From 7eea642c6a5edaa27b176c0bc5c6b9a7806fbf77 Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Wed, 1 Apr 2026 14:00:09 +0000
Subject: [PATCH 1/3] Initial plan


From ded2f83eb74cf6587fe427fb0353835a51cf222a Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Wed, 1 Apr 2026 14:04:27 +0000
Subject: [PATCH 2/3] Implement neural network from scratch using NumPy for
 MNIST digit classification

Agent-Logs-Url: https://github.com/Bitu-Singh-Rathoud/neural-network-numpy/sessions/2f9cae04-079e-450e-9133-bd27adca2411

Co-authored-by: Bitu-Singh-Rathoud <247644259+Bitu-Singh-Rathoud@users.noreply.github.com>
---
 __pycache__/neural_network.cpython-312.pyc    | Bin 0 -> 9698 bytes
 ...eural_network.cpython-312-pytest-9.0.2.pyc | Bin 0 -> 34626 bytes
 neural_network.py                             | 241 ++++++++++++++++
 requirements.txt                              |   1 +
 test_neural_network.py                        | 272 ++++++++++++++++++
 train.py                                      | 131 +++++++++
 6 files changed, 645 insertions(+)
 create mode 100644 __pycache__/neural_network.cpython-312.pyc
 create mode 100644 __pycache__/test_neural_network.cpython-312-pytest-9.0.2.pyc
 create mode 100644 neural_network.py
 create mode 100644 requirements.txt
 create mode 100644 test_neural_network.py
 create mode 100644 train.py

diff --git a/__pycache__/neural_network.cpython-312.pyc b/__pycache__/neural_network.cpython-312.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..881246a70595d72b6f26ecd0b2831dc358edfb88
GIT binary patch
literal 9698
zcmb7KYit`wlAalI_@+ckqF$Emv1~o)=wT_2?I?*IC4MBaqr|bZ37qv2BhE;oOp)}=
z&^CEPoU_OR)UE+h@GeAM5}4jQTofb+_-FQC4mjXmI6z!#&%!{!8GHVd`*Bd@!yb@d
zSJgAamnrY24L4_cx~sdZtLm$&*1vhZE(*#Ueg7W$d>=*q7rvOqYCJr786H+Bfx1Eo
zw7^8^D~w?Igt}r8tWaA87HU?oL2YBGbCh8Jgc2NIG5Y8$_7oFzrvJk=B_<_4%1w#N
z&A4=pi!3Cf;({1ccqI~#aWhhUfs?}$uY~8gq#TLOa#P8LcT*!S*QI145tkI1<E80{
zBJom+n~Mm77~`URN|ZP`k{0D5?mh9s`y3xuA~y&{nVX4AH67(4mmHr_7Wi9SJgFp-
z$`HrL1THMaWqBADBE=IaE($LkOcIHS0(Uc_%yA2mSY&vbtZP={g$VTI1W^vd?}+O=
z+~ic4*7r(k<eCNkKN*2)g;L7j<_(UJUPdL*DrIu&5*1|92i_B53o{Xz{Q|6km$>(1
z5v6KhBNCDcG8z+0Xm;34WFfimC9TnE8E4{tczbj%z95cD$yiL3Mv>Q}F~a)cn9lm)
zSaKnedRAk|=sA)|X;w*$CM743cEf+U6{_3RW4Cu{VsRqJ=lv!3o^|V<C(umQumM>>
z*f0(cE0hYNbL})m&6nAMBlMz_8S|C#m9V1qv?!6oydiS&8FIem&D(fcNhL&$or%VI
zMX~`QefyFF9<^p%bVw+Vmzto}hn)gY-KHLSyg7RD%+krllb`jK+}ygAlTZ(8c7R05
z4)XATJ9zd{;tmDoYs*(S21*(TX6ULhC>V7`rOh2@7?r8&i*wf%m8qi-=K*}Oru#ur
zV5ea|8clJs!cRv<&J^^DKIp8w{3WAVL1tvlaZMBxLS#V>S|wbWW&v5!EaI(%W|5N%
zL5pO^R=YeGnNdVRW8q6EE#u}mj?>v}H>Qy~p%pnRpM(n7+tmE4-j)94{=C27U%R}~
zGG1&NzkP1Y)3kJU@oet+FW<R+X4CEa)%eP><zxB20{`I9M(gpS`}n%`xPA_0i2!=I
z4ss4(!vmZHJrBS2MWDN&DZ?lR_d(Jc@Vn}3BMwsn7&i0#%#(3b8B5hTl&`v?(yArR
zQj|GDWzxpk@SPygF5nmU`qNfs8q`q)E0&N%Arb~2!&7{kpN>Q$N<;*62gWHL6Nl&G
z3Ma<GaRGWn`DrmKgHc49lhUj_L7wm}6+%K2+#slTC>-Tw8Lc8QTp=`(!QCbrsUo3B
zN%2`E>eqw)YZvvNs3fIW)w(W$V1gOrZ-^2<3!TBh>N9?a%0BD=@8OSESFXP@t21@`
zRFKtJd5%wrhFC`9vzj#=jU*(Lu%K15={v<=oL-|BG$(X|?Z>BO8BZPAp-X5BvF;(m
zx`%LbsTX?mz<>D-sBTl+?Ub{5$+PImU0iIwJ-Ov?U1?u#|BSh3U$qx(YwVg_>U_TB
zKf3Nadi%{SFnq@rkLCLFFBV%4mE3)6bkQALw+3~w1%l;bitv-u(6mB{)D>Ew06A&E
zO}D^&LIZLlin0VnZ3IQ_1VtT!2QZYr;uIY4-6c4ocBfoHujbW3)WA^w6_|q@-q|GR
z{%|-BCZ7nBZtyw7jMw7X(;#LDQ+O_-!0@C567XgW@R1un$$fNMjL9PR95KIFxx*Yv
zmkz%n0>3i$(mTP!B^_WJ$tUSeBvGCKY?Bo<>XDct&We)E$q6wWnMon;D{~^wG7ZoU
zJ(2)_G6&M2qkUOc@r=tQ5(`I@f;l)s7EjcN;5-Lr4t4=<h9P<7apuFBk=YUMqvI!z
z4sm1SCx*CV_;BPZ_y!`cfGUE^QtL{<${MePI3q}esPD0(x(~qvwPB)@hU-z$xhP<K
z3|1Zj*9RDXNf8soqap4~{aJB?f;ht`qY8IqWGq-AB8#Z`@4=SQi-H$2S`tAegk(5E
zOiHRU_~^JXHe78Z(Q*<y`?o}3m9iu^a$M%Y-qjb?@|A(#G+1es3YBfK%D`KBE(JXa
zQY!c-(1@4Nt8AB5YQEX*SNZ+Qc%-QmI6+fVH@pWOx~Z^9gaMirIiR_#1WxnTuvxQ2
z#hB(oQPyQx^Mt_5Q$h(|;t>&%M07?P1maQ<s{yF$Op7C=j|rvQYqWC2Y<~&IoPz2$
zwdHUvITxKvO^Z!AWy8_4xu-vC-Eeen?mdjZ-CNyz^Vdq<hu1C^yGOFtJI$L8&qGI7
z(b4sL$DU_+CJEs83Kc=V!904YU==!-jZhGH4P8yQ^E7q!E_y1^x6FH!7MuvJpNx*K
zyY(SoElL?1=h2y}EA!sE26$C1Dr2rgK_8QpdGyxmiU9r*qN*ujh_|qsx?cDySXBB`
z>T?EzmMWU){nZuKBG5C8NpTly)(W*dzqV2-HfT#9c|!!t1t`r&BeD^{K(KLBjLgm{
z@=IJ=1hgkAI33{uhDV5)8VE<ds?i^4^z>BFso8Xa(^!yPFoVy*0(GH7U6J~sBA<j9
z2;_B1(O5zmJhox{L`OB-3?BnA)a>{RQBAFkN<+{s-CZZ1Rg!)dzRQ0D75HrJogjGb
zEq{CNdj8P8(bdtl$zu0#$v?6~S)CKv$<095!$4m#0Pfn_f%{@HFqA#J+1~xIy}#Jr
zU$|au4`$D89qjw!#rrRQaq9l52gkoTTRQmqKTiMC>_5!@b9nvom4Atr&VE!p_)+%s
z()q>n`2+dM{PTHx(Zk`W6ZcQ7eei%^yHGm#Vr_@tdARSLu!5%@Lo`9eV6_u90CA@y
zG&s^UIBbR?0&xJ-)5w4cOq~)4pafRvi>$9JAW~KoDI(_VE`6+3eQYWVE+J4-Z8%HD
zt|DdABdDT&4ca7+f-Wv8s@)XFj3wr-h;MbfYKKsO)n`K?M<o($mG?U{R?;@ZKzLwJ
z|K4Sy-0SfLKy?uX^CmBWy9qubpsf^7&d%wms|SdNdx^tpT}*8G5Zproums9$9ZVf~
z7a$;u+^Qn-luL+oE+--W=LYpiyw>v_;=+6woB;HSbuHq<*8s{ARfwlClShbFNk@PM
z(il`4eMO@ugRJyCwh&#y*ptMrBtik4(Dmj+jU_Wkui*FNSe2<r$Wj?C8LkqG5$hOS
zO9yJz$!NZWqy8Q$a46fT7Eq#`Gs|vzyji-TN;#eTL8W|cEAC}?o-fdOIe)I`JG5Jy
zKR3C|=ILc8bZGYFj;@?qKJ~li-mSLIy#MaGW%_a3p8Vw9w{rC6-hJ6Kxyc-#yObMP
zoT_>K#ng5y%nR%D`Ez~QTiIq^r^+x5%STBhI+Zh^B?IQPp_$bKu3h9h!-7mQDr;(u
zs>8b#O?bwpl%0o)<b(FIwZLR*0sz!s5gkH`b{&pPx)}%PRT*|*?M4BOwU_%iVJ+6G
zcNb_7OEj-pwN*JIE^|f%myA2csJ4t-b*qkfk4d-kOSIc!&O^Ya?5@l=0vk7k@9ft$
zGrR=wvFmFhI_zvThw8>zGPXM(qeg=^JCdz%EKe!4=_bnDjR+4RpRO(ZEP<vfMSNPD
z^$_aEM0{ohs_TYc6J?s<xYMYxy3YwN=yb}oG=reH<wZBN24p07`<0zs6H_{kfe*$8
zhPZ)gtp9NP@8I!+3#8$~K*a&pO#xA7qRGh9%Mu;~F%3bdv<IKjC}<8zgjiG-1>H~(
z8-X5&Mqeg=q#kE_h^>(J<1lpMG#1w;OC0HWL1TnTVmUP1BnI9$bPIy>!lx@TN=!A2
zbR*)xvq9Ti?=zb1aTq3l4;6_^{jF$1T9z&>Uda0w-^tRO9^cYC_^@rGn%f}#M(5<*
zxy4xiMzN`H&0q8$&N7evU7z(8=)zdx`u!I^9WD6}?@%6RHv)Kf{$l=mVQe*-pUuBi
z3=9B(H?%;3JPq*Nnwwp2%1%B4do%Wn<Yu7bv(dE!rS2g>{$=J-pd-gT+PlBd|Ha__
z!PTqPMrUj9{(C=J{Yl}qwTs`hZR|Z+{cY3VveLTTnv<407hl`%f`x7OP`*Ic^EAAp
zS`$<Bzwl6<NF{!412C;MA7D0q3d)1<4@y*D(4RyDaSQ$-tV6HH-mlRQ`=D7!cR-fV
z5+>f|E|dX||18Qzzeber`~`TL)*zeiJy(}<2GE8OMN}XjP9_8h8Fk;Xf?reM9o7LL
zMlg`iL+q1IV}*!Uo5}O)#Ebway{8uAYCirN2Y(C|q=6ebi;VLP8iWlbwaUvYdAP#q
zhlh?84;}j^@U8#bv2PE4bMv?Riige>J?G%#kLkVv4k3%P=ps{nY^ZfaWH3p-Le0WI
zf&Pri*ml7%-J`Qtg-i_C<vPzWM_&cYo^dE;F=`|#^X2db)?kz>ugwX86rjb=8Pz$X
z&*}nHs%>!#){IABGv2E8uz*!o6Ei@hufa;1RF^TbNo`WC=wbexF{n1H?p?q6KwtrX
z>x<fK;@(ezYZ-s0Me&+_RKMyo1GG#lj4xxH*#lI^r?!;)v_T(#<(v7WwwZH4YQUqm
zreTJRNA&`)@yUB<0T2$}(~wDqc51%W99j8=ltMXJ%mh@MU_;fJ$E;voQ9V_&*#!Hp
z{Q_0}Sed@_M+MYdFM>mL2+mKff=ji0N(pY&BY5D=s|N7*b4O)lj=s|ZtG59@H}U^j
zjn9|y9;w>1Hg2|_d8OnsVdH7E{tyzLKoO9Zt98OLlOgkXqb3LSY#-*rz&!93h{Qn%
z_hkY&sGKLRp<rft8)3beh==DOgNJ3RoTbB10`g>W09LYJG7YXl7U7nRbT+C=LVALG
zyK=RLd9&J%Kx#TJLrM(FyAmV{N$Pl%4~O9%F`VK=h!Hrl8t&#Cq|q-!F8I%KGsvyp
z0H%TmNRn>ABssuAI1kCjLD)ZcgO5UL8}p82j;hcD?%u38T635KiT5#O#`hGtp_ASQ
zIg+NJO54sL&$u*m^7$X=(X@jjJDK3pmXV_~X)mXrSO}<1a6ymmsSp#621m~VxM3j+
zN&T)=dKH^{v3e7$^H6CH5f==2pxO9DLX5$MO#+z~Y$tYKvm3Ib*$o-cZ2C!RuF8pM
zc7y4feGWJgmr}aB!;-GjyEu{S!;n0eoSBJ2ip`a(dM15HhFx=`G!e3*49R?jhtvfu
zfVAKO^aofVB<zWEr85h$NptFSfEjW-yc<$e_oI7LERnKwYb{uj#(OYS{w3%l_)$Gn
z&%U+sV#`R@zUgUQdVBHh{DGpUCrdvDhiEK6Q@Fl5UkU`DQ2U)N*)xy3d+r5SgAcn$
zi`}CS=u-E1_WWiS_ZxrVd|_X)>qz#k+P6ntJ<!y>Cwu;J*FNyB4iv7Gx{mBnF7JUH
zyV-r{Hy7awA^9L(9GzSroLujNq{?eK`>r1z?LT-=T~*i4mG&Q9_T<=Ha;qb_*7u;J
z)bV1@y4l%t&%5d^%&sL%oi6}=U)w`pZ_(FVcwxgAeB|rS3-FS2ZTj})uNQsXvU@uK
z3)$|ZydC-Hik?1LYMI2%Qs?pPo4}IX*~NDrdU}hV-onI&X9Qc`Dth)m^z;`!{e{$q
zXY3En?OVRUW_wTm{la+uEyw`b3b(c?`pDkRy#s5O^`0Sk8SdTeeQxdb!v1ZR35@^I
zV)eId9|xXpzeZEOj%?G@9v;d!f$I<90dH6v-40iB!~~w0n%g*}P*(s}>95N`0Y=%9
z{@1bv(u0qxn=m~oABjP(0qhq9^SWKpjcJuRX<&@&5+OAf^v7gE#)ePBpEgK*3a>jV
zmx21q>P;}QEnm_SVl4P{ufmU!{N|-uxWP4YgA!iEO4Cqjbf`8dsM(D<>N6Qe^FtUU
zW9pG)GTJ&<US58=K;J#J=*Y6!>s!tKrPN|7A1GYhXdYO13_P7eD4*)jfXJUd)gO1B
z8EeHkYUGANGR%5yu#zVx=l?(D^EXpqgt_o%7GXVmYZ~%Db>PGd)ug|kaq87Jt8tuN
zGq@gf%WG~JFi)B=2oA<aN8q;VY6U%Y{FwpngEU0oM-OWepET=&$j2Z@HjK?gNHD_H
z+{WSK9hqj+g-*sS4cP=Sx6(y;t&dwx)Oifw<QJeKacjq(mFRM`Fm^Ydg@khJO4o8%
z{@5D5;Xj<Up#a@nyqWvU!c4JasMI{X?ieOAG!<lK|9Tb6Z%=)97)S*z-?hQdcdtTq
zIq1>g!X*ZwPbj3hLZJn?*oxw_CltD#<fFzHdnhEt!=aEg0)t5UTUU`XP)Y)RAWXsi
zd`XeH^$i}vPgr5vT*kb>C)D?hhjl#ZVcDLYb~`(<)7{Q??wsJ*z|IBQ%DzEAX&Yxd
zb`rFk9emR5Wk+;!NCdKtc!j31aLS34egsZvAb*QO@{u}*lUbAa!G#q@y+1f!*7tW(
zx(TlUW0X7!)wYGE>F*hqwmxa1=++$@L-*{s82Z4D%SxZvY4XyC|BpLkp&@0DH6i%_
D`>6hs

literal 0
HcmV?d00001

diff --git a/__pycache__/test_neural_network.cpython-312-pytest-9.0.2.pyc b/__pycache__/test_neural_network.cpython-312-pytest-9.0.2.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..0bac5f5cd1479cee2031860413a6f0fc734c9460
GIT binary patch
literal 34626
zcmeHw32+<NnI;-%0|W_@qGVFCEJ!jXLAEH7I;g|eVau{5JB(xbNL0xnAR8h<@p3mP
zi!?a)WIVfI#WMzN?}&0dV>tGfskO4hWRi`tnb{~+TU)7`q>2<MliP9Cq;^u9(eBPv
zWbH)R?D_rgb-!)^O@fXy<7|~*kw0F)tGnOv-|zKbRab{x2-)TzrGDA$a{Ucmn89mA
zZvH(Y=Us~HluMBmcb|01&3--o-Y&P~>i3=UOFSoVN@h91Q$hTC`a=Cxr>gqHr^5Z!
zr>gsFPSx~BPDLb_+x4JJ@xI_veDAvXJx<lKG(XY;BCU?4$w&)|w0f2nLRytbTfx%8
zNUIiU4J@q&X%UgOlBLxmtxlw^VrlhATOrbJ8*#@P5~}2Kg`OBlrK9Pjme!)D2i0i0
zCmHP+?tgS7I*=Sz6MgteKQpMFiKhC8`jY+0fpj9B8XRZ|g&t>hx(3fC)%~GpbX)XD
zS2}f;<wj2r58z^OK#OislYPT?N43Gz>HfsI7%TY5pr%E;2K$GGd1KKnT`H2|Xj~l}
z8j1Jef|q<Wp(gs1>7*J>;W1N*zLdspz?wNWs6LZWmFN&ECA#^IwVuRKQi~=Al<44a
z8aIfZMI|Mb9${7POLUzntFxtBO(-ecJj!d|9UU0%PpYXdyuHZk>PdE;DSfEnAtjMc
zvX&oL6RCmd_Gl9K8%AQZC)qcYRJG{fKr|6e4~|4pG@&Kaq3bjsI$}P(x`Xv_2k&6L
zzPuB3FZG@tAsAQiNYuS&lX}4D&MrQ+IK*JGaB~NK&byMXQxfJg=J5-zQy#?)=v6#`
zKAM-O{E81SpkO|pl1BnDS(j-*AH$ux%$BG*AO1b~zxe{dc~@HG^&0aU$s%-PjeJ#l
z*C;b4rA1D!xza?6%hemOT^P}H&`isSb`Ue}CAmgiF?UDI^D|!R=ZB1l>#z}#4`1}@
z?tvlQmr&Kjh#sJsof_!Y>l2!mRMT;m98W$yoaj^O-7dOyua@jPtxM-Lx=S?rtH-wY
z4E86ttHT2WNp(9-{_T7%Y%|uvHq4Kq5$8m@Uz``m4_i7dLnHbfEPZHD!z@0VjMF4Z
zYH_TRp2R?RQW@QBc7Ds!1@FKkYFhxFcfB93dFi3?hhB~s!p&LfeL3_}XgoA8H_yt=
zZ!1%+Z*@;SeNAq?iL$Pu5v5*dBl@oq`FKXO;w*Aov#4n7J+`s`o-yP<F(WQl%%k3p
zm$HExx?BNZ46eEgaea+3N`y45(HKptgBXfCi-%$$=~QY1HQ5Fr#^LJ9O@;7g$AP${
z{@J;U=l*S^u>GNH@^Q5h)rv+$tg&w(=R=Q3r4?!G%Q5$FXhpiqMx`5%PT>mg{Axst
zMuQSm8V#g2P<$nTE+@|oCA-o|MWY!RjYj$S)UomDPn}CD#!?wwT|6rGyiKU8wFB73
zqPY;>gt1t$^8DV3y{~<`u;T8l?6gGAU6c2g4~QR)u_f}a5c$vpf;v2W6r^WcB8+`8
z+vLBB+eETu1EE%-h#Cg?*=rXleCN=0dY71AjZjW4fjR>91n5<D@8CdED_<s!Y#1!M
z6F2eK==$Q(a42#Y+SOVC#Gp7Ls=?_*Vbk8f*zmp0-`RX+{dc!slOOnq!$I6xx9FbV
z6b-5{93IEv@OG%HsgKtHTqokP)EM{({VZn?OoX(=AecqA<DS}X05J$TW%6k5^M&w+
zQbZ*NV#{~7TxtF8UDxDe<pUu|$3Y++`?^Te3}eA*&MQ*4Am(OADefkeACnYm%$<=k
z?lT~2W1fsh>r2yJSm>3_nDwim%xOvSQk~Q8cfCBvR>xtrSgb^aa$?#jzSrHOSQ=wq
z#jgZ1-ju|yd}%^sEZABitX2{-KJz`z-!Fj~J131v`1g#t!N$lP98=P~xG4R~eeb<7
zX|%q1%g{(`OpD%oZ<M0DVj36}57G$i`c)0*Vp!gum_~_vVlrFA*6P+nhlfVk(oLv(
zT^wv|UtCQNsl;(7)TADaYl+iIO3`bMFs=xk`U9#usOl@?kOBG=={WFmUt)mq)R?q{
zu=Ogm8t3=lVZYdEzi8tZyp`Sdi*}7r4NyD+$MOxO=!I6m=-Ne53uW&_8(J5@^RAyp
z?s#2&<J^UFQ!A!?UwN(&**-1JM%rfNebZe4Gm*CGuGz@Gd71vM843KVz^-`wDwXF^
zJDijEvH9N-3+i$_J^(Q!9@j(hc>ka>+(+?nJWgCB&rxr~GpTC{L<vC9bg3}{_W)dU
zv9{Q2tHj|lK&7>f057;caJzkNAINUs3F!mb<J<Q^$n86ZsN46TbQ5qtuTNF!<wY=Q
z|IIs*RGH@=@x}al6?O1QE{_$lA%g1tlr?5!6=;r@gkEA=N&=x?lO6%_igj$Qzaj?2
zVRiQs@hS-!cduA=R+i$q<Tclp4LLikMT=G{zDxdh4NS#Ei9pA7dX4LZGgQ71b#LNa
zO8eO#{@u6oxBkQQ!HW_S(t~|iWJEo6zY5WEuwVDl?|@3L$cU*PG3ATCzAnhGN!_Cj
z_p3B47dgYDOQX66GHEfVa~oR&TvUyth89l`#zBooH`+#t%LL_x_Mvs{1OP|x$H$Mq
z^y%?WPac_6IfAcQnM-`RH|r6M&xTofLw@Hw?QgbEOZoP<4t?{!Z$J3;2QT0E=MP^|
zX5<am<cF8Q`Ol;0fb(%X9-IWgWy}h*x~rGwTS>tDa1a2;Jc?(`oAG44XO03V``oUK
z+rqG8{)|5-Y29ftcY2BYDG8t$KF|#xwPMi?#h<2OWucciqmuBhaedaMHQCxH&Z#7%
z#cVcbrE0H17_=mm*BFq9+ARBtRxbTS)~ZwOB^{o`Dv2iu*jjZeD_ETh<;3b#<k!87
zdIprB5?c1#*?Z5azeR5@yT8k7;B7+{-a>E$nM(cQH=jePGTv8C6H#Enb{CTVN<v09
z>zcm?x8Npj0slVyd&m6X8p0hamVirrfPld|NC&|=Y$tXBEJ5g5BNF8t#Lk$u#o#1j
zp|R*e#yaeaX&iMuie_S7Vk35{#4M`BvD4+g(T8lTMk`w34O7f$xy?*N#7No^(^|l7
zgvQt{TfxZSo(ysFo9v*b+jKAZnN|jO#;Q5r@-W@}5q5Jlz#p!S(um;eVPZWxhf{q}
z<!HLUZxB3<rZ>j5p26Whg)ukr?!Lj!L?0v6>Mp8De6P9(AQt2jC%vqlS826h7`zI9
z>Am)xc7A)F)7r+bID^x=qv&dvM#zhCI2M=F#4tSZU>t#KmK4=ohU1-z9YmYj4S+?k
ze5CHBv*TxH<mOkqCljyqB9d<f$o1x1Ck7B<d4)(bqE{o$P!QqoD$nN^0zz~2YB#?s
zSW%IUN|OnZZARy)DV~Y~`BqV!N8Xd0UsOdUGm1*tvm$lId@OriJWM$;B5F+p{N^-(
zT>x<+MkIDX7KE0*Tj~~Q3yg_iT`a^`#F#h`rd~moaAPBYF%cxkOmn$RlA(AV*o|&h
z?(j>8dE_B8viGIKTyB2|1=QmJMWUi$Gt^H}eg}b10~GTT&1|L#-E~|`CpuGoP<_Yy
z5^A@pbR2Chn{}0{+()+}jAMzBXLCLIRe5=8_tfT@m2HLaPRNr>up4jfo9epkWh}<M
zA9++wB#DtU`Cpf_*lA-&?V7J;j<1EZt$C7hGse@-MEW{n?(1F}l8pd-U|lnZ)Uu)6
zf^6M;dKmbP>0j7b($JbJF+x!4I)LJFt!LvpJdh%!6Hg7msxf*;*&r6>KS+(-3t)}d
zlZEi@TqD*#(LT8+zpJofi%lm+tl``3P#f=`I(L~fgZHzM>WBqYYDpzLq#h#h7y)`w
zHASGGz+nPJ+qjTJQ${7Q7ylZilaS-_JtC24vx$V9gEvDS-%*KIqM}{+7r`9ao22nx
zO0pU8!j#fL-Elt|ksjbSBr-7-2{+Ww40Ftts4ko_2kDR$*iO2ObxGz^xjsZKw;kVP
zECI=4YTd-tW_;QOOJs6XeJsYp3M}Nr_?q|#EbK~K`woIqM&L*=jMQH&II+c6Nn%+m
zl(iVafT2|JE3y*I_)}b|5=fJNo&{S=PON9ZT>0j&!6x~}ykLz&9UO&{1)%%8e>wjj
z|2p&Id!bp8?g4vLM?4VNljuE+0Rk7Jd&$gr(XD<K<<%zvinJ@cxprmu#VSj&qOw&k
z*b?;%RP%QT#0ive9Wqy&_($SHYEsehn^uQm5u~Cdxv5_Suxm(m*)R5JBv6>V1Xq|a
zGi@AJ#naH4vsP2g$_j(&Gt6i@y4z_T7$&8~tI>MaUIzfSs9SNqaiVeZ{``u`{ZpQ)
zlljq^)oq2komu}<S&{YmO;ei+k=CqtVbz)$c}+H*OO9ve<u&-b#xMDWfU}g!uL@Rl
zPF|BsiZUW{RbEq0HmHNxl2!rj%s?6oiU%jM;AD2RYc9;m1@c0zv*5TKg-%m}<G>%d
zE8JBqQnO8$OYyM#8h1I(HKJ}y$4EL(p--{U#ZitfAoI2_Eln*d^NI!6>#)ph2&2i?
zJ{itRLR!e#W}iXkwd85S`X!;fh9UDV_lZ{8+IM&&i%<iFE%_u?Ns@UL*+N1U`o!L@
z(-R1ZvEmbml6V5S!qW=WT%nwWY=f`+nAGi8LQ0hl*^)7W1zSsmGD|{cF=Wf+QHF3K
zixN|KJqo~NQT1`efNb{~0w)(4p~JSsLoOJ`w7VG^hU91orR@+(8G=1W?Tt`VK8~DQ
zDB*JiNJb?YvVGL92`6#(DJmb~f+G#Q`Y-`5Q6@*=GC>Nv6qtiw1Ba{PiPXLH05(nd
z9yK9SxJ<ZLO(Mksdvi99h>j%zGP8dv-@@^AQNhnrlfMKY4Bhy8{Toden%-D<Vck^y
z)MJIF9fi79j-OXeteSMcvih|n(2S2<7@1$UV|Lw+>ALCrF1ris4$RzsV7BfaloaTC
zM&3Pr1Yjo8I(=j|vU^^pzxRv;ekI_nyn8OPo?jMhaZ%u^ygR>1<eO1@_LP-vM(5bg
zcp)?}eMFQO5fmxM{)A5%_WuP&)Q<f>1Z!p&AOy`oyG8(h_JR$B+e{G~=(dR+PPJHh
zUe{)b^w`)ySiA&Y_uA4uwsc<x?A{q-4N1T0*4#Gch$vr4fM(MV{idI4+4P%st(Cym
zR@EY0Z?Sw1%wd|=4GT70Z`G(ZbFJU<8V07Nkz?T_va14rR%BO2Ny4;B&=TxE;u8oC
zpyCsVl6ZokZ6qD~Eo0&E-@eha>r0oOPj(T)VLn+RM8&=nv=4n3`mG;5Vflp~ST^+D
zVWm0~FvJ}jr(v514t&_6T467=!TvEHM4lSE$n&EgQuuF&)D2i%>eB!q#fc9|in(~y
z3Q5V(_ma%SHOWveGwKaG&8hcZuHGfYK+;Y3;Gn|Dw(f(nHle^|Mr2DZrru1$tUgC|
zK*#qoE%1&}!+OS;f>v+^EV)g6hF-$Z-V#Hg{xbsqoIoid-p2_s(-a%Z+14WE>`o`;
z?0TkJfs3D(w98XHsqUVn%0;R@i;GlC)g#$Z8-l*z*OsYj>(9#*^5o9EXL4&^o4P;W
zJJZx!sBO#o?DYGIxn0Mu4FBMCVb>Eg@)MxrGrOLcjXW_g)8Bh$3cnO^R(^s}aw=1>
z#YKUu@)Jb9jeIj|&nEJ1WSh}Bb~jOoNAOU1LQz)4MKRp!%CN{cqxS3{oEF(;ly0?)
zZeQ|;bjB%kKz&-Gw93$|d4a$m5jaeskpMCA>YxcegNRlSK>8}L@AePE9^amuA+PUA
zLsLa1yYMf9sS{eBrfMZkry|GS<M3>DR#p8j6xDXlrqsPQBRTRqh8zj?w4mAIVc}yV
zY^hrewbS!9Jmkb#II5j&b(}nI1jVgTPT=+m<t&=$f8ArqHIUkDa*gm2u(jkwcNUXt
zgslyhX6JqlHlZ8gt+ZIbh723^C;iH%Td56m(SEI}(68Z7`jtA=<hz&MuhkX$wdRw4
zUAkYvVMRLNzH<h}t`lZt79QRGR4dfRuh2E+sCNNelt#A^t0fHOmYp0{@D{eQ1m<_c
ztq-genq`d9lhi^hK#}$ttd*KYqVDcA?{%H%4DQ{AdvByl^Z>{g*I&UkcrKEObw3>a
z68)O)V>ebUm9?{u+C4C!wo{w?2v`~oZu8!y9-uav)`PQF`yA98+-IBJ7j}8tg~K;B
zl+ijv9Ze5t6eKH>cA4`^)vI6ISEz2zx))&A9vL5*Y#4uTMsCSIHy^okHgYGTQ+|M%
z$emOE*+|R0On=vm1b$Uu#aYHxmcf$|pYrpl9ir|hxB4g@_@3NSuALzhE7Q*0fni;W
zhpC`77a9h%X9Q4ZCoY9?JvNQK16QI5Zo|~RNUq=5{k?>3E8JhmhSvRoWBeWJ%c!ef
zAn+Q2KPB)60N7aSMQ(zsqZqNKYMubt3~Ywtt}S0hyl98ZIwsh3rm*8ogUQ&~JBRHQ
za4%$Tw#-tr$l6@I+7R7`P!^;bu~LP5A#FX%cfON)Gd1mfbMTw$x6gh3+!cSJ{lPi;
zp~{7>$q$v!BN}NokNN<YoJUlO1(<C;X8#^_<)km1C$0^)5cgMHf8vIZx#Y+fGylS~
z+z|UU8xQF;*I=w=y0#);LR%e&xo7D+;9)0t)m{e<^<CpVN({D6nj)c9D$=1i-CdL<
z9bAPcwRmk=@Y?X)v+>%(T4QU;NrSN%uWi^6V3+u0OfcHNO=FUiUVP0<+bqM_qRQ<%
zH1lli&SttKW`lVm>e5p&Nq46PploU(4v2OyU?(g^VMFIyBt#4)64;^^&YQq7kgyhf
zUo1Z8EE0<NAhHQDU6IrotKzH;d0uxXISMS{Yf9|#mIEL+#k2eAQXgk^-l8{P{LZr!
z`+Wchu7+=cO82PAWV|P(D9HghQ;ooiKic9b$aCh#q5Lpw-{f(wuDQEVy)Em0ziw6T
z>B;r4thwoO2e)L8EUd(Jt}E9*6W%=8`9|tO3Xxp<WNK!`=J_!FUE|mM0vir<?K~<V
zt0}T4J9%^t?WIH+5kUk0)VIV7-0<OL7t_=HOB#)L2uu?=Ou*99<mi$>5B`Oo#^>Aq
zL5;_ETp~pcZ<!V9LLm{*_blApil3!CitPKNNBpsX9-*N=#&?@Pip}Ol+8WZa5be2n
zH*40i4RLLa;=UxPQN|TFAA(jGwudpg4;v5U2ik_5RKVO*V{EEklo)Z>sE)B^CX5Z_
z!$WawiSHRyMsF{9t`>Xl@1fSsS^$vh`gLCoPD^w3ZCURh2Fi&nF<I;ta6|zSSzPkq
zq+Y7*;-7JJ?RWH`6lCf1xbqkhXLqjy`NB3V4?LvEAO^073}V{XmL{Sg_hb%Zb!_G^
zf0~#m7J7-3DhWlT*CNtRq)jNeKzgl`q)RLbaN#fp9cbQPp{yXi70L-lQVAF&8p<&E
zR9Qv_;zn7pwM3f{OTsdY4DhRBV(vv$1;4owKPttG<laS`!As;`a;oy-xQ0#=f0=RM
zbJP%_ef1In!u7G55|?m}w*wG8>F5UaO)6+%{t~2LBH(i4fi8Vk{R>pnrMUV<s{R%M
zG9joI@(<$D^tUpW8<%O#jW;CJ@=cdET)+{&y~qn_M<ZSy3lCfLOLyfgthntwJTT86
znmCkiEv#rBm$Sa?(+iP?m!2Je7CzdO-2mBVC%b1&ujp(1+U^{UIC$+W^ycKv4QFjL
zN{-df@>FuyHoUoc0*5G3jsb+Y7Q$Tpcrkzt`#4|%r>$VX8RYR%5*FDYMi{XqEV)6<
zn)}?v<rpg4Tzi8~b5;F!REUXdpQ2a?fja?WUM{YEj}mA?s#gFsF96LvZi;veS6p?n
z=rptTiH%czT1g*fvI$v??FZZa&b~MIO|Q@ID{S2L&CYLkf4%#1=bxYX0r%GaRQX&H
z)Z%YHc51QoTrsEywu-TBV}5KK1644w@mL@efD2v7h94b-F+oSr%Kt}n#7-ahMrN5g
zJ}j5|HQ9Ql0kf+lSp9DHCLI5kq=Mx&Kn;US*RdQSGIE8ILeNw~#^$SFdbt!ZXugB_
zYbe;6nVmLYjRnB0RoSRuk+U1~EGBI)-&zILE?e24l%ywNs@>5C?R%-5Z0n1VPirrg
zVQVKm2P9&V&Me)7kbpDQ<e2F5RA{*rGvtJ#IMqy0JB)OGnPQ|BR$m0r2vHLP<AFOX
zbIHjdjI^&w93HWylx<Cey*747#qA~kh}xb7_)y5r+z%C<Ow7p9TyiookwS#FK}8FZ
zDB#sdbY7;vt301y2slfr{HkC@=VUY}%J9g0a+G#7QBx>sMp2SBHyNoj=FNBthvit6
zxDUpIj{<(Y6u1YD$!}r9VaDUgi_`X7pUEkY**MxF-0IB)7_Tm8WMJ8cZO*1n6xf@Q
z4Q@S9gk^=I=_5Mj3cecLy5JRh8`0+_!NNJjtwU9|+$XR*K08PO<&Mw4maFt{(otC`
z!z=cQD2YC)<ftqtPlbLKbhtt}3$q7b_cOykRFO~|1`T(rP<o9ygtnG$H`A=&OvuFV
z=C6V0Wn+V_pW%5pH8W}FC~6*r{{>8JM|R-975{Cz`xJgpwK37h^sZtaQrtxfj>uf$
zg@~c>0%kY8rg#=qT1M6)>ajD)5e++N;TVlSz1ln>DNgA)RUnzxeO(D0M+e)0MF$wI
zVR96SP|O>;*r^I?IrS8>;3G{7z;cgf(iNi<x|eRD{*bblOD(Ew)Zz+N{1Jh_BESg6
zmnhar;EM#vHM+<vt4*syl1Vx6%ZUd+7J<Dcdv9{i(r`lGB7AXqK{AeNzW^vD9Zwg+
z>$8%B5GG}J>y3M|)eCpV-Z}B+i5a=|%@c1uf}giN`9}&-Y^iO{j?F|{XCv6+N`LPe
z3H(aHSxV+t1uH7DuTp~|+l<<?^ACt@Gdd@?=1=fc)XzU6qIe!rRYdS$jJ|Zh8eHN8
zbs0UPexCqI8S1|$Fhk%l0m6q|2KffX2t1E}jaC!6LrK1!ANW1ky@!o^9oWd{@vSF?
zTN|%P#k=q?0z^p`mgg8%OI(3@KI1=vmhCHe7WJD1{tba|1K2ccD)kIMd&d3@MH;m?
zsNQ!7P)8Lh0fkI-5R&vqr{0zl(dm#3%ZzDkJi^oP;Ha-7&Y~ocOFN`q;x`b?jYrVi
zEmz_$*4xqH9L{wGt>t$jh@(5arg@HbJt{t8)Jdo3w9{}5CsdBfnLtKn+(a-F)Sh3O
zCgZrKAb1H@$KoYybyiu=V#X}ar9wGDjVhEA2RbN0Lq8W{R4Zgqtty*zVsRKw`nf8T
zTA9BF)$)UC1@Iq4@A#B(g+8qQqz`GVm+Zru3Vj&)qz^0hVQqyztox)7m+nJo8tM%W
z^Z}IQ+IHyKT<WU?{sn=%2>dYtcs97yNdim{aDigXf#h|H{Rsifm849q{ygRX9)YhC
zAf=@)cN!-om#W!!Q4!0c8n*a8UHY7Qg>s4Y(|vSevR2|v#(6=eFL)b;v@n2;=QEf-
zejc28;<D*P#tum-qRC8MP*RNEK$Fo?rTm_86c!Zm$HtH4TEBe!<pke8H{Y~%wrT5B
z>r`U8zR<L5X4S6Q@b1Mp!-d+0^MQ%Ld~I~LHkz-?KTxRMn)N|xpL?8?_FJKIk3?rl
zxv({V3;;rZ{@83}>%2^V*Ng;yRba(g##NTVlM&CuPu{2^BB=YxZRD*VuG8L=x0ds0
zLc{W<Ut7as(hs<e5y!^@%)B^ZSCK3;FE);wnU3c~I9WpJv>3q^**6$EaAb;*ow0E_
zfQ{!%-%aRcMKKq2!8rfBg#Ys>LGW>slP$%vVKO_d58Rs6D|E?LyEt6UFWw@cgqZu4
z4DL=w`Fg9I8Oz9wUIcCYZiAU;{bq!Qb}VEo>BPqcl`5R~6YiESRhw@&R+Xu8()4H(
zNO$14#=@CkhVxU^nd*#p6c0iv_t5WhdZ%VAl3r^*eWnJEUxdFA#%fu67|{>aUR_ap
zb+-2EGqsueQM5<fK1ZRf7=<vM2^r5?E$&j)yN*AB9lmQ^NUlZtM|yF>HnXDkx;);y
zu{@19bsE!5NU2dGN^M3ar@Kpa?~*D7L4rfxi&bzi3dHJ1?|%|U+Mgc5Y5j@5k#x~{
z`&tyIkaqP%Pm}Lra+{JmO(&KPbS0ySAsCR)rI1bE7ihUoA}5@P=vlbtW;4aUNr0st
zr5L4kFtvU6jh?^Be(RS%J$TW}Nh%|nj9wC{)T{c3ampB6CY9tlJ*XxV$}=jSAm-M6
zr~4A!nl29x#aXO-+x@ry*Z;I<H9zTRboGPgR|MFn0-~oAslKGbzNrxN;2?h-BB`FB
zI|P!d8t>P-WA!#o%(v(%Jp>*h@H_!Vnwcr_8eQfA^bk8GINp~!lhpm#(2h?d=rVPZ
z){|7!_o<wh%E^>7kUW>R%#g7aoMtm(&n%GdQAtm7h+8bU>Sms$K@kK)J-KM>HP-1+
zM04TlmDW8e<s1%;R#V-&d#F{tLUn?C!+PJahO{<4#4h_&1H+m|5{{kR8*#4cS<A?I
z@lEJs3~|<t+|8<-{4ZAi5^klv49x@1hsBBb!SUcrHRClobxvN*99E~==BiuZcU5GV
z=Q<%uTbxIzD>Ut#S+#REybC-+O?}q$liQlkk4}u{*S|709)<~jcmW&sswS#%{N`|>
zwmIwj$%?yfx*CEt+4~n7R-HdSaeT6Cq9gmj`)k*|@r4UtnA%-f+nVj5Lsgn4nkIKo
zt(}ToIdEm$T-~GD`xaI;o<H;Q8TuKR7|3}RZo7T*<iu#syRdeB{{9Q6a=u*GL};OL
z^&6oJp?vFCswaFockamhx3^5G)2)Tucjtl?O0L~L?VdhTSi2_|TG+f}djIU^d#|j&
z()In#vzt!jY9_VW+Kp(zxDS7^SF3-1n+q>}=9M%5eYLAP^3sX%6Z7Fsv*Aq(wJYg)
zCOf7ZE+1W3y=(g9^aC@E`+x3pRd4=<%Trx*V<Y;3QDldWd8Bz!7YN)SaF_roShyzY
zD#ZxU>ePtl2u+miyX$7dIv;PCl_6Xpf>1wEW6u6wBc7O-dv`rbbNvbSIgcX64u}^w
z%XDIKLa}2uigUi`Qx|sct}zMZ!9exmArMiRXSkEwl!=0I_?%!T9DUm@rTWt41Y=O4
zJxh(TbbEF+hL?FxgTRyq5G12lz!s4s2@6g;>lu5oXb-V-UE$(oJM#ugp|Fb#_1b?S
z^@&Zv?_fkSdy}2jYAID57z*nt%_oO3#;W1KSd`a*DWq!YI+iJ^LOC(IN_g3CxYE|9
z!y8%*8;m<kzL`}LZxpUjR?IJ@+BjWCUbZ4j%q6qW4Mi52+39>!Zhpaz_Lnh#7{TU5
z{8hSSaxar?nNZ6#u|Gz}=u?JeePA%nT#jK~DK)b!Zd+#uH5-SO;*=V`s~uJu^UzW~
z_*84WGYQADBz&PiOAqon0^13kCGZ>oEbokP=MmSvUHE_zr|Wmpbv6G)P@MZb>oUyS
z?5?)6!`e{J@c=Pn<2$3*|3HO^Hgh}5UW$E*KqrB>0i3qC^IahP-k_)CI2})%24@iG
zk2SiTl<qd%$GzDWuTY>H`51ZS<g|5;+?Y)Qr%Zf$zG2I3!<ItB)@<;n^$q7YPi&st
z^vd@4D@AVveRZL%@8@AxuxegzoRu3VSI)`n=mQ=HCJxNV`zP6fHd753up4au#DU4@
zrh|nQ`w^c3j8J4YvVUH-{;tXUUv1+F1uHry@6R3JsaMhHbM_jDqnYWTsANV_33n+e
zipz2tQ~1#xMIYgNVmR>MA{p42g3|;uIFA%Zm(!F%VB3kEP$JWcFG(~!6_C7GtU$Y7
zTUjTGQV;|j*6fB>-towF>&WdQMDO4=<tQ7v9f>1*PsojuA;IA3OnHeN(H`vh8Skl|
z9O6h}BQEhxgzCOzLLI=Do#HCKwxu%unNS7%ZZ_kG0W(gfAH*3ZA$A`}N{AJ!{(?#~
zlPE1%wyR3LMKR(v*m<ikv=WNd-F+(8l+j|;y~GW(C0X`igO!Y>HCBFqq8lR2u#y~Q
znyj+LR$A<@(YQuS`+3(wbuC!c>eXN=pPt+^U4MD~T=hMRt+w~!_wvm6Gr6;0eh!E5
z-L~fKws-cvxp(SVVdKu(+Fejq$szl~GCvo^Tmak11#s%|HxEr8zv8~~`1gaeyN+Lz
zPZ%7Vuhfe8Kk#%aM7c6%?IvXqAcjpN#RD;GA>V<UjRD`o#sLbH--Uk>%;|4<(2Eb3
z1r+H8Y4J~&!8Kpj!-T*d7*HM`1n4pr6>HfGNY^+27U0)0$yt<<zDweKYG6i!iOJx8
zXqoY^n&8*dgu9YRuDHqh&`IT!77J0d;Ioj06aS<Lvm4Y!-alY2iF6;RW60gzeG(_8
zjqW(fzSI|O+t$?w;u!5q4Pd48jcmiw*-66eoqb9B_vl*qdZ9sJyL#xnSK%R`S1HNU
z2}s0o72i;c;6&j81wPpE0s2ZvpYBg+>5(CP>wpecdmb&IE?X+n09%o|2h~)PHF0Qn
zqBGg2sqf)JBd)VZ2&{?sC(b0}=LXd{eI{>ojeS^KiZlP7TIdCUh*wklQffRkU$b$x
zW@G+np(d8~EQBNT;dQg&bx_kD$@|$ixfWJ5<UTX8YjXYMkqdWD9LTSqt=~9bch_v)
zU6<}J)a}5DXm;6ea#b!h8;(NGi`3@UXVclLazsyWz-CvLA$piUjah(n5loJhDb4Vm
z8*6u{DLoXC0-9xpnnW~#2%Q3{_{bUQl5DC`*j!`hQ79oc1F+FS;n(~$1wZ&`{}~J_
z<DM&hQzx7Wz+_rYb!;Y6dPf$V%)%BiSemq0rd6fJIQ~`UiXp=kYI6d$wd6olEH$9H
zUJMSBxW@oH9ls8rhvU=vA|9N_HXwGLUPb+16l2`h?;>{WqDS{9hX%WPG(FTw%BHxM
z8pT0@_?~4aKH$L)YQ6q{5xBeO)2IWNALi~!d))X*Nnjh*ZPfS<0<Ba=R;vg#((KTE
zY5FXr&6{Qy728c<50$k$Thhv4^hFQ!q_p&)I>N0ejMv*onfnQl6~)ovas{J0Ozom1
zyP&zk*5A=+(L!=4@e68>)L$Sskrnfi4YQFAg~*-b-mH{uB_418^vTOl&s84*-2oG~
zWn#-@>ttfeJ*{5Zo!e5VKb{TzB(f@ZcKo?X6;)^C_WV;*PrezS{>*H1JFaGs*4#cD
zX`h$r@0yvyFAMCF$FCyZY!u@u9L~w@ORT^D{IRS*C`}zHLeYt>Sdqr2B~SpiY3YLX
zbrYvGMSE<q4y#Je@C)0VpB>hPiDlS+Vd3oI<V;~nq>u`&lJyw<l3<l&QsL5NE%~+D
zSm6Ou{@OMptu&jr-F20jX8kU@8Ht~ERx4IFtzT<(tJSpL*y>g&gRO2AAK`VW|C6pC
z2QXJIIqPvpZMbafo<vudnx-<t1kO^yu(f>E2DE{16dIp&wJqZ>Q1Ra(5GPQwstv)7
z=pLhYoNH{N2V(2^i<Cij1-ARhNpM@uRw|wjpWyAnSGPLxm0d8p^ywk)0%x*3Twrr7
z_`h)jEf27al?`rC9+z`<xg(QZQ=ONav+|sLcmdS)%=j5l&Ce8S)<NUBvT;Uk%xbxl
z<0D`t@pp}1@(TeWyk#|hRj{ISa%1kKC?g_Q<;L>WP4kUW*T=oYyE#)=nA`;K<w#*Y
z(9d}<S!T6G>8nXkh^jg{T#bSHV*L_D-C|JGui0j@;N$SrMt~C%QB=6**ePn!hKG-1
zB?oO!?X>E}O0G~&h?dxH!B69ol`18ikp)d(?ur+tJ#!3-X}YCdvM|zEC=+rt-1(Ji
zrKZO{Rs{!je?~Y?!Jrq`{txjS%f_Qkw)c0?G^R<T%!2vU&>t>|Bg<<5uh#z7&j;Z$
zz4Y5zJYI!zVoWNO3!D9_)EQ3S)#>HVsA{vW`D@Im5a}54Uo{qHPBz!?zzAZCwG6b3
z-Kd*!#Sb}PF;Hx7M2t!5T0-1+QNnitXag(0SF8RdC6Kv^DFHc;NQ#Eh9@}Og2ZwNH
zuWzojv<{5|oidNghBmS#rD?;R^zm^1WhH1iNk{oZ+|Kj@Z=ol7k^oUL{!C9%evL)X
z^Z-3Nz8O&J+_NM(W4D1p(Rc(j*g$tEp>9#?q%Ts&TLc)>`YgqMp8z9!U!qtifh2%N
z2+$A$i)2B5PPXx+akhhabX_?cTU3BllG-r<8?DPf2Hvv}*<!J>`=$(aB@9|2*sEG!
zsb)?*eCO@P$>fZ@5j$^_vyqMSGW}iS*Ze}jS)^j;EfpwO(K#y1Q?JTsOB5Fo6n;Na
zpWFTAQSulX<-9g|3?0coo!^59_~-ndLS!q87C8dJ3+P3VR7!&<j2RV9Lhv2R?H1W)
zbdKG{ax~h&OOT_{9ub`}os>jlh?G-gVFz<Vru@gJ$Tk}gzQ9Uu!Wr+G)x<?B0Z6p8
zxrnZC{-f&7$l9OTBu9GEl0fG~VuO(z8|3LgfWT?bGQS%u=pnunC){X0;+Z}6tw>8t
z*rFt8MacGs4*Z67T0c%5<`g-o(5b`LHklP>r`TGeg<KLA-6q3qG{B!S+a?1$O%1Jh
zI^IiNM?hdq{#n~9v>S8(%6-qhYBRE9ZtL5&_wda!5S@7%eIUR{qurClCqs;92#r}<
zNUFBH9B~dz72S{cLv=O9m`J~qE*T$Qg3NYP#vTAvbV6K=EpMkn3`r#@#<14`ioFfs
zyhVk7ZOfDm@C~~3usWc{HGG;bp~8QJ*!Aw@Xj?hvs#GV8m(sq58!r!;m2l|`k$P-#
z1ecDF-_mBs$&&!NuE~?Q6k8CD5|d!ajctm=v0M3wvsxlM*TtiA)L=<TTw)_6!x{7`
ztRGOn0ZWi3<X(y&ByflT8@N?8W@`v+B(RCVVag@0iW~jilt$q1kff~vc)@kk?e{f&
zQ0w#U{h-G0Yogd)H*37U^}HUHjy39mcw8Cm!q<6a^K*gB(xx}X`6rQEIB|ySP>Y&m
zo;`X{B;g|wJdI5$-3KQV!<dHy71<Y+;SH07NQTc0Cet|m`3#?hUV3oS-0=@FGi4L8
zgX{xX{Al4EU4DbWRRW{~XD?Xt=D(8f;olcNsP4cE;m@U=1NgZ|lB65%D#?3mwM&v-
za1~tjKXz5$^7<wBKezyHH8x87Z@2(%?WvcJN;g~tZ?)A*Pa;n6R!g0<<%SF3){fhx
kwi_;hTlZ{}YHqjyZuLkF(!(f9@YbG<()tfv0IYBS9}d(e%>V!Z

literal 0
HcmV?d00001

diff --git a/neural_network.py b/neural_network.py
new file mode 100644
index 0000000..dc640ee
--- /dev/null
+++ b/neural_network.py
@@ -0,0 +1,241 @@
+"""
+Neural Network implementation from scratch using NumPy.
+
+Supports arbitrary hidden layer sizes, ReLU activations for hidden layers,
+softmax output, and cross-entropy loss trained with mini-batch gradient descent.
+"""
+
+import numpy as np
+
+
+def relu(z):
+    """Rectified Linear Unit activation."""
+    return np.maximum(0, z)
+
+
+def relu_derivative(z):
+    """Derivative of ReLU."""
+    return (z > 0).astype(float)
+
+
+def softmax(z):
+    """Numerically stable softmax activation."""
+    shifted = z - np.max(z, axis=0, keepdims=True)
+    exp_z = np.exp(shifted)
+    return exp_z / np.sum(exp_z, axis=0, keepdims=True)
+
+
+def cross_entropy_loss(y_pred, y_true):
+    """
+    Cross-entropy loss between predicted probabilities and one-hot encoded labels.
+
+    Args:
+        y_pred: (num_classes, batch_size) predicted probabilities.
+        y_true: (num_classes, batch_size) one-hot encoded true labels.
+
+    Returns:
+        Scalar average loss.
+    """
+    m = y_true.shape[1]
+    log_probs = -np.log(np.clip(y_pred, 1e-12, 1.0))
+    return np.sum(y_true * log_probs) / m
+
+
+class NeuralNetwork:
+    """
+    Fully-connected neural network trained with gradient descent.
+
+    Architecture: Input -> [Dense + ReLU] * num_hidden_layers -> Dense -> Softmax
+
+    Args:
+        layer_sizes: List of integers specifying the number of units per layer
+                     including the input and output dimensions.
+                     E.g. [784, 128, 64, 10] creates two hidden layers of
+                     sizes 128 and 64 with a 10-class output.
+        learning_rate: Step size for gradient descent (default 0.1).
+        seed: Random seed for reproducibility (default 42).
+    """
+
+    def __init__(self, layer_sizes, learning_rate=0.1, seed=42):
+        np.random.seed(seed)
+        self.layer_sizes = layer_sizes
+        self.learning_rate = learning_rate
+        self.num_layers = len(layer_sizes) - 1  # number of weight matrices
+        self._init_params()
+
+    # ------------------------------------------------------------------
+    # Parameter initialisation
+    # ------------------------------------------------------------------
+
+    def _init_params(self):
+        """He initialisation for weights; zeros for biases."""
+        self.params = {}
+        for l in range(1, self.num_layers + 1):
+            fan_in = self.layer_sizes[l - 1]
+            fan_out = self.layer_sizes[l]
+            self.params[f"W{l}"] = np.random.randn(fan_out, fan_in) * np.sqrt(2.0 / fan_in)
+            self.params[f"b{l}"] = np.zeros((fan_out, 1))
+
+    # ------------------------------------------------------------------
+    # Forward propagation
+    # ------------------------------------------------------------------
+
+    def forward(self, X):
+        """
+        Compute forward pass through the network.
+
+        Args:
+            X: (input_size, batch_size) input matrix.
+
+        Returns:
+            Tuple (output probabilities, cache dict for backprop).
+        """
+        cache = {"A0": X}
+        A = X
+        for l in range(1, self.num_layers + 1):
+            W = self.params[f"W{l}"]
+            b = self.params[f"b{l}"]
+            Z = W @ A + b
+            if l < self.num_layers:
+                A = relu(Z)
+            else:
+                A = softmax(Z)
+            cache[f"Z{l}"] = Z
+            cache[f"A{l}"] = A
+        return A, cache
+
+    # ------------------------------------------------------------------
+    # Backpropagation
+    # ------------------------------------------------------------------
+
+    def backward(self, y_true, cache):
+        """
+        Compute gradients via backpropagation.
+
+        Args:
+            y_true: (num_classes, batch_size) one-hot encoded labels.
+            cache:  Dict produced by forward().
+
+        Returns:
+            Dict of gradients keyed by 'W1', 'b1', …, 'WL', 'bL'.
+        """
+        grads = {}
+        m = y_true.shape[1]
+        L = self.num_layers
+
+        # Gradient of loss w.r.t. softmax output (combined softmax + CE)
+        dA = (cache[f"A{L}"] - y_true) / m
+
+        for l in reversed(range(1, L + 1)):
+            A_prev = cache[f"A{l - 1}"]
+            W = self.params[f"W{l}"]
+            Z = cache[f"Z{l}"]
+
+            if l < L:
+                dZ = dA * relu_derivative(Z)
+            else:
+                dZ = dA  # Already the combined gradient for softmax + CE
+
+            grads[f"W{l}"] = dZ @ A_prev.T
+            grads[f"b{l}"] = np.sum(dZ, axis=1, keepdims=True)
+            dA = W.T @ dZ
+
+        return grads
+
+    # ------------------------------------------------------------------
+    # Parameter update
+    # ------------------------------------------------------------------
+
+    def update_params(self, grads):
+        """Gradient descent parameter update."""
+        for l in range(1, self.num_layers + 1):
+            self.params[f"W{l}"] -= self.learning_rate * grads[f"W{l}"]
+            self.params[f"b{l}"] -= self.learning_rate * grads[f"b{l}"]
+
+    # ------------------------------------------------------------------
+    # Training
+    # ------------------------------------------------------------------
+
+    def train(self, X_train, y_train, epochs=20, batch_size=64, verbose=True):
+        """
+        Train the network using mini-batch gradient descent.
+
+        Args:
+            X_train:    (input_size, num_samples) training data.
+            y_train:    (num_classes, num_samples) one-hot labels.
+            epochs:     Number of passes over the training set.
+            batch_size: Mini-batch size.
+            verbose:    Print loss/accuracy each epoch when True.
+
+        Returns:
+            Dict with 'loss' and 'accuracy' lists (one value per epoch).
+        """
+        history = {"loss": [], "accuracy": []}
+        m = X_train.shape[1]
+
+        for epoch in range(1, epochs + 1):
+            # Shuffle
+            permutation = np.random.permutation(m)
+            X_shuffled = X_train[:, permutation]
+            y_shuffled = y_train[:, permutation]
+
+            epoch_loss = 0.0
+            num_batches = 0
+
+            for start in range(0, m, batch_size):
+                X_batch = X_shuffled[:, start: start + batch_size]
+                y_batch = y_shuffled[:, start: start + batch_size]
+
+                probs, cache = self.forward(X_batch)
+                loss = cross_entropy_loss(probs, y_batch)
+                grads = self.backward(y_batch, cache)
+                self.update_params(grads)
+
+                epoch_loss += loss
+                num_batches += 1
+
+            epoch_loss /= num_batches
+            epoch_acc = self.evaluate(X_train, y_train)
+            history["loss"].append(epoch_loss)
+            history["accuracy"].append(epoch_acc)
+
+            if verbose:
+                print(
+                    f"Epoch {epoch:>3}/{epochs}  "
+                    f"loss: {epoch_loss:.4f}  "
+                    f"train_acc: {epoch_acc:.4f}"
+                )
+
+        return history
+
+    # ------------------------------------------------------------------
+    # Inference
+    # ------------------------------------------------------------------
+
+    def predict(self, X):
+        """
+        Return predicted class indices for each sample.
+
+        Args:
+            X: (input_size, num_samples) input matrix.
+
+        Returns:
+            1-D array of predicted class labels.
+        """
+        probs, _ = self.forward(X)
+        return np.argmax(probs, axis=0)
+
+    def evaluate(self, X, y_true):
+        """
+        Compute classification accuracy.
+
+        Args:
+            X:      (input_size, num_samples) input matrix.
+            y_true: (num_classes, num_samples) one-hot labels.
+
+        Returns:
+            Accuracy in [0, 1].
+        """
+        predictions = self.predict(X)
+        labels = np.argmax(y_true, axis=0)
+        return np.mean(predictions == labels)
diff --git a/requirements.txt b/requirements.txt
new file mode 100644
index 0000000..e0485aa
--- /dev/null
+++ b/requirements.txt
@@ -0,0 +1 @@
+numpy>=1.21
diff --git a/test_neural_network.py b/test_neural_network.py
new file mode 100644
index 0000000..c52fe21
--- /dev/null
+++ b/test_neural_network.py
@@ -0,0 +1,272 @@
+"""
+Unit tests for the NumPy neural network implementation.
+
+Tests cover:
+  - Activation functions (relu, softmax)
+  - Loss computation (cross_entropy_loss)
+  - Parameter initialisation
+  - Forward propagation shapes and output validity
+  - Backward propagation (gradient shapes, numerical gradient check)
+  - Parameter update
+  - Train / evaluate helpers on a toy dataset
+"""
+
+import numpy as np
+import pytest
+
+from neural_network import (
+    NeuralNetwork,
+    cross_entropy_loss,
+    relu,
+    relu_derivative,
+    softmax,
+)
+
+# ---------------------------------------------------------------------------
+# Activation function tests
+# ---------------------------------------------------------------------------
+
+class TestRelu:
+    def test_positive_values_unchanged(self):
+        x = np.array([1.0, 2.0, 3.0])
+        np.testing.assert_array_equal(relu(x), x)
+
+    def test_negative_values_zeroed(self):
+        x = np.array([-1.0, -0.5, 0.0])
+        np.testing.assert_array_equal(relu(x), np.array([0.0, 0.0, 0.0]))
+
+    def test_mixed_values(self):
+        x = np.array([-2.0, 0.0, 3.0])
+        expected = np.array([0.0, 0.0, 3.0])
+        np.testing.assert_array_equal(relu(x), expected)
+
+    def test_derivative_positive(self):
+        x = np.array([1.0, 2.0, 0.1])
+        np.testing.assert_array_equal(relu_derivative(x), np.ones(3))
+
+    def test_derivative_negative(self):
+        x = np.array([-1.0, -0.5])
+        np.testing.assert_array_equal(relu_derivative(x), np.zeros(2))
+
+    def test_derivative_zero(self):
+        # Subgradient at 0 should be 0 (our implementation)
+        assert relu_derivative(np.array([0.0]))[0] == 0.0
+
+
+class TestSoftmax:
+    def test_output_sums_to_one(self):
+        z = np.random.randn(5, 10)
+        out = softmax(z)
+        np.testing.assert_allclose(out.sum(axis=0), np.ones(10), atol=1e-6)
+
+    def test_all_outputs_positive(self):
+        z = np.random.randn(5, 10)
+        assert np.all(softmax(z) > 0)
+
+    def test_numerical_stability_large_values(self):
+        z = np.array([[1000.0], [1001.0], [999.0]])
+        out = softmax(z)
+        np.testing.assert_allclose(out.sum(), 1.0, atol=1e-6)
+
+    def test_uniform_input(self):
+        z = np.zeros((4, 3))
+        expected = np.full((4, 3), 0.25)
+        np.testing.assert_allclose(softmax(z), expected, atol=1e-6)
+
+
+# ---------------------------------------------------------------------------
+# Loss function tests
+# ---------------------------------------------------------------------------
+
+class TestCrossEntropyLoss:
+    def test_perfect_prediction_low_loss(self):
+        y_pred = np.array([[1.0, 0.0], [0.0, 1.0]])
+        y_true = np.array([[1.0, 0.0], [0.0, 1.0]])
+        loss = cross_entropy_loss(y_pred, y_true)
+        assert loss < 1e-10
+
+    def test_loss_positive(self):
+        y_pred = softmax(np.random.randn(10, 32))
+        y_true = np.eye(10)[:, np.random.randint(0, 10, 32)]
+        assert cross_entropy_loss(y_pred, y_true) > 0
+
+    def test_worse_prediction_higher_loss(self):
+        y_true = np.array([[1.0, 0.0], [0.0, 1.0]])
+        y_good = np.array([[0.9, 0.1], [0.1, 0.9]])
+        y_bad = np.array([[0.1, 0.9], [0.9, 0.1]])
+        assert cross_entropy_loss(y_good, y_true) < cross_entropy_loss(y_bad, y_true)
+
+
+# ---------------------------------------------------------------------------
+# NeuralNetwork: initialisation
+# ---------------------------------------------------------------------------
+
+class TestNeuralNetworkInit:
+    def test_param_shapes(self):
+        nn = NeuralNetwork([4, 8, 3])
+        assert nn.params["W1"].shape == (8, 4)
+        assert nn.params["b1"].shape == (8, 1)
+        assert nn.params["W2"].shape == (3, 8)
+        assert nn.params["b2"].shape == (3, 1)
+
+    def test_biases_initialised_to_zero(self):
+        nn = NeuralNetwork([4, 8, 3])
+        np.testing.assert_array_equal(nn.params["b1"], np.zeros((8, 1)))
+        np.testing.assert_array_equal(nn.params["b2"], np.zeros((3, 1)))
+
+    def test_three_hidden_layers(self):
+        nn = NeuralNetwork([10, 16, 8, 4, 2])
+        for l in range(1, 5):
+            assert f"W{l}" in nn.params
+            assert f"b{l}" in nn.params
+
+
+# ---------------------------------------------------------------------------
+# NeuralNetwork: forward propagation
+# ---------------------------------------------------------------------------
+
+class TestForwardProp:
+    def setup_method(self):
+        self.nn = NeuralNetwork([4, 8, 3], seed=0)
+
+    def test_output_shape(self):
+        X = np.random.randn(4, 16)
+        probs, _ = self.nn.forward(X)
+        assert probs.shape == (3, 16)
+
+    def test_output_sums_to_one(self):
+        X = np.random.randn(4, 16)
+        probs, _ = self.nn.forward(X)
+        np.testing.assert_allclose(probs.sum(axis=0), np.ones(16), atol=1e-6)
+
+    def test_output_all_positive(self):
+        X = np.random.randn(4, 16)
+        probs, _ = self.nn.forward(X)
+        assert np.all(probs > 0)
+
+    def test_cache_keys_present(self):
+        X = np.random.randn(4, 5)
+        _, cache = self.nn.forward(X)
+        expected_keys = {"A0", "Z1", "A1", "Z2", "A2"}
+        assert expected_keys == set(cache.keys())
+
+
+# ---------------------------------------------------------------------------
+# NeuralNetwork: backward propagation
+# ---------------------------------------------------------------------------
+
+class TestBackwardProp:
+    def setup_method(self):
+        self.nn = NeuralNetwork([4, 8, 3], seed=1)
+
+    def test_gradient_shapes(self):
+        X = np.random.randn(4, 10)
+        Y = np.eye(3)[:, np.random.randint(0, 3, 10)]
+        _, cache = self.nn.forward(X)
+        grads = self.nn.backward(Y, cache)
+        assert grads["W1"].shape == (8, 4)
+        assert grads["b1"].shape == (8, 1)
+        assert grads["W2"].shape == (3, 8)
+        assert grads["b2"].shape == (3, 1)
+
+    def test_numerical_gradient_check(self):
+        """Verify analytical gradients match finite-difference approximations."""
+        np.random.seed(99)
+        nn = NeuralNetwork([3, 4, 2], seed=99)
+        X = np.random.randn(3, 5)
+        Y = np.eye(2)[:, np.random.randint(0, 2, 5)]
+        eps = 1e-5
+
+        probs, cache = nn.forward(X)
+        grads = nn.backward(Y, cache)
+
+        for key in ["W1", "b1", "W2", "b2"]:
+            param = nn.params[key]
+            numerical_grad = np.zeros_like(param)
+            it = np.nditer(param, flags=["multi_index"], op_flags=["readwrite"])
+            while not it.finished:
+                idx = it.multi_index
+                orig = param[idx]
+                param[idx] = orig + eps
+                p1, _ = nn.forward(X)
+                loss_plus = cross_entropy_loss(p1, Y)
+                param[idx] = orig - eps
+                p2, _ = nn.forward(X)
+                loss_minus = cross_entropy_loss(p2, Y)
+                numerical_grad[idx] = (loss_plus - loss_minus) / (2 * eps)
+                param[idx] = orig
+                it.iternext()
+
+            np.testing.assert_allclose(
+                grads[key], numerical_grad, rtol=1e-4, atol=1e-6,
+                err_msg=f"Gradient check failed for {key}"
+            )
+
+
+# ---------------------------------------------------------------------------
+# NeuralNetwork: parameter update
+# ---------------------------------------------------------------------------
+
+class TestParamUpdate:
+    def test_params_change_after_update(self):
+        nn = NeuralNetwork([4, 8, 3], seed=2)
+        W1_before = nn.params["W1"].copy()
+        grads = {
+            "W1": np.ones((8, 4)),
+            "b1": np.ones((8, 1)),
+            "W2": np.ones((3, 8)),
+            "b2": np.ones((3, 1)),
+        }
+        nn.update_params(grads)
+        assert not np.allclose(nn.params["W1"], W1_before)
+
+    def test_update_magnitude(self):
+        lr = 0.5
+        nn = NeuralNetwork([2, 2], learning_rate=lr, seed=3)
+        W1_before = nn.params["W1"].copy()
+        grad = np.ones((2, 2))
+        grads = {"W1": grad, "b1": np.zeros((2, 1))}
+        nn.update_params(grads)
+        np.testing.assert_allclose(nn.params["W1"], W1_before - lr * grad)
+
+
+# ---------------------------------------------------------------------------
+# NeuralNetwork: end-to-end training on toy data
+# ---------------------------------------------------------------------------
+
+class TestEndToEnd:
+    def _make_xor_data(self, n=200, seed=0):
+        """Simple 2-class linearly-inseparable toy dataset."""
+        rng = np.random.default_rng(seed)
+        X = rng.standard_normal((2, n))
+        labels = ((X[0] * X[1]) > 0).astype(int)
+        Y = np.eye(2)[:, labels]
+        return X, Y
+
+    def test_loss_decreases(self):
+        X, Y = self._make_xor_data()
+        nn = NeuralNetwork([2, 16, 2], learning_rate=0.1, seed=7)
+        history = nn.train(X, Y, epochs=50, batch_size=32, verbose=False)
+        # Loss should be lower at the end than at the start
+        assert history["loss"][-1] < history["loss"][0]
+
+    def test_accuracy_above_random(self):
+        X, Y = self._make_xor_data()
+        nn = NeuralNetwork([2, 32, 2], learning_rate=0.1, seed=7)
+        nn.train(X, Y, epochs=100, batch_size=32, verbose=False)
+        acc = nn.evaluate(X, Y)
+        # A trained network should do better than random (0.5 for 2 classes)
+        assert acc > 0.5
+
+    def test_predict_shape(self):
+        X, Y = self._make_xor_data(n=30)
+        nn = NeuralNetwork([2, 8, 2], seed=5)
+        preds = nn.predict(X)
+        assert preds.shape == (30,)
+        assert set(preds).issubset({0, 1})
+
+    def test_evaluate_returns_scalar_in_range(self):
+        X, Y = self._make_xor_data(n=50)
+        nn = NeuralNetwork([2, 8, 2], seed=5)
+        acc = nn.evaluate(X, Y)
+        assert 0.0 <= acc <= 1.0
diff --git a/train.py b/train.py
new file mode 100644
index 0000000..372b40c
--- /dev/null
+++ b/train.py
@@ -0,0 +1,131 @@
+"""
+Train a NumPy neural network on the MNIST dataset and report accuracy.
+
+Usage:
+    python train.py
+
+The script:
+  1. Downloads / loads the MNIST dataset.
+  2. Pre-processes the images (normalise, flatten).
+  3. Trains the network for a configurable number of epochs.
+  4. Evaluates and prints the final test accuracy.
+"""
+
+import numpy as np
+
+from neural_network import NeuralNetwork
+
+# ---------------------------------------------------------------------------
+# Hyper-parameters
+# ---------------------------------------------------------------------------
+LAYER_SIZES = [784, 128, 64, 10]
+LEARNING_RATE = 0.1
+EPOCHS = 20
+BATCH_SIZE = 64
+SEED = 42
+
+
+# ---------------------------------------------------------------------------
+# Data loading helpers
+# ---------------------------------------------------------------------------
+
+def load_mnist():
+    """
+    Load MNIST returning train/test splits as NumPy arrays.
+
+    The function tries several backends in order:
+      1. tensorflow.keras  (if TensorFlow is installed)
+      2. sklearn fetch_openml  (downloads the dataset on first run)
+
+    Returns:
+        (X_train, y_train, X_test, y_test) where
+            X_*: float32 arrays of shape (60000/10000, 784) in [0, 1]
+            y_*: int arrays of shape (60000/10000,) with class labels 0-9
+    """
+    # ---- Option 1: TensorFlow / Keras ----
+    try:
+        from tensorflow.keras.datasets import mnist  # type: ignore
+        (X_train, y_train), (X_test, y_test) = mnist.load_data()
+        X_train = X_train.reshape(-1, 784).astype(np.float32) / 255.0
+        X_test = X_test.reshape(-1, 784).astype(np.float32) / 255.0
+        print("Loaded MNIST via TensorFlow/Keras.")
+        return X_train, y_train, X_test, y_test
+    except Exception:
+        pass
+
+    # ---- Option 2: scikit-learn openml ----
+    try:
+        from sklearn.datasets import fetch_openml  # type: ignore
+        print("Downloading MNIST via scikit-learn (this may take a moment)…")
+        mnist = fetch_openml("mnist_784", version=1, as_frame=False, parser="auto")
+        X = mnist.data.astype(np.float32) / 255.0
+        y = mnist.target.astype(np.int32)
+        X_train, y_train = X[:60000], y[:60000]
+        X_test, y_test = X[60000:], y[60000:]
+        print("Loaded MNIST via scikit-learn.")
+        return X_train, y_train, X_test, y_test
+    except Exception as exc:
+        raise RuntimeError(
+            "Could not load MNIST. Please install tensorflow or scikit-learn.\n"
+            f"  pip install tensorflow   OR   pip install scikit-learn\n"
+            f"Original error: {exc}"
+        ) from exc
+
+
+def one_hot_encode(y, num_classes=10):
+    """Convert integer label vector to one-hot matrix (num_classes x num_samples)."""
+    m = len(y)
+    oh = np.zeros((num_classes, m), dtype=np.float32)
+    oh[y, np.arange(m)] = 1.0
+    return oh
+
+
+# ---------------------------------------------------------------------------
+# Main
+# ---------------------------------------------------------------------------
+
+def main():
+    np.random.seed(SEED)
+
+    # -- Load data -----------------------------------------------------------
+    X_train, y_train, X_test, y_test = load_mnist()
+
+    # Transpose to (features, samples) convention expected by the network
+    X_train = X_train.T  # (784, 60000)
+    X_test = X_test.T    # (784, 10000)
+
+    Y_train = one_hot_encode(y_train)  # (10, 60000)
+    Y_test = one_hot_encode(y_test)    # (10, 10000)
+
+    print(
+        f"\nDataset: {X_train.shape[1]} training samples, "
+        f"{X_test.shape[1]} test samples."
+    )
+
+    # -- Build & train network -----------------------------------------------
+    print(f"\nNetwork architecture: {LAYER_SIZES}")
+    print(f"Learning rate: {LEARNING_RATE}, Epochs: {EPOCHS}, Batch size: {BATCH_SIZE}\n")
+
+    model = NeuralNetwork(
+        layer_sizes=LAYER_SIZES,
+        learning_rate=LEARNING_RATE,
+        seed=SEED,
+    )
+
+    history = model.train(
+        X_train,
+        Y_train,
+        epochs=EPOCHS,
+        batch_size=BATCH_SIZE,
+        verbose=True,
+    )
+
+    # -- Final evaluation ----------------------------------------------------
+    test_acc = model.evaluate(X_test, Y_test)
+    print(f"\nFinal test accuracy: {test_acc * 100:.2f}%")
+
+    return history, test_acc
+
+
+if __name__ == "__main__":
+    main()

From f8c6a3400eba2113523cba0fe8aab8993b33df6b Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Wed, 1 Apr 2026 14:04:52 +0000
Subject: [PATCH 3/3] Add .gitignore to exclude pycache and build artifacts

Agent-Logs-Url: https://github.com/Bitu-Singh-Rathoud/neural-network-numpy/sessions/2f9cae04-079e-450e-9133-bd27adca2411

Co-authored-by: Bitu-Singh-Rathoud <247644259+Bitu-Singh-Rathoud@users.noreply.github.com>
---
 .gitignore                                      |   8 ++++++++
 __pycache__/neural_network.cpython-312.pyc      | Bin 9698 -> 0 bytes
 ..._neural_network.cpython-312-pytest-9.0.2.pyc | Bin 34626 -> 0 bytes
 3 files changed, 8 insertions(+)
 create mode 100644 .gitignore
 delete mode 100644 __pycache__/neural_network.cpython-312.pyc
 delete mode 100644 __pycache__/test_neural_network.cpython-312-pytest-9.0.2.pyc

diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..27b878d
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,8 @@
+__pycache__/
+*.pyc
+*.pyo
+.pytest_cache/
+*.egg-info/
+dist/
+build/
+.env
diff --git a/__pycache__/neural_network.cpython-312.pyc b/__pycache__/neural_network.cpython-312.pyc
deleted file mode 100644
index 881246a70595d72b6f26ecd0b2831dc358edfb88..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001

literal 9698
zcmb7KYit`wlAalI_@+ckqF$Emv1~o)=wT_2?I?*IC4MBaqr|bZ37qv2BhE;oOp)}=
z&^CEPoU_OR)UE+h@GeAM5}4jQTofb+_-FQC4mjXmI6z!#&%!{!8GHVd`*Bd@!yb@d
zSJgAamnrY24L4_cx~sdZtLm$&*1vhZE(*#Ueg7W$d>=*q7rvOqYCJr786H+Bfx1Eo
zw7^8^D~w?Igt}r8tWaA87HU?oL2YBGbCh8Jgc2NIG5Y8$_7oFzrvJk=B_<_4%1w#N
z&A4=pi!3Cf;({1ccqI~#aWhhUfs?}$uY~8gq#TLOa#P8LcT*!S*QI145tkI1<E80{
zBJom+n~Mm77~`URN|ZP`k{0D5?mh9s`y3xuA~y&{nVX4AH67(4mmHr_7Wi9SJgFp-
z$`HrL1THMaWqBADBE=IaE($LkOcIHS0(Uc_%yA2mSY&vbtZP={g$VTI1W^vd?}+O=
z+~ic4*7r(k<eCNkKN*2)g;L7j<_(UJUPdL*DrIu&5*1|92i_B53o{Xz{Q|6km$>(1
z5v6KhBNCDcG8z+0Xm;34WFfimC9TnE8E4{tczbj%z95cD$yiL3Mv>Q}F~a)cn9lm)
zSaKnedRAk|=sA)|X;w*$CM743cEf+U6{_3RW4Cu{VsRqJ=lv!3o^|V<C(umQumM>>
z*f0(cE0hYNbL})m&6nAMBlMz_8S|C#m9V1qv?!6oydiS&8FIem&D(fcNhL&$or%VI
zMX~`QefyFF9<^p%bVw+Vmzto}hn)gY-KHLSyg7RD%+krllb`jK+}ygAlTZ(8c7R05
z4)XATJ9zd{;tmDoYs*(S21*(TX6ULhC>V7`rOh2@7?r8&i*wf%m8qi-=K*}Oru#ur
zV5ea|8clJs!cRv<&J^^DKIp8w{3WAVL1tvlaZMBxLS#V>S|wbWW&v5!EaI(%W|5N%
zL5pO^R=YeGnNdVRW8q6EE#u}mj?>v}H>Qy~p%pnRpM(n7+tmE4-j)94{=C27U%R}~
zGG1&NzkP1Y)3kJU@oet+FW<R+X4CEa)%eP><zxB20{`I9M(gpS`}n%`xPA_0i2!=I
z4ss4(!vmZHJrBS2MWDN&DZ?lR_d(Jc@Vn}3BMwsn7&i0#%#(3b8B5hTl&`v?(yArR
zQj|GDWzxpk@SPygF5nmU`qNfs8q`q)E0&N%Arb~2!&7{kpN>Q$N<;*62gWHL6Nl&G
z3Ma<GaRGWn`DrmKgHc49lhUj_L7wm}6+%K2+#slTC>-Tw8Lc8QTp=`(!QCbrsUo3B
zN%2`E>eqw)YZvvNs3fIW)w(W$V1gOrZ-^2<3!TBh>N9?a%0BD=@8OSESFXP@t21@`
zRFKtJd5%wrhFC`9vzj#=jU*(Lu%K15={v<=oL-|BG$(X|?Z>BO8BZPAp-X5BvF;(m
zx`%LbsTX?mz<>D-sBTl+?Ub{5$+PImU0iIwJ-Ov?U1?u#|BSh3U$qx(YwVg_>U_TB
zKf3Nadi%{SFnq@rkLCLFFBV%4mE3)6bkQALw+3~w1%l;bitv-u(6mB{)D>Ew06A&E
zO}D^&LIZLlin0VnZ3IQ_1VtT!2QZYr;uIY4-6c4ocBfoHujbW3)WA^w6_|q@-q|GR
z{%|-BCZ7nBZtyw7jMw7X(;#LDQ+O_-!0@C567XgW@R1un$$fNMjL9PR95KIFxx*Yv
zmkz%n0>3i$(mTP!B^_WJ$tUSeBvGCKY?Bo<>XDct&We)E$q6wWnMon;D{~^wG7ZoU
zJ(2)_G6&M2qkUOc@r=tQ5(`I@f;l)s7EjcN;5-Lr4t4=<h9P<7apuFBk=YUMqvI!z
z4sm1SCx*CV_;BPZ_y!`cfGUE^QtL{<${MePI3q}esPD0(x(~qvwPB)@hU-z$xhP<K
z3|1Zj*9RDXNf8soqap4~{aJB?f;ht`qY8IqWGq-AB8#Z`@4=SQi-H$2S`tAegk(5E
zOiHRU_~^JXHe78Z(Q*<y`?o}3m9iu^a$M%Y-qjb?@|A(#G+1es3YBfK%D`KBE(JXa
zQY!c-(1@4Nt8AB5YQEX*SNZ+Qc%-QmI6+fVH@pWOx~Z^9gaMirIiR_#1WxnTuvxQ2
z#hB(oQPyQx^Mt_5Q$h(|;t>&%M07?P1maQ<s{yF$Op7C=j|rvQYqWC2Y<~&IoPz2$
zwdHUvITxKvO^Z!AWy8_4xu-vC-Eeen?mdjZ-CNyz^Vdq<hu1C^yGOFtJI$L8&qGI7
z(b4sL$DU_+CJEs83Kc=V!904YU==!-jZhGH4P8yQ^E7q!E_y1^x6FH!7MuvJpNx*K
zyY(SoElL?1=h2y}EA!sE26$C1Dr2rgK_8QpdGyxmiU9r*qN*ujh_|qsx?cDySXBB`
z>T?EzmMWU){nZuKBG5C8NpTly)(W*dzqV2-HfT#9c|!!t1t`r&BeD^{K(KLBjLgm{
z@=IJ=1hgkAI33{uhDV5)8VE<ds?i^4^z>BFso8Xa(^!yPFoVy*0(GH7U6J~sBA<j9
z2;_B1(O5zmJhox{L`OB-3?BnA)a>{RQBAFkN<+{s-CZZ1Rg!)dzRQ0D75HrJogjGb
zEq{CNdj8P8(bdtl$zu0#$v?6~S)CKv$<095!$4m#0Pfn_f%{@HFqA#J+1~xIy}#Jr
zU$|au4`$D89qjw!#rrRQaq9l52gkoTTRQmqKTiMC>_5!@b9nvom4Atr&VE!p_)+%s
z()q>n`2+dM{PTHx(Zk`W6ZcQ7eei%^yHGm#Vr_@tdARSLu!5%@Lo`9eV6_u90CA@y
zG&s^UIBbR?0&xJ-)5w4cOq~)4pafRvi>$9JAW~KoDI(_VE`6+3eQYWVE+J4-Z8%HD
zt|DdABdDT&4ca7+f-Wv8s@)XFj3wr-h;MbfYKKsO)n`K?M<o($mG?U{R?;@ZKzLwJ
z|K4Sy-0SfLKy?uX^CmBWy9qubpsf^7&d%wms|SdNdx^tpT}*8G5Zproums9$9ZVf~
z7a$;u+^Qn-luL+oE+--W=LYpiyw>v_;=+6woB;HSbuHq<*8s{ARfwlClShbFNk@PM
z(il`4eMO@ugRJyCwh&#y*ptMrBtik4(Dmj+jU_Wkui*FNSe2<r$Wj?C8LkqG5$hOS
zO9yJz$!NZWqy8Q$a46fT7Eq#`Gs|vzyji-TN;#eTL8W|cEAC}?o-fdOIe)I`JG5Jy
zKR3C|=ILc8bZGYFj;@?qKJ~li-mSLIy#MaGW%_a3p8Vw9w{rC6-hJ6Kxyc-#yObMP
zoT_>K#ng5y%nR%D`Ez~QTiIq^r^+x5%STBhI+Zh^B?IQPp_$bKu3h9h!-7mQDr;(u
zs>8b#O?bwpl%0o)<b(FIwZLR*0sz!s5gkH`b{&pPx)}%PRT*|*?M4BOwU_%iVJ+6G
zcNb_7OEj-pwN*JIE^|f%myA2csJ4t-b*qkfk4d-kOSIc!&O^Ya?5@l=0vk7k@9ft$
zGrR=wvFmFhI_zvThw8>zGPXM(qeg=^JCdz%EKe!4=_bnDjR+4RpRO(ZEP<vfMSNPD
z^$_aEM0{ohs_TYc6J?s<xYMYxy3YwN=yb}oG=reH<wZBN24p07`<0zs6H_{kfe*$8
zhPZ)gtp9NP@8I!+3#8$~K*a&pO#xA7qRGh9%Mu;~F%3bdv<IKjC}<8zgjiG-1>H~(
z8-X5&Mqeg=q#kE_h^>(J<1lpMG#1w;OC0HWL1TnTVmUP1BnI9$bPIy>!lx@TN=!A2
zbR*)xvq9Ti?=zb1aTq3l4;6_^{jF$1T9z&>Uda0w-^tRO9^cYC_^@rGn%f}#M(5<*
zxy4xiMzN`H&0q8$&N7evU7z(8=)zdx`u!I^9WD6}?@%6RHv)Kf{$l=mVQe*-pUuBi
z3=9B(H?%;3JPq*Nnwwp2%1%B4do%Wn<Yu7bv(dE!rS2g>{$=J-pd-gT+PlBd|Ha__
z!PTqPMrUj9{(C=J{Yl}qwTs`hZR|Z+{cY3VveLTTnv<407hl`%f`x7OP`*Ic^EAAp
zS`$<Bzwl6<NF{!412C;MA7D0q3d)1<4@y*D(4RyDaSQ$-tV6HH-mlRQ`=D7!cR-fV
z5+>f|E|dX||18Qzzeber`~`TL)*zeiJy(}<2GE8OMN}XjP9_8h8Fk;Xf?reM9o7LL
zMlg`iL+q1IV}*!Uo5}O)#Ebway{8uAYCirN2Y(C|q=6ebi;VLP8iWlbwaUvYdAP#q
zhlh?84;}j^@U8#bv2PE4bMv?Riige>J?G%#kLkVv4k3%P=ps{nY^ZfaWH3p-Le0WI
zf&Pri*ml7%-J`Qtg-i_C<vPzWM_&cYo^dE;F=`|#^X2db)?kz>ugwX86rjb=8Pz$X
z&*}nHs%>!#){IABGv2E8uz*!o6Ei@hufa;1RF^TbNo`WC=wbexF{n1H?p?q6KwtrX
z>x<fK;@(ezYZ-s0Me&+_RKMyo1GG#lj4xxH*#lI^r?!;)v_T(#<(v7WwwZH4YQUqm
zreTJRNA&`)@yUB<0T2$}(~wDqc51%W99j8=ltMXJ%mh@MU_;fJ$E;voQ9V_&*#!Hp
z{Q_0}Sed@_M+MYdFM>mL2+mKff=ji0N(pY&BY5D=s|N7*b4O)lj=s|ZtG59@H}U^j
zjn9|y9;w>1Hg2|_d8OnsVdH7E{tyzLKoO9Zt98OLlOgkXqb3LSY#-*rz&!93h{Qn%
z_hkY&sGKLRp<rft8)3beh==DOgNJ3RoTbB10`g>W09LYJG7YXl7U7nRbT+C=LVALG
zyK=RLd9&J%Kx#TJLrM(FyAmV{N$Pl%4~O9%F`VK=h!Hrl8t&#Cq|q-!F8I%KGsvyp
z0H%TmNRn>ABssuAI1kCjLD)ZcgO5UL8}p82j;hcD?%u38T635KiT5#O#`hGtp_ASQ
zIg+NJO54sL&$u*m^7$X=(X@jjJDK3pmXV_~X)mXrSO}<1a6ymmsSp#621m~VxM3j+
zN&T)=dKH^{v3e7$^H6CH5f==2pxO9DLX5$MO#+z~Y$tYKvm3Ib*$o-cZ2C!RuF8pM
zc7y4feGWJgmr}aB!;-GjyEu{S!;n0eoSBJ2ip`a(dM15HhFx=`G!e3*49R?jhtvfu
zfVAKO^aofVB<zWEr85h$NptFSfEjW-yc<$e_oI7LERnKwYb{uj#(OYS{w3%l_)$Gn
z&%U+sV#`R@zUgUQdVBHh{DGpUCrdvDhiEK6Q@Fl5UkU`DQ2U)N*)xy3d+r5SgAcn$
zi`}CS=u-E1_WWiS_ZxrVd|_X)>qz#k+P6ntJ<!y>Cwu;J*FNyB4iv7Gx{mBnF7JUH
zyV-r{Hy7awA^9L(9GzSroLujNq{?eK`>r1z?LT-=T~*i4mG&Q9_T<=Ha;qb_*7u;J
z)bV1@y4l%t&%5d^%&sL%oi6}=U)w`pZ_(FVcwxgAeB|rS3-FS2ZTj})uNQsXvU@uK
z3)$|ZydC-Hik?1LYMI2%Qs?pPo4}IX*~NDrdU}hV-onI&X9Qc`Dth)m^z;`!{e{$q
zXY3En?OVRUW_wTm{la+uEyw`b3b(c?`pDkRy#s5O^`0Sk8SdTeeQxdb!v1ZR35@^I
zV)eId9|xXpzeZEOj%?G@9v;d!f$I<90dH6v-40iB!~~w0n%g*}P*(s}>95N`0Y=%9
z{@1bv(u0qxn=m~oABjP(0qhq9^SWKpjcJuRX<&@&5+OAf^v7gE#)ePBpEgK*3a>jV
zmx21q>P;}QEnm_SVl4P{ufmU!{N|-uxWP4YgA!iEO4Cqjbf`8dsM(D<>N6Qe^FtUU
zW9pG)GTJ&<US58=K;J#J=*Y6!>s!tKrPN|7A1GYhXdYO13_P7eD4*)jfXJUd)gO1B
z8EeHkYUGANGR%5yu#zVx=l?(D^EXpqgt_o%7GXVmYZ~%Db>PGd)ug|kaq87Jt8tuN
zGq@gf%WG~JFi)B=2oA<aN8q;VY6U%Y{FwpngEU0oM-OWepET=&$j2Z@HjK?gNHD_H
z+{WSK9hqj+g-*sS4cP=Sx6(y;t&dwx)Oifw<QJeKacjq(mFRM`Fm^Ydg@khJO4o8%
z{@5D5;Xj<Up#a@nyqWvU!c4JasMI{X?ieOAG!<lK|9Tb6Z%=)97)S*z-?hQdcdtTq
zIq1>g!X*ZwPbj3hLZJn?*oxw_CltD#<fFzHdnhEt!=aEg0)t5UTUU`XP)Y)RAWXsi
zd`XeH^$i}vPgr5vT*kb>C)D?hhjl#ZVcDLYb~`(<)7{Q??wsJ*z|IBQ%DzEAX&Yxd
zb`rFk9emR5Wk+;!NCdKtc!j31aLS34egsZvAb*QO@{u}*lUbAa!G#q@y+1f!*7tW(
zx(TlUW0X7!)wYGE>F*hqwmxa1=++$@L-*{s82Z4D%SxZvY4XyC|BpLkp&@0DH6i%_
D`>6hs

diff --git a/__pycache__/test_neural_network.cpython-312-pytest-9.0.2.pyc b/__pycache__/test_neural_network.cpython-312-pytest-9.0.2.pyc
deleted file mode 100644
index 0bac5f5cd1479cee2031860413a6f0fc734c9460..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001

literal 34626
zcmeHw32+<NnI;-%0|W_@qGVFCEJ!jXLAEH7I;g|eVau{5JB(xbNL0xnAR8h<@p3mP
zi!?a)WIVfI#WMzN?}&0dV>tGfskO4hWRi`tnb{~+TU)7`q>2<MliP9Cq;^u9(eBPv
zWbH)R?D_rgb-!)^O@fXy<7|~*kw0F)tGnOv-|zKbRab{x2-)TzrGDA$a{Ucmn89mA
zZvH(Y=Us~HluMBmcb|01&3--o-Y&P~>i3=UOFSoVN@h91Q$hTC`a=Cxr>gqHr^5Z!
zr>gsFPSx~BPDLb_+x4JJ@xI_veDAvXJx<lKG(XY;BCU?4$w&)|w0f2nLRytbTfx%8
zNUIiU4J@q&X%UgOlBLxmtxlw^VrlhATOrbJ8*#@P5~}2Kg`OBlrK9Pjme!)D2i0i0
zCmHP+?tgS7I*=Sz6MgteKQpMFiKhC8`jY+0fpj9B8XRZ|g&t>hx(3fC)%~GpbX)XD
zS2}f;<wj2r58z^OK#OislYPT?N43Gz>HfsI7%TY5pr%E;2K$GGd1KKnT`H2|Xj~l}
z8j1Jef|q<Wp(gs1>7*J>;W1N*zLdspz?wNWs6LZWmFN&ECA#^IwVuRKQi~=Al<44a
z8aIfZMI|Mb9${7POLUzntFxtBO(-ecJj!d|9UU0%PpYXdyuHZk>PdE;DSfEnAtjMc
zvX&oL6RCmd_Gl9K8%AQZC)qcYRJG{fKr|6e4~|4pG@&Kaq3bjsI$}P(x`Xv_2k&6L
zzPuB3FZG@tAsAQiNYuS&lX}4D&MrQ+IK*JGaB~NK&byMXQxfJg=J5-zQy#?)=v6#`
zKAM-O{E81SpkO|pl1BnDS(j-*AH$ux%$BG*AO1b~zxe{dc~@HG^&0aU$s%-PjeJ#l
z*C;b4rA1D!xza?6%hemOT^P}H&`isSb`Ue}CAmgiF?UDI^D|!R=ZB1l>#z}#4`1}@
z?tvlQmr&Kjh#sJsof_!Y>l2!mRMT;m98W$yoaj^O-7dOyua@jPtxM-Lx=S?rtH-wY
z4E86ttHT2WNp(9-{_T7%Y%|uvHq4Kq5$8m@Uz``m4_i7dLnHbfEPZHD!z@0VjMF4Z
zYH_TRp2R?RQW@QBc7Ds!1@FKkYFhxFcfB93dFi3?hhB~s!p&LfeL3_}XgoA8H_yt=
zZ!1%+Z*@;SeNAq?iL$Pu5v5*dBl@oq`FKXO;w*Aov#4n7J+`s`o-yP<F(WQl%%k3p
zm$HExx?BNZ46eEgaea+3N`y45(HKptgBXfCi-%$$=~QY1HQ5Fr#^LJ9O@;7g$AP${
z{@J;U=l*S^u>GNH@^Q5h)rv+$tg&w(=R=Q3r4?!G%Q5$FXhpiqMx`5%PT>mg{Axst
zMuQSm8V#g2P<$nTE+@|oCA-o|MWY!RjYj$S)UomDPn}CD#!?wwT|6rGyiKU8wFB73
zqPY;>gt1t$^8DV3y{~<`u;T8l?6gGAU6c2g4~QR)u_f}a5c$vpf;v2W6r^WcB8+`8
z+vLBB+eETu1EE%-h#Cg?*=rXleCN=0dY71AjZjW4fjR>91n5<D@8CdED_<s!Y#1!M
z6F2eK==$Q(a42#Y+SOVC#Gp7Ls=?_*Vbk8f*zmp0-`RX+{dc!slOOnq!$I6xx9FbV
z6b-5{93IEv@OG%HsgKtHTqokP)EM{({VZn?OoX(=AecqA<DS}X05J$TW%6k5^M&w+
zQbZ*NV#{~7TxtF8UDxDe<pUu|$3Y++`?^Te3}eA*&MQ*4Am(OADefkeACnYm%$<=k
z?lT~2W1fsh>r2yJSm>3_nDwim%xOvSQk~Q8cfCBvR>xtrSgb^aa$?#jzSrHOSQ=wq
z#jgZ1-ju|yd}%^sEZABitX2{-KJz`z-!Fj~J131v`1g#t!N$lP98=P~xG4R~eeb<7
zX|%q1%g{(`OpD%oZ<M0DVj36}57G$i`c)0*Vp!gum_~_vVlrFA*6P+nhlfVk(oLv(
zT^wv|UtCQNsl;(7)TADaYl+iIO3`bMFs=xk`U9#usOl@?kOBG=={WFmUt)mq)R?q{
zu=Ogm8t3=lVZYdEzi8tZyp`Sdi*}7r4NyD+$MOxO=!I6m=-Ne53uW&_8(J5@^RAyp
z?s#2&<J^UFQ!A!?UwN(&**-1JM%rfNebZe4Gm*CGuGz@Gd71vM843KVz^-`wDwXF^
zJDijEvH9N-3+i$_J^(Q!9@j(hc>ka>+(+?nJWgCB&rxr~GpTC{L<vC9bg3}{_W)dU
zv9{Q2tHj|lK&7>f057;caJzkNAINUs3F!mb<J<Q^$n86ZsN46TbQ5qtuTNF!<wY=Q
z|IIs*RGH@=@x}al6?O1QE{_$lA%g1tlr?5!6=;r@gkEA=N&=x?lO6%_igj$Qzaj?2
zVRiQs@hS-!cduA=R+i$q<Tclp4LLikMT=G{zDxdh4NS#Ei9pA7dX4LZGgQ71b#LNa
zO8eO#{@u6oxBkQQ!HW_S(t~|iWJEo6zY5WEuwVDl?|@3L$cU*PG3ATCzAnhGN!_Cj
z_p3B47dgYDOQX66GHEfVa~oR&TvUyth89l`#zBooH`+#t%LL_x_Mvs{1OP|x$H$Mq
z^y%?WPac_6IfAcQnM-`RH|r6M&xTofLw@Hw?QgbEOZoP<4t?{!Z$J3;2QT0E=MP^|
zX5<am<cF8Q`Ol;0fb(%X9-IWgWy}h*x~rGwTS>tDa1a2;Jc?(`oAG44XO03V``oUK
z+rqG8{)|5-Y29ftcY2BYDG8t$KF|#xwPMi?#h<2OWucciqmuBhaedaMHQCxH&Z#7%
z#cVcbrE0H17_=mm*BFq9+ARBtRxbTS)~ZwOB^{o`Dv2iu*jjZeD_ETh<;3b#<k!87
zdIprB5?c1#*?Z5azeR5@yT8k7;B7+{-a>E$nM(cQH=jePGTv8C6H#Enb{CTVN<v09
z>zcm?x8Npj0slVyd&m6X8p0hamVirrfPld|NC&|=Y$tXBEJ5g5BNF8t#Lk$u#o#1j
zp|R*e#yaeaX&iMuie_S7Vk35{#4M`BvD4+g(T8lTMk`w34O7f$xy?*N#7No^(^|l7
zgvQt{TfxZSo(ysFo9v*b+jKAZnN|jO#;Q5r@-W@}5q5Jlz#p!S(um;eVPZWxhf{q}
z<!HLUZxB3<rZ>j5p26Whg)ukr?!Lj!L?0v6>Mp8De6P9(AQt2jC%vqlS826h7`zI9
z>Am)xc7A)F)7r+bID^x=qv&dvM#zhCI2M=F#4tSZU>t#KmK4=ohU1-z9YmYj4S+?k
ze5CHBv*TxH<mOkqCljyqB9d<f$o1x1Ck7B<d4)(bqE{o$P!QqoD$nN^0zz~2YB#?s
zSW%IUN|OnZZARy)DV~Y~`BqV!N8Xd0UsOdUGm1*tvm$lId@OriJWM$;B5F+p{N^-(
zT>x<+MkIDX7KE0*Tj~~Q3yg_iT`a^`#F#h`rd~moaAPBYF%cxkOmn$RlA(AV*o|&h
z?(j>8dE_B8viGIKTyB2|1=QmJMWUi$Gt^H}eg}b10~GTT&1|L#-E~|`CpuGoP<_Yy
z5^A@pbR2Chn{}0{+()+}jAMzBXLCLIRe5=8_tfT@m2HLaPRNr>up4jfo9epkWh}<M
zA9++wB#DtU`Cpf_*lA-&?V7J;j<1EZt$C7hGse@-MEW{n?(1F}l8pd-U|lnZ)Uu)6
zf^6M;dKmbP>0j7b($JbJF+x!4I)LJFt!LvpJdh%!6Hg7msxf*;*&r6>KS+(-3t)}d
zlZEi@TqD*#(LT8+zpJofi%lm+tl``3P#f=`I(L~fgZHzM>WBqYYDpzLq#h#h7y)`w
zHASGGz+nPJ+qjTJQ${7Q7ylZilaS-_JtC24vx$V9gEvDS-%*KIqM}{+7r`9ao22nx
zO0pU8!j#fL-Elt|ksjbSBr-7-2{+Ww40Ftts4ko_2kDR$*iO2ObxGz^xjsZKw;kVP
zECI=4YTd-tW_;QOOJs6XeJsYp3M}Nr_?q|#EbK~K`woIqM&L*=jMQH&II+c6Nn%+m
zl(iVafT2|JE3y*I_)}b|5=fJNo&{S=PON9ZT>0j&!6x~}ykLz&9UO&{1)%%8e>wjj
z|2p&Id!bp8?g4vLM?4VNljuE+0Rk7Jd&$gr(XD<K<<%zvinJ@cxprmu#VSj&qOw&k
z*b?;%RP%QT#0ive9Wqy&_($SHYEsehn^uQm5u~Cdxv5_Suxm(m*)R5JBv6>V1Xq|a
zGi@AJ#naH4vsP2g$_j(&Gt6i@y4z_T7$&8~tI>MaUIzfSs9SNqaiVeZ{``u`{ZpQ)
zlljq^)oq2komu}<S&{YmO;ei+k=CqtVbz)$c}+H*OO9ve<u&-b#xMDWfU}g!uL@Rl
zPF|BsiZUW{RbEq0HmHNxl2!rj%s?6oiU%jM;AD2RYc9;m1@c0zv*5TKg-%m}<G>%d
zE8JBqQnO8$OYyM#8h1I(HKJ}y$4EL(p--{U#ZitfAoI2_Eln*d^NI!6>#)ph2&2i?
zJ{itRLR!e#W}iXkwd85S`X!;fh9UDV_lZ{8+IM&&i%<iFE%_u?Ns@UL*+N1U`o!L@
z(-R1ZvEmbml6V5S!qW=WT%nwWY=f`+nAGi8LQ0hl*^)7W1zSsmGD|{cF=Wf+QHF3K
zixN|KJqo~NQT1`efNb{~0w)(4p~JSsLoOJ`w7VG^hU91orR@+(8G=1W?Tt`VK8~DQ
zDB*JiNJb?YvVGL92`6#(DJmb~f+G#Q`Y-`5Q6@*=GC>Nv6qtiw1Ba{PiPXLH05(nd
z9yK9SxJ<ZLO(Mksdvi99h>j%zGP8dv-@@^AQNhnrlfMKY4Bhy8{Toden%-D<Vck^y
z)MJIF9fi79j-OXeteSMcvih|n(2S2<7@1$UV|Lw+>ALCrF1ris4$RzsV7BfaloaTC
zM&3Pr1Yjo8I(=j|vU^^pzxRv;ekI_nyn8OPo?jMhaZ%u^ygR>1<eO1@_LP-vM(5bg
zcp)?}eMFQO5fmxM{)A5%_WuP&)Q<f>1Z!p&AOy`oyG8(h_JR$B+e{G~=(dR+PPJHh
zUe{)b^w`)ySiA&Y_uA4uwsc<x?A{q-4N1T0*4#Gch$vr4fM(MV{idI4+4P%st(Cym
zR@EY0Z?Sw1%wd|=4GT70Z`G(ZbFJU<8V07Nkz?T_va14rR%BO2Ny4;B&=TxE;u8oC
zpyCsVl6ZokZ6qD~Eo0&E-@eha>r0oOPj(T)VLn+RM8&=nv=4n3`mG;5Vflp~ST^+D
zVWm0~FvJ}jr(v514t&_6T467=!TvEHM4lSE$n&EgQuuF&)D2i%>eB!q#fc9|in(~y
z3Q5V(_ma%SHOWveGwKaG&8hcZuHGfYK+;Y3;Gn|Dw(f(nHle^|Mr2DZrru1$tUgC|
zK*#qoE%1&}!+OS;f>v+^EV)g6hF-$Z-V#Hg{xbsqoIoid-p2_s(-a%Z+14WE>`o`;
z?0TkJfs3D(w98XHsqUVn%0;R@i;GlC)g#$Z8-l*z*OsYj>(9#*^5o9EXL4&^o4P;W
zJJZx!sBO#o?DYGIxn0Mu4FBMCVb>Eg@)MxrGrOLcjXW_g)8Bh$3cnO^R(^s}aw=1>
z#YKUu@)Jb9jeIj|&nEJ1WSh}Bb~jOoNAOU1LQz)4MKRp!%CN{cqxS3{oEF(;ly0?)
zZeQ|;bjB%kKz&-Gw93$|d4a$m5jaeskpMCA>YxcegNRlSK>8}L@AePE9^amuA+PUA
zLsLa1yYMf9sS{eBrfMZkry|GS<M3>DR#p8j6xDXlrqsPQBRTRqh8zj?w4mAIVc}yV
zY^hrewbS!9Jmkb#II5j&b(}nI1jVgTPT=+m<t&=$f8ArqHIUkDa*gm2u(jkwcNUXt
zgslyhX6JqlHlZ8gt+ZIbh723^C;iH%Td56m(SEI}(68Z7`jtA=<hz&MuhkX$wdRw4
zUAkYvVMRLNzH<h}t`lZt79QRGR4dfRuh2E+sCNNelt#A^t0fHOmYp0{@D{eQ1m<_c
ztq-genq`d9lhi^hK#}$ttd*KYqVDcA?{%H%4DQ{AdvByl^Z>{g*I&UkcrKEObw3>a
z68)O)V>ebUm9?{u+C4C!wo{w?2v`~oZu8!y9-uav)`PQF`yA98+-IBJ7j}8tg~K;B
zl+ijv9Ze5t6eKH>cA4`^)vI6ISEz2zx))&A9vL5*Y#4uTMsCSIHy^okHgYGTQ+|M%
z$emOE*+|R0On=vm1b$Uu#aYHxmcf$|pYrpl9ir|hxB4g@_@3NSuALzhE7Q*0fni;W
zhpC`77a9h%X9Q4ZCoY9?JvNQK16QI5Zo|~RNUq=5{k?>3E8JhmhSvRoWBeWJ%c!ef
zAn+Q2KPB)60N7aSMQ(zsqZqNKYMubt3~Ywtt}S0hyl98ZIwsh3rm*8ogUQ&~JBRHQ
za4%$Tw#-tr$l6@I+7R7`P!^;bu~LP5A#FX%cfON)Gd1mfbMTw$x6gh3+!cSJ{lPi;
zp~{7>$q$v!BN}NokNN<YoJUlO1(<C;X8#^_<)km1C$0^)5cgMHf8vIZx#Y+fGylS~
z+z|UU8xQF;*I=w=y0#);LR%e&xo7D+;9)0t)m{e<^<CpVN({D6nj)c9D$=1i-CdL<
z9bAPcwRmk=@Y?X)v+>%(T4QU;NrSN%uWi^6V3+u0OfcHNO=FUiUVP0<+bqM_qRQ<%
zH1lli&SttKW`lVm>e5p&Nq46PploU(4v2OyU?(g^VMFIyBt#4)64;^^&YQq7kgyhf
zUo1Z8EE0<NAhHQDU6IrotKzH;d0uxXISMS{Yf9|#mIEL+#k2eAQXgk^-l8{P{LZr!
z`+Wchu7+=cO82PAWV|P(D9HghQ;ooiKic9b$aCh#q5Lpw-{f(wuDQEVy)Em0ziw6T
z>B;r4thwoO2e)L8EUd(Jt}E9*6W%=8`9|tO3Xxp<WNK!`=J_!FUE|mM0vir<?K~<V
zt0}T4J9%^t?WIH+5kUk0)VIV7-0<OL7t_=HOB#)L2uu?=Ou*99<mi$>5B`Oo#^>Aq
zL5;_ETp~pcZ<!V9LLm{*_blApil3!CitPKNNBpsX9-*N=#&?@Pip}Ol+8WZa5be2n
zH*40i4RLLa;=UxPQN|TFAA(jGwudpg4;v5U2ik_5RKVO*V{EEklo)Z>sE)B^CX5Z_
z!$WawiSHRyMsF{9t`>Xl@1fSsS^$vh`gLCoPD^w3ZCURh2Fi&nF<I;ta6|zSSzPkq
zq+Y7*;-7JJ?RWH`6lCf1xbqkhXLqjy`NB3V4?LvEAO^073}V{XmL{Sg_hb%Zb!_G^
zf0~#m7J7-3DhWlT*CNtRq)jNeKzgl`q)RLbaN#fp9cbQPp{yXi70L-lQVAF&8p<&E
zR9Qv_;zn7pwM3f{OTsdY4DhRBV(vv$1;4owKPttG<laS`!As;`a;oy-xQ0#=f0=RM
zbJP%_ef1In!u7G55|?m}w*wG8>F5UaO)6+%{t~2LBH(i4fi8Vk{R>pnrMUV<s{R%M
zG9joI@(<$D^tUpW8<%O#jW;CJ@=cdET)+{&y~qn_M<ZSy3lCfLOLyfgthntwJTT86
znmCkiEv#rBm$Sa?(+iP?m!2Je7CzdO-2mBVC%b1&ujp(1+U^{UIC$+W^ycKv4QFjL
zN{-df@>FuyHoUoc0*5G3jsb+Y7Q$Tpcrkzt`#4|%r>$VX8RYR%5*FDYMi{XqEV)6<
zn)}?v<rpg4Tzi8~b5;F!REUXdpQ2a?fja?WUM{YEj}mA?s#gFsF96LvZi;veS6p?n
z=rptTiH%czT1g*fvI$v??FZZa&b~MIO|Q@ID{S2L&CYLkf4%#1=bxYX0r%GaRQX&H
z)Z%YHc51QoTrsEywu-TBV}5KK1644w@mL@efD2v7h94b-F+oSr%Kt}n#7-ahMrN5g
zJ}j5|HQ9Ql0kf+lSp9DHCLI5kq=Mx&Kn;US*RdQSGIE8ILeNw~#^$SFdbt!ZXugB_
zYbe;6nVmLYjRnB0RoSRuk+U1~EGBI)-&zILE?e24l%ywNs@>5C?R%-5Z0n1VPirrg
zVQVKm2P9&V&Me)7kbpDQ<e2F5RA{*rGvtJ#IMqy0JB)OGnPQ|BR$m0r2vHLP<AFOX
zbIHjdjI^&w93HWylx<Cey*747#qA~kh}xb7_)y5r+z%C<Ow7p9TyiookwS#FK}8FZ
zDB#sdbY7;vt301y2slfr{HkC@=VUY}%J9g0a+G#7QBx>sMp2SBHyNoj=FNBthvit6
zxDUpIj{<(Y6u1YD$!}r9VaDUgi_`X7pUEkY**MxF-0IB)7_Tm8WMJ8cZO*1n6xf@Q
z4Q@S9gk^=I=_5Mj3cecLy5JRh8`0+_!NNJjtwU9|+$XR*K08PO<&Mw4maFt{(otC`
z!z=cQD2YC)<ftqtPlbLKbhtt}3$q7b_cOykRFO~|1`T(rP<o9ygtnG$H`A=&OvuFV
z=C6V0Wn+V_pW%5pH8W}FC~6*r{{>8JM|R-975{Cz`xJgpwK37h^sZtaQrtxfj>uf$
zg@~c>0%kY8rg#=qT1M6)>ajD)5e++N;TVlSz1ln>DNgA)RUnzxeO(D0M+e)0MF$wI
zVR96SP|O>;*r^I?IrS8>;3G{7z;cgf(iNi<x|eRD{*bblOD(Ew)Zz+N{1Jh_BESg6
zmnhar;EM#vHM+<vt4*syl1Vx6%ZUd+7J<Dcdv9{i(r`lGB7AXqK{AeNzW^vD9Zwg+
z>$8%B5GG}J>y3M|)eCpV-Z}B+i5a=|%@c1uf}giN`9}&-Y^iO{j?F|{XCv6+N`LPe
z3H(aHSxV+t1uH7DuTp~|+l<<?^ACt@Gdd@?=1=fc)XzU6qIe!rRYdS$jJ|Zh8eHN8
zbs0UPexCqI8S1|$Fhk%l0m6q|2KffX2t1E}jaC!6LrK1!ANW1ky@!o^9oWd{@vSF?
zTN|%P#k=q?0z^p`mgg8%OI(3@KI1=vmhCHe7WJD1{tba|1K2ccD)kIMd&d3@MH;m?
zsNQ!7P)8Lh0fkI-5R&vqr{0zl(dm#3%ZzDkJi^oP;Ha-7&Y~ocOFN`q;x`b?jYrVi
zEmz_$*4xqH9L{wGt>t$jh@(5arg@HbJt{t8)Jdo3w9{}5CsdBfnLtKn+(a-F)Sh3O
zCgZrKAb1H@$KoYybyiu=V#X}ar9wGDjVhEA2RbN0Lq8W{R4Zgqtty*zVsRKw`nf8T
zTA9BF)$)UC1@Iq4@A#B(g+8qQqz`GVm+Zru3Vj&)qz^0hVQqyztox)7m+nJo8tM%W
z^Z}IQ+IHyKT<WU?{sn=%2>dYtcs97yNdim{aDigXf#h|H{Rsifm849q{ygRX9)YhC
zAf=@)cN!-om#W!!Q4!0c8n*a8UHY7Qg>s4Y(|vSevR2|v#(6=eFL)b;v@n2;=QEf-
zejc28;<D*P#tum-qRC8MP*RNEK$Fo?rTm_86c!Zm$HtH4TEBe!<pke8H{Y~%wrT5B
z>r`U8zR<L5X4S6Q@b1Mp!-d+0^MQ%Ld~I~LHkz-?KTxRMn)N|xpL?8?_FJKIk3?rl
zxv({V3;;rZ{@83}>%2^V*Ng;yRba(g##NTVlM&CuPu{2^BB=YxZRD*VuG8L=x0ds0
zLc{W<Ut7as(hs<e5y!^@%)B^ZSCK3;FE);wnU3c~I9WpJv>3q^**6$EaAb;*ow0E_
zfQ{!%-%aRcMKKq2!8rfBg#Ys>LGW>slP$%vVKO_d58Rs6D|E?LyEt6UFWw@cgqZu4
z4DL=w`Fg9I8Oz9wUIcCYZiAU;{bq!Qb}VEo>BPqcl`5R~6YiESRhw@&R+Xu8()4H(
zNO$14#=@CkhVxU^nd*#p6c0iv_t5WhdZ%VAl3r^*eWnJEUxdFA#%fu67|{>aUR_ap
zb+-2EGqsueQM5<fK1ZRf7=<vM2^r5?E$&j)yN*AB9lmQ^NUlZtM|yF>HnXDkx;);y
zu{@19bsE!5NU2dGN^M3ar@Kpa?~*D7L4rfxi&bzi3dHJ1?|%|U+Mgc5Y5j@5k#x~{
z`&tyIkaqP%Pm}Lra+{JmO(&KPbS0ySAsCR)rI1bE7ihUoA}5@P=vlbtW;4aUNr0st
zr5L4kFtvU6jh?^Be(RS%J$TW}Nh%|nj9wC{)T{c3ampB6CY9tlJ*XxV$}=jSAm-M6
zr~4A!nl29x#aXO-+x@ry*Z;I<H9zTRboGPgR|MFn0-~oAslKGbzNrxN;2?h-BB`FB
zI|P!d8t>P-WA!#o%(v(%Jp>*h@H_!Vnwcr_8eQfA^bk8GINp~!lhpm#(2h?d=rVPZ
z){|7!_o<wh%E^>7kUW>R%#g7aoMtm(&n%GdQAtm7h+8bU>Sms$K@kK)J-KM>HP-1+
zM04TlmDW8e<s1%;R#V-&d#F{tLUn?C!+PJahO{<4#4h_&1H+m|5{{kR8*#4cS<A?I
z@lEJs3~|<t+|8<-{4ZAi5^klv49x@1hsBBb!SUcrHRClobxvN*99E~==BiuZcU5GV
z=Q<%uTbxIzD>Ut#S+#REybC-+O?}q$liQlkk4}u{*S|709)<~jcmW&sswS#%{N`|>
zwmIwj$%?yfx*CEt+4~n7R-HdSaeT6Cq9gmj`)k*|@r4UtnA%-f+nVj5Lsgn4nkIKo
zt(}ToIdEm$T-~GD`xaI;o<H;Q8TuKR7|3}RZo7T*<iu#syRdeB{{9Q6a=u*GL};OL
z^&6oJp?vFCswaFockamhx3^5G)2)Tucjtl?O0L~L?VdhTSi2_|TG+f}djIU^d#|j&
z()In#vzt!jY9_VW+Kp(zxDS7^SF3-1n+q>}=9M%5eYLAP^3sX%6Z7Fsv*Aq(wJYg)
zCOf7ZE+1W3y=(g9^aC@E`+x3pRd4=<%Trx*V<Y;3QDldWd8Bz!7YN)SaF_roShyzY
zD#ZxU>ePtl2u+miyX$7dIv;PCl_6Xpf>1wEW6u6wBc7O-dv`rbbNvbSIgcX64u}^w
z%XDIKLa}2uigUi`Qx|sct}zMZ!9exmArMiRXSkEwl!=0I_?%!T9DUm@rTWt41Y=O4
zJxh(TbbEF+hL?FxgTRyq5G12lz!s4s2@6g;>lu5oXb-V-UE$(oJM#ugp|Fb#_1b?S
z^@&Zv?_fkSdy}2jYAID57z*nt%_oO3#;W1KSd`a*DWq!YI+iJ^LOC(IN_g3CxYE|9
z!y8%*8;m<kzL`}LZxpUjR?IJ@+BjWCUbZ4j%q6qW4Mi52+39>!Zhpaz_Lnh#7{TU5
z{8hSSaxar?nNZ6#u|Gz}=u?JeePA%nT#jK~DK)b!Zd+#uH5-SO;*=V`s~uJu^UzW~
z_*84WGYQADBz&PiOAqon0^13kCGZ>oEbokP=MmSvUHE_zr|Wmpbv6G)P@MZb>oUyS
z?5?)6!`e{J@c=Pn<2$3*|3HO^Hgh}5UW$E*KqrB>0i3qC^IahP-k_)CI2})%24@iG
zk2SiTl<qd%$GzDWuTY>H`51ZS<g|5;+?Y)Qr%Zf$zG2I3!<ItB)@<;n^$q7YPi&st
z^vd@4D@AVveRZL%@8@AxuxegzoRu3VSI)`n=mQ=HCJxNV`zP6fHd753up4au#DU4@
zrh|nQ`w^c3j8J4YvVUH-{;tXUUv1+F1uHry@6R3JsaMhHbM_jDqnYWTsANV_33n+e
zipz2tQ~1#xMIYgNVmR>MA{p42g3|;uIFA%Zm(!F%VB3kEP$JWcFG(~!6_C7GtU$Y7
zTUjTGQV;|j*6fB>-towF>&WdQMDO4=<tQ7v9f>1*PsojuA;IA3OnHeN(H`vh8Skl|
z9O6h}BQEhxgzCOzLLI=Do#HCKwxu%unNS7%ZZ_kG0W(gfAH*3ZA$A`}N{AJ!{(?#~
zlPE1%wyR3LMKR(v*m<ikv=WNd-F+(8l+j|;y~GW(C0X`igO!Y>HCBFqq8lR2u#y~Q
znyj+LR$A<@(YQuS`+3(wbuC!c>eXN=pPt+^U4MD~T=hMRt+w~!_wvm6Gr6;0eh!E5
z-L~fKws-cvxp(SVVdKu(+Fejq$szl~GCvo^Tmak11#s%|HxEr8zv8~~`1gaeyN+Lz
zPZ%7Vuhfe8Kk#%aM7c6%?IvXqAcjpN#RD;GA>V<UjRD`o#sLbH--Uk>%;|4<(2Eb3
z1r+H8Y4J~&!8Kpj!-T*d7*HM`1n4pr6>HfGNY^+27U0)0$yt<<zDweKYG6i!iOJx8
zXqoY^n&8*dgu9YRuDHqh&`IT!77J0d;Ioj06aS<Lvm4Y!-alY2iF6;RW60gzeG(_8
zjqW(fzSI|O+t$?w;u!5q4Pd48jcmiw*-66eoqb9B_vl*qdZ9sJyL#xnSK%R`S1HNU
z2}s0o72i;c;6&j81wPpE0s2ZvpYBg+>5(CP>wpecdmb&IE?X+n09%o|2h~)PHF0Qn
zqBGg2sqf)JBd)VZ2&{?sC(b0}=LXd{eI{>ojeS^KiZlP7TIdCUh*wklQffRkU$b$x
zW@G+np(d8~EQBNT;dQg&bx_kD$@|$ixfWJ5<UTX8YjXYMkqdWD9LTSqt=~9bch_v)
zU6<}J)a}5DXm;6ea#b!h8;(NGi`3@UXVclLazsyWz-CvLA$piUjah(n5loJhDb4Vm
z8*6u{DLoXC0-9xpnnW~#2%Q3{_{bUQl5DC`*j!`hQ79oc1F+FS;n(~$1wZ&`{}~J_
z<DM&hQzx7Wz+_rYb!;Y6dPf$V%)%BiSemq0rd6fJIQ~`UiXp=kYI6d$wd6olEH$9H
zUJMSBxW@oH9ls8rhvU=vA|9N_HXwGLUPb+16l2`h?;>{WqDS{9hX%WPG(FTw%BHxM
z8pT0@_?~4aKH$L)YQ6q{5xBeO)2IWNALi~!d))X*Nnjh*ZPfS<0<Ba=R;vg#((KTE
zY5FXr&6{Qy728c<50$k$Thhv4^hFQ!q_p&)I>N0ejMv*onfnQl6~)ovas{J0Ozom1
zyP&zk*5A=+(L!=4@e68>)L$Sskrnfi4YQFAg~*-b-mH{uB_418^vTOl&s84*-2oG~
zWn#-@>ttfeJ*{5Zo!e5VKb{TzB(f@ZcKo?X6;)^C_WV;*PrezS{>*H1JFaGs*4#cD
zX`h$r@0yvyFAMCF$FCyZY!u@u9L~w@ORT^D{IRS*C`}zHLeYt>Sdqr2B~SpiY3YLX
zbrYvGMSE<q4y#Je@C)0VpB>hPiDlS+Vd3oI<V;~nq>u`&lJyw<l3<l&QsL5NE%~+D
zSm6Ou{@OMptu&jr-F20jX8kU@8Ht~ERx4IFtzT<(tJSpL*y>g&gRO2AAK`VW|C6pC
z2QXJIIqPvpZMbafo<vudnx-<t1kO^yu(f>E2DE{16dIp&wJqZ>Q1Ra(5GPQwstv)7
z=pLhYoNH{N2V(2^i<Cij1-ARhNpM@uRw|wjpWyAnSGPLxm0d8p^ywk)0%x*3Twrr7
z_`h)jEf27al?`rC9+z`<xg(QZQ=ONav+|sLcmdS)%=j5l&Ce8S)<NUBvT;Uk%xbxl
z<0D`t@pp}1@(TeWyk#|hRj{ISa%1kKC?g_Q<;L>WP4kUW*T=oYyE#)=nA`;K<w#*Y
z(9d}<S!T6G>8nXkh^jg{T#bSHV*L_D-C|JGui0j@;N$SrMt~C%QB=6**ePn!hKG-1
zB?oO!?X>E}O0G~&h?dxH!B69ol`18ikp)d(?ur+tJ#!3-X}YCdvM|zEC=+rt-1(Ji
zrKZO{Rs{!je?~Y?!Jrq`{txjS%f_Qkw)c0?G^R<T%!2vU&>t>|Bg<<5uh#z7&j;Z$
zz4Y5zJYI!zVoWNO3!D9_)EQ3S)#>HVsA{vW`D@Im5a}54Uo{qHPBz!?zzAZCwG6b3
z-Kd*!#Sb}PF;Hx7M2t!5T0-1+QNnitXag(0SF8RdC6Kv^DFHc;NQ#Eh9@}Og2ZwNH
zuWzojv<{5|oidNghBmS#rD?;R^zm^1WhH1iNk{oZ+|Kj@Z=ol7k^oUL{!C9%evL)X
z^Z-3Nz8O&J+_NM(W4D1p(Rc(j*g$tEp>9#?q%Ts&TLc)>`YgqMp8z9!U!qtifh2%N
z2+$A$i)2B5PPXx+akhhabX_?cTU3BllG-r<8?DPf2Hvv}*<!J>`=$(aB@9|2*sEG!
zsb)?*eCO@P$>fZ@5j$^_vyqMSGW}iS*Ze}jS)^j;EfpwO(K#y1Q?JTsOB5Fo6n;Na
zpWFTAQSulX<-9g|3?0coo!^59_~-ndLS!q87C8dJ3+P3VR7!&<j2RV9Lhv2R?H1W)
zbdKG{ax~h&OOT_{9ub`}os>jlh?G-gVFz<Vru@gJ$Tk}gzQ9Uu!Wr+G)x<?B0Z6p8
zxrnZC{-f&7$l9OTBu9GEl0fG~VuO(z8|3LgfWT?bGQS%u=pnunC){X0;+Z}6tw>8t
z*rFt8MacGs4*Z67T0c%5<`g-o(5b`LHklP>r`TGeg<KLA-6q3qG{B!S+a?1$O%1Jh
zI^IiNM?hdq{#n~9v>S8(%6-qhYBRE9ZtL5&_wda!5S@7%eIUR{qurClCqs;92#r}<
zNUFBH9B~dz72S{cLv=O9m`J~qE*T$Qg3NYP#vTAvbV6K=EpMkn3`r#@#<14`ioFfs
zyhVk7ZOfDm@C~~3usWc{HGG;bp~8QJ*!Aw@Xj?hvs#GV8m(sq58!r!;m2l|`k$P-#
z1ecDF-_mBs$&&!NuE~?Q6k8CD5|d!ajctm=v0M3wvsxlM*TtiA)L=<TTw)_6!x{7`
ztRGOn0ZWi3<X(y&ByflT8@N?8W@`v+B(RCVVag@0iW~jilt$q1kff~vc)@kk?e{f&
zQ0w#U{h-G0Yogd)H*37U^}HUHjy39mcw8Cm!q<6a^K*gB(xx}X`6rQEIB|ySP>Y&m
zo;`X{B;g|wJdI5$-3KQV!<dHy71<Y+;SH07NQTc0Cet|m`3#?hUV3oS-0=@FGi4L8
zgX{xX{Al4EU4DbWRRW{~XD?Xt=D(8f;olcNsP4cE;m@U=1NgZ|lB65%D#?3mwM&v-
za1~tjKXz5$^7<wBKezyHH8x87Z@2(%?WvcJN;g~tZ?)A*Pa;n6R!g0<<%SF3){fhx
kwi_;hTlZ{}YHqjyZuLkF(!(f9@YbG<()tfv0IYBS9}d(e%>V!Z