJqCn_$#^;-^pR
z0Qh8NWJHj75B&Fn?BVgCCXyon>=RGw&~GHHWw^K$fz@^B`6Kl0&@j2DL2I7d-N0DM
zcEm4^L4V^u>_pXZ63b|7gJp3-RP=F9wiY(o69sf3%2m)7z40;e{ZZCRhW|dp>zj<t
z50-Gw+`E^`H{Mrw@jvc=^x=|5YAD1w5L*M98rb;z(+xVA>P+=iSitfwg>j&42D$*H
z6Tx}`<&qx-Xw2;FZ7cq{TmSpI?u8uK7jSOCWZiYO$Nu#VsdBSshMx=ep3isZOR%oB
zTP-Z*>>a3o=E(U{VxzH_cVF4S?LehJk)jp@f-ns7aQBnE0L2P?(!siO0a%B4TMVy5
zlu1{B#epPKYB%Kb{Q2|v_)EY|f%`&$3#;!eh;NF)xhMlbouKoV2uEA93dOR-1jQWd
zm#(Z8jKn(+j1PTNy>(;2u%uq_D2K0w)?4!Y64!?;i)xN1#)sC9D>`+y8<Mkp=$V`H
zHVvGw``(tOz_zvQ035bpj?VfP{Dz+1-aF({Zmt{a6)-P9_qoKz1jQNPGb@=Dkj^>g
zZ-OUD60$4-)2rZwl`WN51Df|JkXe4hxNa0cV}TQ4q51DKyk~w+sqxP3ko&(l#|!)1
z+P8L16tDY>mU^ErDU9N@lSy5-p)XpSi)T7Ez9d_<pvarHKU=mQZwOCJAlqGNej9y^
zNKBN@Oaqw+6(`^WixDuw0f}Ui_khcJ7x`RnWgu$^yF1Z`$VhUPRj{?U|3)`U#mMLk
zvm!#%0(&<#U0vVs((Q!lbikgpx~a4&_@9(-iJ+TtzIBI}n=`C00`&IeEv2`>okMTR
zJqX%zpekw%`}aSGt4mFd)crr;c@{)%YVzU(FL9SlQ$vG<>spKfYbaU>Xf>iKx;vkQ
zQ49$Q316C{034oCLNF6LSn9lFwH$76I+DRghr_wN1ONa~Q$W8HbHi2A>M+u06<6S*
zd2K6aPD9<+)&@p-JDgVNEwI7i){)tQ=EZejSn8r6mmrg(8Jdw(P#~!2s2>X^;SRdZ
zR9slN7}Umf5Uj9jz+XRmmglv8)Q)oU)nSugGkr?{g;J!VoCOyoJN>ruA~7T!aW3`5
zOE^+9CM74cQM99bn30{`bd*vRJJ|n3NSBg1%?<#Z;x@x_ei>lpBgkKW1n7bh4RRj1
zw7NIt^2vDSzH=l$aBOWX`U)fjcLA)T6>jzJ0AxPJmOUOLtKcG_izCI#>TaNRgY_r=
zmX*FETU$j(=W}ao7}C0+pQ#E5VJ)k_7et^wHEiVq{hC_sb%%b$vbsNos$Z@YfD3N*
z;HK<<hYj@&L5Kjdv$8t#E;Fv?p*3CTOjl7+0Y?O{CNw&Jzy&Yx+lfN17sbWswH~xn
zL)=H<K@!)D4GXgYWg4hnvZ0L3gR$1#`%#BOBsl7Y4!|?Ria&rb`?sXnR(cY4K!U(c
z>2rX!FTd*!>jLsxsIR3n3I|D9y=f4Tf?vUhBt0kR;mAq5(b^q<56Ba@ZZ=-+C`Hzs
zGZiguI&cY8@n~Y-hfG3BYUSgaCkZUN9$43D<-wO@OUAnbAJjrXsn>m3Tw|o&Dhmtw
z#i;@&;Bb9J#3>3BBnHvRU;pzxzndQ=7PnO8r*zvCvd-)}cp;<UCw#IBI&hb7^LaKw
zUIBj=bSBn*wO;@#JmlaK84&@PbZ@J1bu6Ye0QT3b@b#?+vOGtJ6Rx@s?fxPsD%b%f
zkX^u11h0%7duo=>jt?yfcy+PmfYb4_J!2ma+bP5AwQto|abg!(*Yr|K{{H<tXZR9H
zb_xQHIXVEcBiVl8nVf>?(C+e;2M+5U4__o~2X%kdAp1oLZtk1NeGA3`YWmNl8uxtY
z?9|rRXHYb3NX4V}u`S(xrQrE<M!1FxJp=&-XOYg2>Lb}CgacXgwaN&}Y^t6-RK&2G
zG!ZrdL0J~{t9S$+CaxH~xxyPmm^eaV7mh~Lw<hgP^C>tij|)*A)?hL(he$^JypgG@
z<0Ft76z-$HlCxQj)58Sx@7`i-HvscROSK+ET)6mV6L*w7qoqxEg-qx#(B%lyWOt+i
z>n{Yo{Td@n30|5(NKum@7aC`iJ!r`B_1(4U>8|2X+-f5u8w_M=0Uq6jVGQ0ohs8Bs
zn!_PQ4Qi4=ll!;2Tce7LVJp6S&9@2k6gtPJ>Q+`iz=4I;LNY<tfu<+UKy7zJD$!@k
z22_T1Nn3oH6*f)?`b-kNPh_@2!eHb)`z7O7ykAL5h!eY@4j1Fh7Uq=$Q~WUrOZvYx
zUS{*ZoUOKTzy*8!YB_lBQu3h~QlVy3^}f)E1^9F|i4_DteE7XRBLw99amXw8jz~E#
z1|DUvJfNMfFe)S|S%IhgIq$MFmMuCOnSsH<pF2CfgIs~UBVXRkluo&V$4>|czslr6
zh}-W`bvnLVgH0Xl7D7tP{C<z3IWYc!Z@ue3@g2GXBYmrPOPmry_Xj`d*5__nt6pDK
zN~^QxI>jq2`TQL3kokezlh@_+&N)p)H%d!r*S3hOei?jz$L4m(z>VRVKT^pJj&2$z
z+LY9fvhTIEFFI}$qRz`cAt9*qXNKTF0RIFxXves6kgV)j4*$`kQ1<4cw}?<a0@)+n
zbk+Pva&i28X&`$<EbaNxT^4-wEG$XriV>PYjB$Z?9oL5E&M-xsB-lXGgb6V=eXaE_
z9**uN5S(CYME2%<tiU9ok$#WtL&C!Q++4FJMQs`%+bBN{n_~d#5wj*MU9lQ7%@F$Q
zt)SC#b_S5<#8;EG9~Pv3Y+m7tIv!4X!{E|q@WDY}ibD0=%Gvxkmy5R?V7V)8IcYt6
zz4vabePXh~a0a!Qy6LsCKie|O>uJJ=nIfkdVy0__OINrm=2UD);x4&R$KM}zjCgoq
zt=WDqIMJqjvedL6h?OVZx@VLWbIuB3P}wLR9BpR)lyrAtK7{s%h~j`V?Pc1Co|zCN
z`g^(@URS-H`P{5D_}te)D}fZo09vnVmU?F9Il(G7h;`gokf(Uxnm~n@%gpX|Pw((H
z{lSc?kukJmy<cNt-b1?`!5`x+^08qXQaw^ldt{Ubz$OQEn94NzsKJh(cgW9fDeHRc
z*Rjti$;E@*&c)UL!_Rs9TNlWg17QR3>u5$pTrd@y3%blch9!)aPOBpfWpDQjJy6Vd
z6LI<N`Ss<#7*Q*xtJj!r|GDwStdG^Cj8$_rB<91wuFBiTmzz>4JI-DF<~P8;t`<G<
z#>O@J_<HP9h0*7cFD%a~XZ0v2?N;4&M1QVvLv`2ojXKWj^B1~b${)T)^LTK3dV`fs
zxRzU>_3x(CPu<34(#+UKiQ8zV>H;U8I!*qKHcH-2MY*nQ^<BmL)^XZm@n(xRd?&v$
znOfR~kHtN`<-+!SZu|O}D))KAjd)>I{SB8XpKmbG92RGIF>UjvHiYh{TaapS#pU-}
zCOUSB^ArbAS3P`@n`_n|YicrI{FV!J^{V{Enxj6>BcSwJ&^&rthpZML!Rezo+c8hY
zMELz%1CU(*Ai%2mr>{v`HEp*<3$Yi6#}&7~FHw-|v0ZDNzj=7eWKn3q1q&micb#;c
z!pDF5nAFISOLN4zx4(K|IwLLXN*7Ya$KpHSKxeeX-nM>y^i0<TbUX(ojl*`Zp8x26
zEh(gN-6#o;6V=A;9K-WVt=%?f3YC+J3R3F_(T>}>->$4jROD^Lnc$iaHm#LidW`1X
z23<(m9LYWx$FrXwBzvpAC73zj9GCUB#1}^d5=yB2QuhveRa2}vcT8>Fu<id-z200I
z`svf73(h%`@dck-$Mv^W-dg-kmwaj9_AI%SBcmq^ldt0Oijmo}1>Y*<ip(tG{VSZK
zDsMt*u2eQk|JpBp_wCaVb+c~S+nEAa8&gZ2@{bMfdr<3WAR8o^+!P!Uo^zvW^RW(*
zyUop-!dq>o4c6%WI`mRA1q7jXe+7mU5?8)E2zn%XdI4BbfeZ{Rlo;FezH)oaJ)~TI
zA#hUQ`FZAh^9MxB8}DPi_eS!%?VI&=>Jz2jfEAB9l=I7@EwgJJ{<e1X@1(zmdK4>G
zdkq}6i=BMD(naN{326Mh74ycn0->5R{tc-<`^E$QSKhuCuljU`R@?#<b$R9|yC*K<
zFr$mOVYg~>)gg<CLMccv;H;(z%*?>VXQge)NJ}fwe1gi+STg+0_UTp^7Z*J)4JG*7
z`%JUZXzvvIdv@s08>T6rg_%RHO=2dDUj22CR21nu$l~ta1%Rxx{YpjSW8%A?6SM<y
z|L^N=pBFD*f@P+2<i1qlKp;K8tcnWx0xBLglj}i|M=uiL?REa#>iN$6IzZmJ>RLr4
zM4`q<ydlUnkm?XlKMObm-AO8C(Pzr>8C+viKAP`|q#>=2<V}`Ia2l|B32HP>VkFwh
zmxQggPNmiT+>%wO@Lzf8f1-c>3hDp+nFMzpoVSKhMh1CwaGxYD2H)*Imj;J{oXhe5
z>KzDNhO!j0_`y>~vj`2_dWkG`CDwnFdasV<`&>pkCaBbrT_VgIzR8OZi4F}t0S^LH
z-7OI$?`{9pheEyEMO^v*T-(&y{PWaZq;34Wo{@SSf?^P7U65b^BWI?jt|DpCmn@<y
z3jd+EYY1W(9kB^jG!S#H*Tq}z@%^<-Qk>;%<}xBb?~s<({+<!<lR-jHvJ>W>LN#LS
zsU99x)^GlTGcF7Qj}ZNgaClr?85p@NZfkSs#a*JEOZcCc&xVsi=QEn)P_csJ=NVeh
z1A^p{qeuUuf$Oq*h9AZqh0&`>`cQ;Se*fOwNj<ddzwb2RHHF+nEr|8x$vY78Saxi~
z=%mFH-yp}us(%Tid@xmpaG3KJ*jJfj&7<4-3mX8++U)EkD%HsC|GiNr<A;BeoIOHL
z1DIGw@mR(TjDpNSKR-Q~D)xWIj2zJFx9!~dyv_rL3eZ;mgoL+h0<_3UiR@X{E#`Z6
z??&YG1O@fqW$@tNN5ip|6cH9C<Koo~N(Jmx)Edh+!Lh3g?McG9q3-bfbkC{W%^Gqe
z0I;L&1MO4JW>E{GNem8{_UGe49o<NDF+s&7u|vr`mzh8ws9gm;@C6}p+ed%m3vy_B
zIHDjs)BlSeI(BkE6KY!&ia^LGs<(stqmyqK<mX?}1GW?>*rnUOKB+aD*HG;rPfblF
zbipQdd$_o`Ui~sqf|Ux2N>JnY-QZ54VG@?$Cqlr(fU$WB%K8xH&^rxdJ;0~;1*5&1
zP%+{*gNFvXf)$dc+{1n$Aq3+QnQ<6jYsrn^d@gFOSW@AAIJPB9e%hQxHJ69532w~>
z%73`R>&5M^7N71dG}%e<lIQz*gH|FfhjGzQalj%B;6A`u1b3^>b)QV=Wp~_T%t_iz
z@q<64g1v1N4+U!mGuEuEt*=O^QdgZx&%y|gLS%Q{CvTst`LQ}TKYtNfGoI*R-eqx+
z<Gs9O_gNe}4yFOJYB=B@));|ZOG$66S4mE*P5jeTXtm}o@^^cZyw}Eq9$WCCwI-(1
zPBmmuXsvOOGfb1C8bp&h_si}z>j7NSY1P;v8%Uf0qe=m&T8zp~q`b(Z8U&1n*iWZ!
z><RYw*~4%Llht%K&@)DZ^mDd0xvvUJ%V*sW;N}m;Gd4!pGbphm4Adeh6$QnDJ6(b|
zQd>8q7+(#yZW3ITTF85^Z|(ogf7NZ>mFzjmX5vZ7hm@sqi>-76%&*s4C)}78ig&;N
zHk_tRrOnaM&`?lN06lTdx;eDHAa`L}lVA9~sSfb8(V5`r3Z}HeB?c}pFB6;ORv%ql
zE!_H{#8H2okZ>QY1MpL1L$N|J^Fh?&5*^yuZy+8ff-m~a*o9e*T{*S#K`WEDTS)cl
z8)6h!e#@Cp&bX52?NfgqUp6-Wdg~(BQA4L?1p?utHIho+J$v+P*~J`dY18(gciD6+
zLtn%Dss$5=4QlOqcbI9|b~gC=`-|exLg1&QEjSg=Z=z7T=v-r;5#HHAwtF`!ZB4>>
z5*8qkf-eTX!$1i;gYU#80St9>bNgH(@F~&W#u;?M>0vung~(1l0VWE47n7Ev98Ez!
zVP@CcE1wzbyEd!ac)Tvvm9FaF>zg!Da?QmVBT_6yT-=I$YA3Gi?#$}(lZmCdCwLM^
zAY?c}S3@ziQk#lh_EBco*Xd~qL<a&zgBynkbZdPg5|WZc>fC4j2FvsZc=<LqHo5%L
zN{>Oq$9Y`0xeH($Foou$nLU27Bo$@fIO|+_u@|<kBTV6u;XE%SspRB?Bw%Go9ich*
zY`9E5R!brF5_grEh6L0K$FwsQ+?kSsg5HBw!B&dlFPC8!(221*gpFNhB#subYmbQS
zixqdM7KP#c--$gWZ|+R>(_-Zz*CLWAOyoy)(uxYweN~Bdsd)Egoo4St^%b*uAF_01
zP|WUUT@*YC{Ev{7SKfCnQInB*hj~DcfKDJNLmn0!^3Cov8yg$a+Z^n2fMM`1!H|!d
zsa-w=*<r_-%>4|-`1Y#na)k89#~+H~B+pX~I2#LH_fYhd8pqc0J_tpxY=>kd>P?0Y
zhBOfpYrAPOduSlwq<@#3Ov+0+sN8m<A-!5&SmAh?n0)n}>aVNO(}oY#@WyOEeKd{q
zVr#6u1VkMl%|8A2qNM~+QZhhHvI-t9Y<j<<9>>lnP&uF<X;88{BBD)O8ciHL9_V5%
zNT=*;KOa#qOKvFU5@Z|0=QLR$;I%p)bavAl8v9!s+|IAVN&Y`o?&VOScZLoadM$tI
zS=_u>zhh_<mR;NHQA&wQU()up>Yc-36e?$?JN34$1<O9-R1T_H8WYb9L-WrCsuy5p
ze1_#QRxRA*o#_hew#zTn#5xqkTvmEja)E67ec#oYU!@Yk6_BR<TfFn}zb`M#3JZ&g
ziHTnN%W|KcS}UETqT)+eZcBgUDK4|-DdPuWjZjYMtBiVALOB(WKYP0R->2q;af?5C
zR!7~QJl6kS*g^|;OT&Z;_e$qo+cnn<UD@X%(iDbQ!D(XPITIw`_H~Bs;hVHGTk}lK
zR8@R2)`sSsmZe6c*&X?@w0~|Z0Tt3}n>0?|{xBXF`)5&Ulb{4s)Q+sez}XCCOLe6P
zPaY0c;!Y7=V(t)<$~oVzPzsyCkLlGhDfiEjxhe(GZZ(^KY~dQ)A?79Em|kTT4r=8U
zqLbLwp(iVLuqB$}a)s<|+Y6Hgfj0-%W+%=hQz<0ZIF2L8;Jzs6@`V5M<-Cy(MWVkn
zIm@12ZT<F(;p`obkm{)Q9ao-K9Jn8yFspPnT~R~JtIDkCg+)a9UTA?`Oo@*<ASm`#
zuS`$u#nnkl#feA?UhkIN{@Lzd&;fkfc={spf_`bKZGNk;I__x2;_mUb(Yo;VL$B^)
zD`03rz#cV|5Yeu5UD4S#w&2Q_CAP~C+Su3v<Z4GV8Xvw=idf>zmy0d0YfJl;ogo>C
zhi~tRD~+%C(Qnk-C>nRG)OLd5++-*vqh74-SbF38S2Neo$3J(hJ<91fMSQ1iUNw>~
zlZum}nY*rvyG@K9pud<IaY9Cm5o`fC&2eudnCjaIy$t{pu!i^U1&4RcJZBS(S&*BL
zkGEZhRG%P1kltZn2AFV!OEuUFunGDB8!1W21Opv-A7?^i2no$Y-u+%=pmN}^Badj(
z!J3ki5|NuA%Yev}qDRZ&i?L8H2osR{5Y9<*L09oRV2zi)E|>x(2XM6l$yP3wr^8II
zAQC!EZPBTA`130Zo|IswQ&X3iZY50Qdk_gLU{JYtYmTgBc^Rq*Z1jbWiQ!=nLA{F?
zVHjEvi=bCpR#sM8ngDGMTHM;yUW5v#sR>}1gR`*gs5=R&&N^BtHwD5(a{?2pPW&7h
zIR#QRru4wD7JY-=oF=yj2QXm2AXlKRc@;iQnWYdi;=H<A2|VeQS-VdcwQ?ayPK1Ax
z69&jaMqO>i3)JfhQ02MLh$L{nF$Pbc2bT760%jhh+^;zBxAU@+eMKrx0Y=ExBySVD
z>Fl?Ucy+@gc-<_IVDcG0^UikJ4(=3%O%IN6b6-`^(TRG$$x<G~Z?M2+4MFHjuBedY
zfopSod_2JPi?1w8;lta%1&m>mz|Bp4X!7UckoHT)ag9lCM8ak~#{;s9-y*(V3_R!|
z^Q=bvJ%Na;u}ufS(m}K9H*Tc(uzKy)NENvaI1w__&5C1iO+l~bZQjn+lUDj;516=z
zx9GCU1yOg3fV_E79y4T=^z~QZR)C1rQ*`|yKx!|T)_FHH2p-+^861QAa2Mj;v17-u
z_X(DR__GU!(H&q^;F(gKB1^!3aF+eTMHP+%T`f411BDPAT3$XmDnoW8O~K}ETN`u%
z4;q3mm71$*XvChpzMJ@bc0x|Wg28?`HJ^yp$N?N%s|?uAr&qvw^r$MNm42kK6}1S=
z0So9Jgv4?c*c@)Lv5)o^plna;P~8I>+_!i=XFGs@Imr&cCXZL!*wR8twVN|-;N>81
zrL(ZGaApgHy@e24Z(%vPx%kV1DZu?hD%05o6G((JK{X<>G1&~1B3KX(8Kni(FXRpz
z21|&ckD?Qz6_9(v;xpg0!sSbs^bV8n?0zsjgAuX%huIFD4njUS4i-41G8@YyAt-~8
zP~B`x+x0V>Ed*&EJIB1eQ3W0j<PlVK`mNBxqiKVRg|JeM^#5Jpv`x*<PDWR^80A`S
zI137rn=!nUUTQdoccL#qFi@w*ub#{dzQ^f;$y{h&K$}_wNFTR+9jILQ%jCiK_UBl=
z$hBcbAB#1Et%TYE`?nuyET-j6e~1;fuEbQK7l0`hNc7@%e2e$H3z1r?Ks)feVK3+w
zH9z(Vd6NnaY7D7=>t%R()j(8HAf$7KY@gDa2)z@wIe62^KIafZSqBT{{q7T&g3it9
z)5L58*pGxBv03XtKMdgO!7;Em7)T^`PP~GuUh>C}A0o`LCyc6)-4x6{>4sX_@<#WI
zf~>15)hY0UD5va$;EZmnP!cO}C)Ivy-&;F&8cjdjC5!QU0b2<lxjAQ;`aVe~QW}hV
zeFRC0NS!oXg)FvK8-4V9jHQ@-IA>c!EclY(uuJ9Ve*tWX<LDbK#c>*9eHArFK|F(1
zY8359<S$_28!F9Uj23=(0;6C8E}L3f^5t-O?LL_QMHh3q_u6(Mak5jYp{uQ}HOiU5
zpg(9N6$0sPFgf=XS-l3)>XUFd@hS(YTgo`f6#RJG!BcEZFxRHZ{jz|{$i~a72`d!V
zCCodObA6|<h377Is3<Gr5Fn?PKO9+pvo*z|i^?nXp9g-qY=<-kgVB}o%wisp7BU`_
z19jqs{uPfP8oY6198Mnv8^(_U!A%pk><;CH8!a<4MJv0fF$)(E-A>naq?`wHH=iP7
zz<19G52uzrWczT9m@?DBvG=OFjDhULTb&2p2f(gmVtOyW(zq|!41(#1@ZFMcn0@8C
zQyjVu$Y44y#$<i2e)I<1Hg~5M*5mU-qn=B$lKBf1Ps-fMzWU<OW938;ng97_Q-|Lf
z*FpdH#F-E$^h{0=OiyAW^rhr4V(bBGsZmin{^oELJHozYp1Vx929ucK7WW(Z92v{@
zTf5+gM~{>0DhQgeykQ@>9gq|D`?z<-^6LI!%E7w29mib$#xe>mg4>CjMCKJHPGv*+
z()@&MKUVrR3*^;m*jwbicM6`Hs&S+vFxG6s8@hLp`Ofs0?mr2Q4O#dl2yz}iR6$mX
z*)w{Katd^tK`4irVj!7c9X1Y9O!`q@k#3l*!Hf;l{(hqfym6Vq)6<h+*<cbvM{{Fi
zXa1F>fu_9{@TpTnr!}}Al@6h=;=hD3oFI;$f*62(Z<hMO+4*^oD8#h%%uGKw984{F
zjOcXIZ4?O+I^??UHB|3C12>={xhib^qyoc!bW+mNYOJJ8K*OrC%fx)Q*Px8RJ|f4I
zS5a~lRNJa!Oz(;cPFkqFho)-8EdU0vGA_siWDn|9INj4{R-~k*^-g^NU6sw)os{(S
zT4rY*|Fox1t$K>aunsRhW+h`Ie2%tKG7q2B<z~t(1YaNjE|tcEys-#s$fe})hL)CT
z%p`Er;369?L)uXDVA$O24&sXtdhu{Bf<~mjzaKm5P&2^oAbNgd>>_;kn`jxHYT@PP
zw(NTqbZ-vMTm9Hn;1Q~yx;x$i?;W(cV2P58h91Hh&Rl2>u*i-%M7Ur`p5xL7PbgNt
zj>Qm0VC9adq&m@lTVZjB8Q`(w$MZ8YZyFn;=)hq6aKf5|BtRyJ#)bS={Epc7e)~YQ
zsxd^|vxD?(E#!;agQaCQ57ul9XJ(_I(d7s{eB?+e?hLVOtooM`L*T=PIe$klzHlD}
z&;q+LXehhj?n;<lH%)fs4*?Lt5eg3wMubLu7l6%^Sl2jkkrFh+S2R9W=+<tN1lY|k
z6$rSezxlfu2UDGKwpOTK(0Mf)L!UHB!m%hUjH>)$le_yS0k}NW6KY*xtT;-Cx1K4a
zyoUOfaKenjP%oU35FMy%Xc#A?%{CG_-2n_y)sU4nM%=u)E8*K4ei=$_(99`!2IN3p
zon$ZEj9Mvl<HUZ#+yGyv=nte3$1lH*e2=4&<g+k02eqPS4rt-}6#)TFU_E}7Xd(L3
z9@l<LP;#Rs|DhMz)w?Vtbza?YtOx_$zL(V^W%FPK0;}4Q90X@y8@3T3s`{`RJX=11
z83qoidT;Zg{Eb29N);JX%#~dp1Si)^L~Zz`0{VI}7tWt2dd`+2Y4E8<T7YfBvVaa0
zr8d$Cs#ePBFVZz9OM)atqwFQ`U`t{lUJpOBEDl?%oNqhltl&HlBXba{1`9-|L5fLA
z7JtdY3K*%8<*S{fX8U>Y$krQ7C$BXC=i=YMG1=_|jfN!^<}b6<_%mdNNm@T(^)vrx
zvDbFOL;rk?H69u#Cnu;2nlRiy+(9-|=zB>~LE+WSR%5vS-n?pQN%Rb^U*9f@A)!=%
zuHD-7xkb|isxk}@eLuAV>mEX63}izZhY%uf$__T#{3@}+8}4d;#jeY<MrP*a7>Yg~
z5fMSf0N}8j{p!`LZ#4-|L;b^eY<r2v_8IcNYh<TO0+wzjA~<opHrmmPuipnqSpPog
z5Co-voB^rc=FOYav#2JY%fWLrmqite1R^d^LL0rVNU>S25r7S(TC?XfZ(O?uGkz{k
zPFde_=4_?M=y}{&IE}MT`92WVa;kS^|2-P_+#Lt<currw{1++OZDjy%=w6Xo5ZwLg
zfXn9QJmD<;CQvd^Q9Xzu(}mn>8BpczQNXOBUnzgiR)Kc&T+l?)VON=@C|p#?E?l{i
z6vX;&pOiZ?1z-(5F|;@a(9s6A&1Veo&N-B`XZ!hb72I-rw9%0#boVwllQ(%S%+l#S
z!?XxM(|~6X!7sT~A#{vH6zp8i%FYH!gl8iGm+eHXsN1+bNZXreHtl=o*x)w3A_Z#D
zNg*LA*EJD~m5El47jnVZOcYN)z!8p4aNDO(?oq%-FbEoWa(&qVVvVe#qVKMKk8t2B
z270!2YcPi%-onC26|q?y4}5qlYM*#%x5P%;?*JkT(Ci)jg`-d@D3cgL$LH5sl<`8^
zZT;YJnH}Dbz@S*@(*E?-4$sOmk<~Ao5CGuF|9{Mc|NsB>yZUb*H+I$ytzf3g#T}dQ
OPg(xFT%PPzkN*Wni%g#Y

literal 0
HcmV?d00001

diff --git a/moses/latentgan/README.md b/moses/latentgan/README.md
new file mode 100644
index 0000000..ccc021d
--- /dev/null
+++ b/moses/latentgan/README.md
@@ -0,0 +1,26 @@
+LatentGAN
+=========
+<p align="center">
+  <img src="../../images/LatentGAN.png">
+</p>
+
+LatentGAN [1] with heteroencoder trained on ChEMBL 25 [2], which encodes SMILES strings into latent vector representations of size 512. A Wasserstein Generative Adversarial network with Gradient Penalty [3] is then trained to generate latent vectors resembling that of the training set, which are then decoded using the heteroencoder. This model uses the Deep-Drug-Coder heteroencoder implementation [4].
+
+
+Important!
+==========
+Currently, the Deep-Drug-Coder [4] and its dependency package molvecgen [5] are not available in pypi, these have to be installed from there respective repositories (links provided below).
+
+The pretrained models of the LatentGAN are currently not shared in this repository due to file size constraints. These will be added in the near future.
+
+## References
+
+[1] [A De Novo Molecular Generation Method Using Latent Vector Based Generative Adversarial Network](https://chemrxiv.org/articles/A_De_Novo_Molecular_Generation_Method_Using_Latent_Vector_Based_Generative_Adversarial_Network/8299544)
+
+[2] [ChEMBL](https://www.ebi.ac.uk/chembl/)
+
+[3] [Improved training of Wasserstein GANs](https://arxiv.org/abs/1704.00028)
+
+[4] [Deep-Drug-Coder](https://github.com/pcko1/Deep-Drug-Coder)
+
+[5] [molvecgen](https://github.com/EBjerrum/molvecgen)
diff --git a/moses/latentgan/__init__.py b/moses/latentgan/__init__.py
new file mode 100644
index 0000000..bc77b73
--- /dev/null
+++ b/moses/latentgan/__init__.py
@@ -0,0 +1,5 @@
+from .config import get_parser as latentGAN_parser
+from .model import LatentGAN
+from .trainer import LatentGANTrainer
+
+__all__ = ['latentGAN_parser', 'LatentGAN', 'LatentGANTrainer']
diff --git a/moses/latentgan/config.py b/moses/latentgan/config.py
new file mode 100644
index 0000000..47d4e37
--- /dev/null
+++ b/moses/latentgan/config.py
@@ -0,0 +1,92 @@
+import argparse
+
+
+def get_parser(parser=None):
+    if parser is None:
+        parser = argparse.ArgumentParser()
+
+    # Model
+    model_arg = parser.add_argument_group('Model')
+    model_arg.add_argument("--heteroencoder_version", type=str, default='new',
+                           help="Which heteroencoder model version to use")
+    # Train
+    train_arg = parser.add_argument_group('Training')
+
+    train_arg.add_argument('--gp', type=int, default=10,
+                           help='Gradient Penalty Coefficient')
+    train_arg.add_argument('--n_critic', type=int, default=5,
+                           help='Ratio of discriminator to'
+                                ' generator training frequency')
+    train_arg.add_argument('--train_epochs', type=int, default=2000,
+                           help='Number of epochs for model training')
+    train_arg.add_argument('--n_batch', type=int, default=64,
+                           help='Size of batch')
+    train_arg.add_argument('--lr', type=float, default=0.0002,
+                           help='Learning rate')
+    train_arg.add_argument('--b1', type=float, default=0.5,
+                           help='Adam optimizer parameter beta 1')
+    train_arg.add_argument('--b2', type=float, default=0.999,
+                           help='Adam optimizer parameter beta 2')
+    train_arg.add_argument('--step_size', type=int, default=10,
+                           help='Period of learning rate decay')
+    train_arg.add_argument('--latent_vector_dim', type=int, default=512,
+                           help='Size of latentgan vector')
+    train_arg.add_argument('--gamma', type=float, default=1,
+                           help='Multiplicative factor of'
+                                ' learning rate decay')
+    train_arg.add_argument('--n_jobs', type=int, default=1,
+                           help='Number of threads')
+    train_arg.add_argument('--n_workers', type=int, default=1,
+                           help='Number of workers for DataLoaders')
+
+    # Arguments used if training a new heteroencoder
+    heteroencoder_arg = parser.add_argument_group('heteroencoder')
+
+    heteroencoder_arg.add_argument('--heteroencoder_layer_dim', type=int,
+                                   default=512,
+                                   help='Layer size for heteroencoder '
+                                        '(if training new heteroencoder)')
+    heteroencoder_arg.add_argument('--heteroencoder_noise_std', type=float,
+                                   default=0.1,
+                                   help='Noise amplitude for heteroencoder')
+    heteroencoder_arg.add_argument('--heteroencoder_dec_layers', type=int,
+                                   default=4,
+                                   help='Number of decoding layers'
+                                        ' for heteroencoder')
+    heteroencoder_arg.add_argument('--heteroencoder_batch_size',
+                                   type=int, default=128,
+                                   help='Batch size for heteroencoder')
+    heteroencoder_arg.add_argument('--heteroencoder_epochs', type=int,
+                                   default=100,
+                                   help='Number of epochs for heteroencoder')
+    heteroencoder_arg.add_argument('--heteroencoder_lr', type=float,
+                                   default=1e-3,
+                                   help='learning rate for heteroencoder')
+    heteroencoder_arg.add_argument('--heteroencoder_mini_epochs', type=int,
+                                   default=10,
+                                   help='How many sub-epochs to '
+                                        'split each epoch for heteroencoder')
+    heteroencoder_arg.add_argument('--heteroencoder_lr_decay',
+                                   default=True, action='store_false',
+                                   help='Use learning rate decay '
+                                        'for heteroencoder ')
+    heteroencoder_arg.add_argument('--heteroencoder_patience', type=int,
+                                   default=100,
+                                   help='Patience for adaptive learning '
+                                        'rate for heteroencoder')
+    heteroencoder_arg.add_argument('--heteroencoder_lr_decay_start', type=int,
+                                   default=500,
+                                   help='Which sub-epoch to start decaying '
+                                        'learning rate for heteroencoder ')
+    heteroencoder_arg.add_argument('--heteroencoder_save_period', type=int,
+                                   default=100,
+                                   help='How often in sub-epochs to '
+                                        'save model checkpoints for'
+                                        ' heteroencoder')
+
+    return parser
+
+
+def get_config():
+    parser = get_parser()
+    return parser.parse_known_args()[0]
diff --git a/moses/latentgan/model.py b/moses/latentgan/model.py
new file mode 100644
index 0000000..d4ae374
--- /dev/null
+++ b/moses/latentgan/model.py
@@ -0,0 +1,226 @@
+import torch.nn as nn
+import numpy as np
+import torch
+from ddc_pub import ddc_v3 as ddc
+import os
+from rdkit import Chem
+import sys
+from torch.utils import data
+import torch.autograd as autograd
+
+
+class LatentGAN(nn.Module):
+    def __init__(self, vocabulary, config):
+        super(LatentGAN, self).__init__()
+        self.vocabulary = vocabulary
+        self.generator = Generator()
+        self.model_version = config.heteroencoder_version
+        self.discriminator = Discriminator()
+        self.sample_decoder = None
+        self.model_loaded = False
+        self.new_batch_size = 256
+        # init params
+        cuda = True if torch.cuda.is_available() else False
+        if cuda:
+            self.discriminator.cuda()
+            self.generator.cuda()
+        self.Tensor = torch.cuda.FloatTensor if cuda else torch.FloatTensor
+
+    def forward(self, n_batch):
+        out = self.sample(n_batch)
+        return out
+
+    def encode_smiles(self, smiles_in, encoder=None):
+
+        model = load_model(model_version=encoder)
+
+        # MUST convert SMILES to binary mols for the model to accept them
+        # (it re-converts them to SMILES internally)
+        mols_in = [Chem.rdchem.Mol.ToBinary(Chem.MolFromSmiles(smiles))
+                   for smiles in smiles_in]
+        latent = model.transform(model.vectorize(mols_in))
+
+        return latent.tolist()
+
+    def compute_gradient_penalty(self, real_samples,
+                                 fake_samples, discriminator):
+        """Calculates the gradient penalty loss for WGAN GP"""
+        # Random weight term for interpolation between real and fake samples
+        alpha = self.Tensor(np.random.random((real_samples.size(0), 1)))
+
+        # Get random interpolation between real and fake samples
+        interpolates = (alpha * real_samples +
+                        ((1 - alpha) * fake_samples)).requires_grad_(True)
+        d_interpolates = discriminator(interpolates)
+        fake = self.Tensor(real_samples.shape[0], 1).fill_(1.0)
+
+        # Get gradient w.r.t. interpolates
+        gradients = autograd.grad(
+            outputs=d_interpolates,
+            inputs=interpolates,
+            grad_outputs=fake,
+            create_graph=True,
+            retain_graph=True,
+            only_inputs=True,
+        )[0]
+        gradients = gradients.view(gradients.size(0), -1)
+        gradient_penalty = ((gradients.norm(2, dim=1) - 1) ** 2).mean()
+
+        return gradient_penalty
+
+    @property
+    def device(self):
+        return next(self.parameters()).device
+
+    def sample(self, n_batch, max_length=100):
+        if not self.model_loaded:
+            # Checking for first batch of model to only load model once
+            print('Heteroencoder for Sampling Loaded')
+            self.sample_decoder = load_model(model_version=self.model_version)
+            # load generator
+
+            self.Gen = self.generator
+            self.Gen.eval()
+
+            self.D = self.discriminator
+            torch.no_grad()
+            cuda = True if torch.cuda.is_available() else False
+            if cuda:
+                self.Gen.cuda()
+                self.D.cuda()
+            self.S = Sampler(generator=self.Gen)
+            self.model_loaded = True
+
+            if n_batch <= 256:
+                print('Batch size of {} detected. Decoding '
+                      'performs poorly when Batch size != 256. \
+                 Setting batch size to 256'.format(n_batch))
+        # Sampling performs very poorly on default sampling batch parameters.
+        #  This takes care of the default scenario.
+        if n_batch == 32:
+            n_batch = 256
+
+        latent = self.S.sample(n_batch)
+        sanitycheck = self.D(latent)
+        print('mean latent values')
+        print(torch.mean(latent))
+        print('var latent values')
+        print(torch.var(latent))
+        print('generator loss of sample')
+        print(-torch.mean(sanitycheck))
+        latent = latent.detach().cpu().numpy()
+
+        if self.new_batch_size != n_batch:
+            # The batch decoder creates a new instance of the decoder
+            # every time a new batch size is given, e.g. for the
+            # final batch of the generation.
+            self.new_batch_size = n_batch
+            self.sample_decoder.batch_input_length = self.new_batch_size
+        lat = latent
+
+        sys.stdout.flush()
+
+        smi, _ = self.sample_decoder.predict_batch(lat, temp=0)
+        return smi
+
+
+def load_model(model_version=None):
+    # Import model
+    currentDirectory = os.getcwd()
+
+    if model_version == 'chembl':
+        model_name = 'chembl_pretrained'
+    elif model_version == 'moses':
+        model_name = 'moses_pretrained'
+    elif model_version == 'new':
+        model_name = 'new_model'
+    else:
+        print('No predefined model of that name found. '
+              'using the default pre-trained MOSES heteroencoder')
+        model_name = 'moses_pretrained'
+
+    path = '{}/moses/latentgan/heteroencoder_models/{}' \
+        .format(currentDirectory, model_name)
+    print("Loading heteroencoder model titled {}".format(model_version))
+    print("Path to model file: {}".format(path))
+    model = ddc.DDC(model_name=path)
+    sys.stdout.flush()
+
+    return model
+
+
+class LatentMolsDataset(data.Dataset):
+    def __init__(self, latent_space_mols):
+        self.data = latent_space_mols
+
+    def __len__(self):
+        return len(self.data)
+
+    def __getitem__(self, index):
+        return self.data[index]
+
+
+class Discriminator(nn.Module):
+    def __init__(self, data_shape=(1, 512)):
+        super(Discriminator, self).__init__()
+        self.data_shape = data_shape
+
+        self.model = nn.Sequential(
+            nn.Linear(int(np.prod(self.data_shape)), 512),
+            nn.LeakyReLU(0.2, inplace=True),
+            nn.Linear(512, 256),
+            nn.LeakyReLU(0.2, inplace=True),
+            nn.Linear(256, 1),
+        )
+
+    def forward(self, mol):
+        validity = self.model(mol)
+        return validity
+
+
+class Generator(nn.Module):
+    def __init__(self, data_shape=(1, 512), latent_dim=None):
+        super(Generator, self).__init__()
+        self.data_shape = data_shape
+
+        # latent dim of the generator is one of the hyperparams.
+        # by default it is set to the prod of data_shapes
+        self.latent_dim = int(np.prod(self.data_shape)) \
+            if latent_dim is None else latent_dim
+
+        def block(in_feat, out_feat, normalize=True):
+            layers = [nn.Linear(in_feat, out_feat)]
+            if normalize:
+                layers.append(nn.BatchNorm1d(out_feat, 0.8))
+            layers.append(nn.LeakyReLU(0.2, inplace=True))
+            return layers
+
+        self.model = nn.Sequential(
+            *block(self.latent_dim, 128, normalize=False),
+            *block(128, 256),
+            *block(256, 512),
+            *block(512, 1024),
+            nn.Linear(1024, int(np.prod(self.data_shape))),
+            # nn.Tanh() # expecting latent vectors to be not normalized
+        )
+
+    def forward(self, z):
+        out = self.model(z)
+        return out
+
+
+class Sampler(object):
+    """
+    Sampling the mols the generator.
+    All scripts should use this class for sampling.
+    """
+
+    def __init__(self, generator: Generator):
+        self.G = generator
+
+    def sample(self, n):
+        # Sample noise as generator input
+        z = torch.cuda.FloatTensor(np.random.uniform(-1, 1,
+                                                     (n, self.G.latent_dim)))
+        # Generate a batch of mols
+        return self.G(z)
diff --git a/moses/latentgan/trainer.py b/moses/latentgan/trainer.py
new file mode 100644
index 0000000..69a57ff
--- /dev/null
+++ b/moses/latentgan/trainer.py
@@ -0,0 +1,271 @@
+import torch
+import torch.optim as optim
+import numpy as np
+import pandas as pd
+from tqdm.auto import tqdm
+from collections import Counter
+import os
+import sys
+from moses.interfaces import MosesTrainer
+from moses.utils import CharVocab, Logger
+from .model import LatentMolsDataset
+from .model import load_model
+from .model import Generator
+from .model import Discriminator
+from .model import Sampler
+from rdkit import Chem
+from ddc_pub import ddc_v3 as ddc
+
+
+class LatentGANTrainer(MosesTrainer):
+
+    def __init__(self, config):
+        self.config = config
+        self.latent_size = self.config.latent_vector_dim
+        self.generator = Generator(data_shape=(1, self.latent_size))
+        self.discriminator = Discriminator(data_shape=(1, self.latent_size))
+        cuda = True if torch.cuda.is_available() else False
+        if cuda:
+            self.discriminator.cuda()
+            self.generator.cuda()
+
+    def _train_epoch(self, model, tqdm_data,
+                     optimizer_disc=None, optimizer_gen=None):
+        if optimizer_disc is None:
+            model.eval()
+            optimizer_gen = None
+        else:
+            model.train()
+        self.Sampler = Sampler(generator=self.generator)
+
+        postfix = {'generator_loss': 0,
+                   'discriminator_loss': 0}
+        disc_loss_batch = []
+        g_loss_batch = []
+
+        for i, real_mols in enumerate(tqdm_data):
+
+            real_mols = real_mols.type(model.Tensor)
+            if optimizer_disc is not None:
+                optimizer_disc.zero_grad()
+            fake_mols = self.Sampler.sample(real_mols.shape[0])
+
+            real_validity = self.discriminator(real_mols)
+            fake_validity = self.discriminator(fake_mols)
+            # Gradient penalty
+            gradient_penalty = model.compute_gradient_penalty(
+                real_mols.data, fake_mols.data, self.discriminator)
+
+            d_loss = -torch.mean(real_validity) \
+                + torch.mean(fake_validity) \
+                + self.config.gp * gradient_penalty
+
+            disc_loss_batch.append(d_loss.item())
+
+            if optimizer_disc is not None:
+
+                d_loss.backward()
+                optimizer_disc.step()
+
+                # Train the generator every n_critic steps
+                if i % self.config.n_critic == 0:
+                    # -----------------
+                    #  Train Generator
+                    # -----------------
+                    optimizer_gen.zero_grad()
+                    # Generate a batch of mols
+                    fake_mols = self.Sampler.sample(real_mols.shape[0])
+
+                    # Loss measures generator's ability to
+                    # fool the discriminator
+                    # Train on fake images
+                    fake_validity = self.discriminator(fake_mols)
+                    g_loss = -torch.mean(fake_validity)
+
+                    g_loss.backward()
+                    optimizer_gen.step()
+
+                    g_loss_batch.append(g_loss.item())
+                    postfix['generator_loss'] = np.mean(g_loss_batch)
+
+            postfix['discriminator_loss'] = np.mean(disc_loss_batch)
+            tqdm_data.set_postfix(postfix)
+        postfix['mode'] = 'Eval' if optimizer_disc is None else 'Train'
+        return postfix
+
+    def _train(self, model, train_loader, val_loader=None, logger=None):
+
+        device = model.device
+        optimizer_disc = optim.Adam(self.discriminator.parameters(),
+                                    lr=self.config.lr,
+                                    betas=(self.config.b1, self.config.b2))
+        optimizer_gen = optim.Adam(self.generator.parameters(),
+                                   lr=self.config.lr,
+                                   betas=(self.config.b1, self.config.b2))
+        scheduler_disc = optim.lr_scheduler.StepLR(optimizer_disc,
+                                                   self.config.step_size,
+                                                   self.config.gamma)
+        scheduler_gen = optim.lr_scheduler.StepLR(optimizer_gen,
+                                                  self.config.step_size,
+                                                  self.config.gamma)
+        sys.stdout.flush()
+
+        for epoch in range(self.config.train_epochs):
+            scheduler_disc.step()
+            scheduler_gen.step()
+
+            tqdm_data = tqdm(train_loader,
+                             desc='Training (epoch #{})'.format(epoch))
+
+            postfix = self._train_epoch(model, tqdm_data,
+                                        optimizer_disc, optimizer_gen)
+            if logger is not None:
+                logger.append(postfix)
+                logger.save(self.config.log_file)
+
+            if val_loader is not None:
+                tqdm_data = tqdm(val_loader,
+                                 desc='Validation (epoch #{})'.format(epoch))
+                postfix = self._train_epoch(model, tqdm_data)
+                if logger is not None:
+                    logger.append(postfix)
+                    logger.save(self.config.log_file)
+
+            # Putting the generator and discriminator weights onto
+            # the model instance for checkpoint purposes
+            # The parameters of the model are
+            # the weights used during the sampling stage later.
+
+            gen_params = self.generator.named_parameters()
+            disc_params = self.discriminator.named_parameters()
+            model_params = dict(model.named_parameters())
+            for name, param in gen_params:
+                iter_name = 'generator.{}'.format(name)
+                if iter_name in model_params:
+                    model_params[iter_name].data.copy_(param.data)
+            for name, param in disc_params:
+                iter_name = 'discriminator.{}'.format(name)
+                if iter_name in model_params:
+                    model_params[iter_name].data.copy_(param.data)
+            model.load_state_dict(model_params, strict=False)
+            sys.stdout.flush()
+            if (self.config.model_save is not None) and \
+                    (epoch % self.config.save_frequency == 0):
+                model = model.to('cpu')
+                torch.save(
+                    model.state_dict(),
+                    self.config.model_save[:-3] + '_{0:03d}.pt'.format(epoch)
+                )
+                model = model.to(device)
+
+    def get_vocabulary(self, data):
+        return CharVocab.from_data(data)
+
+    def get_collate_fn(self, model):
+        device = self.get_collate_device(model)
+
+        def collate(data):
+            tensors = torch.tensor([t for t in data],
+                                   dtype=torch.float64, device=device)
+            return tensors
+
+        return collate
+
+    def _get_dataset_info(self, data, name=None):
+        df = pd.DataFrame(data)
+        maxlen = df.iloc[:, 0].map(len).max()
+        ctr = Counter(''.join(df.unstack().values))
+        charset = ''
+        for c in list(ctr):
+            charset += c
+        return {"maxlen": maxlen, "charset": charset, "name": name}
+
+    def fit(self,
+            model,
+            train_data,
+            val_data=None):
+
+        logger = Logger() if self.config.log_file is not None else None
+
+        if self.config.heteroencoder_version == 'new':
+            # Train the heteroencoder first
+            print("Training heteroencoder.")
+            currentDirectory = os.getcwd()
+            path = '{}/moses/latentgan/heteroencoder_models/new_model' \
+                .format(currentDirectory)
+            encoder_checkpoint_path = \
+                '{}/moses/latentgan/heteroencoder_models/checkpoints/' \
+                .format(currentDirectory)
+            # Convert all SMILES to binary RDKit mols to be
+            #  compatible with the heteroencoder
+            heteroencoder_mols = [Chem.rdchem.Mol
+                                      .ToBinary(Chem.MolFromSmiles(smiles))
+                                  for smiles in train_data]
+            # Dataset information
+            dataset_info = self._get_dataset_info(
+                train_data, name="heteroencoder_train_data")
+            # Initialize heteroencoder with default parameters
+            heteroencoder_model = ddc.DDC(x=np.array(heteroencoder_mols),
+                                          y=np.array(heteroencoder_mols),
+                                          dataset_info=dataset_info,
+                                          scaling=False,
+                                          noise_std=self.config.
+                                          heteroencoder_noise_std,
+                                          lstm_dim=self.config.
+                                          heteroencoder_layer_dim,
+                                          dec_layers=self.config.
+                                          heteroencoder_dec_layers,
+                                          td_dense_dim=0,
+                                          batch_size=self.config.
+                                          heteroencoder_batch_size,
+                                          codelayer_dim=self.latent_size)
+            # Train heteroencoder
+            heteroencoder_model.fit(epochs=self.config.heteroencoder_epochs,
+                                    lr=self.config.heteroencoder_lr,
+                                    model_name="new_model",
+                                    mini_epochs=self.config.
+                                    heteroencoder_mini_epochs,
+                                    patience=self.config.
+                                    heteroencoder_patience,
+                                    save_period=self.config.
+                                    heteroencoder_save_period,
+                                    checkpoint_dir=encoder_checkpoint_path,
+                                    gpus=1,
+                                    use_multiprocessing=False,
+                                    workers=1,
+                                    lr_decay=self.config.
+                                    heteroencoder_lr_decay,
+                                    sch_epoch_to_start=self.config.
+                                    heteroencoder_lr_decay_start)
+
+            heteroencoder_model.save(path)
+
+        heteroencoder = load_model(
+            model_version=self.config.heteroencoder_version)
+        print("Training GAN.")
+        mols_in = [Chem.rdchem.Mol.ToBinary(
+            Chem.MolFromSmiles(smiles)) for smiles in train_data]
+        latent_train = heteroencoder.transform(
+            heteroencoder.vectorize(mols_in))
+        # Now encode the GAN training set to latent vectors
+
+        latent_train = latent_train.reshape(latent_train.shape[0],
+                                            self.latent_size)
+
+        if val_data is not None:
+            mols_val = [Chem.rdchem.Mol.ToBinary(Chem.MolFromSmiles(smiles))
+                        for smiles in val_data]
+            latent_val = heteroencoder.transform(
+                heteroencoder.vectorize(mols_val))
+            latent_val = latent_val.reshape(latent_val.shape[0],
+                                            self.latent_size)
+
+        train_loader = self.get_dataloader(model,
+                                           LatentMolsDataset(latent_train),
+                                           shuffle=True)
+        val_loader = None if val_data is None else self.get_dataloader(
+            model, LatentMolsDataset(latent_val), shuffle=False
+        )
+
+        self._train(model, train_loader, val_loader, logger)
+        return model
diff --git a/moses/models_storage.py b/moses/models_storage.py
index 0a651fa..7ff0384 100644
--- a/moses/models_storage.py
+++ b/moses/models_storage.py
@@ -2,6 +2,7 @@
 from moses.organ import ORGAN, ORGANTrainer, organ_parser
 from moses.aae import AAE, AAETrainer, aae_parser
 from moses.char_rnn import CharRNN, CharRNNTrainer, char_rnn_parser
+from moses.latentgan import LatentGAN, LatentGANTrainer, latentGAN_parser
 
 
 class ModelsStorage():
@@ -12,6 +13,8 @@ def __init__(self):
         self.add_model('char_rnn', CharRNN, CharRNNTrainer, char_rnn_parser)
         self.add_model('vae', VAE, VAETrainer, vae_parser)
         self.add_model('organ', ORGAN, ORGANTrainer, organ_parser)
+        self.add_model('latentgan', LatentGAN, LatentGANTrainer,
+                       latentGAN_parser)
 
     def add_model(self, name, class_, trainer_, parser_):
         self._models[name] = {'class': class_,
diff --git a/scripts/table_config.csv b/scripts/table_config.csv
index 02a9660..e11a32f 100644
--- a/scripts/table_config.csv
+++ b/scripts/table_config.csv
@@ -1,5 +1,6 @@
 name,path
 Train,../data/samples/train_metrics.csv
+LatentGAN,../data/samples/LatentGAN.csv
 CharRNN,../data/samples/char_rnn/metrics_char_rnn_1.csv
 CharRNN,../data/samples/char_rnn/metrics_char_rnn_2.csv
 CharRNN,../data/samples/char_rnn/metrics_char_rnn_3.csv
@@ -9,4 +10,5 @@ AAE,../data/samples/aae/metrics_aae_3.csv
 VAE,../data/samples/vae/metrics_vae_1.csv
 VAE,../data/samples/vae/metrics_vae_2.csv
 VAE,../data/samples/vae/metrics_vae_3.csv
-JTN-VAE,../data/samples/jtn_metrics.csv
\ No newline at end of file
+JTN-VAE,../data/samples/jtn_metrics.csv
+