Modeling High-Dimensional Index Structures using Sampling Christian A. Lang
Ambuj K. Singh
Department of Computer Science University of California Santa Barbara, CA 93106
Department of Computer Science University of California Santa Barbara, CA 93106
[email protected] [email protected] *"./ (1
02 3 & & ) 4* -4 $ 1 5!4#" $ & 0%" $ 769 &8:,' )).(** (*;+," *$%*" %?(@ H " * & %.5( *#"J$I9 * &%>" K - &&AB" (*7"3 &$ C0)#"$7". -*D%" " *E (&8=M0 * " $(> %?-(C & *4*^ . *-/^4-$ ".G* &$("BaKK1--] " (*$(*1-1]S ]d-" " \c(%&-"^^7"D44 & *$ &K=-" T '*-a]D TB $ &-#" &$ 02 &8 &a% " R5 13 W " ( B & 3" (--$ &6[02$ W Q-%-$$ * #"$.4*4" & " *&8^ &^ 9-ea". =$ ++ . *D =B Wb --b -7" - D[" &(*1 T] =02" (. D-" *4* ^ 1'-B%?" *- *" V T-^ 44* " -Q"O]D " (* U'*f b]_" (*"N 4 8
ABSTRACT
*".@INTRODUCTION g ( 1-02 E * &C h 4*V -D4- $ 1*h E h#"$ 4** % &" 0* *U >" (* U * ?fA_4-- \$w& &. - R}r2:#k-D-k
Ql&`!q2n>41K= ]T$-" .-" 76(o- [o$+ Ul14*=l4" 8--$ " 1DZ9%&6{ & -%. 3 " -} W*D
Q" +,- )2k-n- mXk9l9Ab6G:$U:$Z9 +
1.
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. ACM SIGMOD 2001 May 21-24, Santa Barbara, California, USA Copyright 2001 ACM 1-58113-332-4/01/05 ... 5.00.
]_ &7" #"((1%0C 3*" b" -* yJ/>" (* \ " * ? &f EAT" (*$`c /8N'*V *N%& \ *" ( ?
%1 - U >* & *"1 8 .-769A^" ( "$$6}" E$-0 /%1 >* ?+ $% 1W'1]_& 2-¦§-TK?-¦P?5¢aG-¦T' $'$N9¦K$-K K$>-b-¦§??£>. #1$9?9 ?T¦,3[ D®V-41" ]D(* 1 * &Q02% " 1A* [- ?**-Q* *(]TC-$% .(*`(+, &*%1- E & &&8 *$$>- - \" (#" ;$ %?*" ?* "¯($ -%?z"" *W [".`2 9 D" $ %?&}" 02- 4 .^%$c"[14--( W$ " &(* W&8 =^024uT ? (4%& _%11A¯ W]T >9 6" U(*$ & * /' * V bQ" [$((%& W1-]|4*(* 1=] ++ $ $ 1W@ -"O4*]D4* & & -%.U(YA^1]W- +,}I9-7("O6-(*+,I9"B--#"7"OP6)4* &*"[%$" 1]| 34*f 1U%? &" - @-$6-+ b*%*" $` *9E+§" -(*$
&X- 44 -2W-%.(Y 8N R}} ;4-4* $ C& 9 @"V4"O]D(-$ " & &&8%?)(* I9C & 1_%?'(}\" 4*%? ( ?++ %&I9$ - AD 1@]T '-* &*02 7vc- 4S & 9%&-"#" &' C$65$b"w& )&A_ &#" =".b-$ " " /$" (*w& /&A_:#±- %&-#"7+ * & \ *-$ -< ",]T7" C &&8.U*R5 _$(*¯1b]--" ( 7=" "U N" '(E-#"" %?Q(*" (I9 > &N 1 $ * "O[]D &" ( &
4$7" QB$(-%%&1 &]$$ " &(B]=_"B7" *(b^U*$`9 &+§*4- $ =&0 &a( c8a(CuT4*( _ 1 %?4" - - &% 9%&".*. % ?6 + ]_( G="" ( / >" b 1**%& E" ( /4 1*%" - S%&-#"K
*( *I* &/"V -$ $ 1&E%& G*&8" (* V* 7'%& -4",6G $$-$ H]_4*7" " (!- )4* 1 %" " ( ; HK." &%%?".(-+ K" -(* (* & $(*b +, )-7-"O 6)" ($$( ?< 1" ]_- W\"%?(** (".=I9"[8@ >" (*uz" &G($ b" "b(- Z &p*%b" -" $% E &&%.k($8 *$ &IW*$ -8! UR} E44 & %1=" ]_- " ( uK- b(*RELATED T C %&$
7K-- $%>74"OJ6d 1 E%&E- * 3"%& & 9G"$" .(*°* G " #" 1/02 & $#+ + T X 6I9$ E$6 -4N
_ & &-+§\"$ Y 9 &" AN] T$ &-%?3" " 1-(*) &0 S9 &6 %&-" (*6}%& 3I9=-" X- *" &" (*&8$ GC uT102?(" [.--°-8h E#"\i7 l&`*j *sQC=]_ " ]T - & &f#"-$ >%" 1 " *5" $-- i +,7*l&sY4* & * &$$ & 9 "__%&=".#"_8y:, S &Xc$" bC4-%&4- *1+ A % &$`4*"-$%%& &$$ &V" K -N "O]T `2+,1*]_$` &_ $$- -C"z X=".4* 8_ 1_%$"363" ( >"V- 8;Ci n=sa ?-C0 "$" (* * C6 ?". --$$ 1 *G{#" /$ " (%?6=" $*" " &(U >C * N%&* C*I9 * *>C -; &->* K-*".GX$*46G ?%.vc 9`2" &" %" -&8 [1]_®V- b 1y]T `9024+O -1++ A *l? &+§ &V- 9^" -=I97".*6- $ ]_" &7&8" W( D" ( ( %.(9(" +,*E & &$"\ $ t--8< 7-i"Oq-6 cs^$4]_4 -( &%&$ %. &(d8P 9"\uTb(*b& 6 %&6=".-N$$* ;4* 1" ¯" *N9 &N ;-4*"Q 1(*%?/" -' N & 1$*a7" -a / ;" ((- (*$ + %& *- $ &"1&88 PuT(T$(*-$" %&- [ &" [4=.- "$ %Q* &D*$%&$$ & 2.
uT(* &$ _* &B"$$6;" ;-%&%&- 9"^'-^" ( W * +,* 7' 7"O6[-" (*
2.2 Globally Parametric Models
K°. =#"".%Q3".-%&9_-6>#"V & &$ &c '3--V7"N7"O
Q69+t8)$"$ &b Q-*K- " ]$ $1->3 -"C -" ¯(8S4 U=Ji7.l1.--n=s_%".4* " a ? * $ &A& 9" & 1`2 *5 \$" " (( *
+ -/7*"O6[".]_$ ("1%.83(3uTB%(*&U°- #" 1;" U* T ¯" ]W(* -UO * (#" $ &%? 9" "P 1X*" b &. *-$ * I9", 6 ?U - " 02 & "_$* 1-C E" 7K_"B-]T' N9 +§l?`>V+§V96>hI9La*I9- 4 ? &&8 ;^4D -*
_+tN"$ - & / &&8 " %& Q ? "U*8_#"t$8;uT (i7l& ?o6 +s " V! BI9 ? -. &8B D:O b" \%1-4$ N 1*-Y%"^$4" (%& N VT4LB$H" 7" - 3 :$ *±*hA" %&(-#"P"D &f%.(. * I-* V Q ? + $- 7X" B Z& &$$ &i 9" j-lsB76-0 3 3(*b#" --.* &z^K-=\024 ?^ 1" *(* _%" *=".b-" $ (*"1 8B4uT ?(*$' &- *b *%&
]_" -(c( C.%.(b 9%&- QPTI9 &$ ? %&$- 7P" 76 3 &-04$% .*B]$" $ D-1 G- _`96E *" D$4(=w" *J`;¯= *U- "."U-4-- 83$ U&i &8=" s§ 8 % &*uT4#" & (* X *&A &Y 9*-"N0- -- 9B".$" %1#"] _ 9%&"$%& *.*%%6 1( d _$ -&%&$%-$" #"*W f- &ANc3.]D- X * $( V* 1-] 3(7 *" 1 1 1] @ -#"-T- 4* 9- *"(-N-%&W- hI9 &* E -$ &69&8$ 8X^1uTSuT((*"
* "_$ &%" ]_¯ 1-c]_7" (/" ( V7" 1 &z02-> 1*}D9 6)=*°" & **#" [P" 02-[ 4$4 1 X *%-"PP" (*" " E%&#"V &$- $6T" ( \ G" 1(*= 4 -`2- ^ &_--4*z " 1(*%" O [ -7+§%&% *. - %?69A= "zz" ( V4I-*$". $ 6"c]T" _ &` *$* c a8D" *(-=T"Y" " (*(*
uT(* *W+§#" $ %- " (* U-(- &7" -((x"N-'a-_" (*" U(* V"$ & 8 7P+§6> * ?#+ N Ub%&" ( N $ KAWf" ( *b -c$A9$]D* U &%1`- / -%.1(]_ & 102* N-" G(-WU#"" $(* %" .- c$#"$ J=%" 7"O 69 8 " " ?(* &&A^ 1- 1U4c-8/ :, & 5$4$ &" & 9(%" 3 1°G%1K-%&-7=6 ]_G¬a * * CA^]TZ4 447'"9 1%&
* " +ti ?+t"$Uj- q- "st$ ?8) +,®V%?`2" 31*]T#" $ & V02* -%1+ A ]D &Z8 %& D" (* D 7+§ ;a7"B CN$*$ "B" ( D".$ "1A]D
(7Y1]T02 V V" *C$ U 1W[%& N " ( U44 -l \l&%&o-4-f%&" (7"O N6/-- %&%- .*-f *=769".8PA 1--_%.(< *4-- V4 A " 4(*- C " &8U 7N+§U G /" ]_( BN%&]T- 9.".-&A )c" ( X ;1402 . [--%1 /-4l -%&l17"Oo/6PN" (* 1 C**%& 1- 9-6 " (*- -*%? " -3l 1l1 &o&A8 " Z( U *%& 1 b
Q%&+t"$ & > & 2- W *-w&B /44-- 9" " _4[ Q1b *b]_ %&(-102$
U#"$" (* %?" *=.X-c" $$( * 7`"O698/AuT" ((* V4$(-*-= U]_&;"69 (=" "[W &-02¯ &" 5(* V" (*'**c() ]D
¯>-" (** U l; *7 +§4 *%" ?_"< ( \ \%& - "Q % &" %& $ 7",6/]T 1 &I9 /-" (8 \ - -¯- " $ (*?" Tz ¯--*4* T b4l2 TQ*& &694*%"1" 8BQ"¯ (* V*l2 t/a#"$$( *1%]_" B" [(* D% #"-$ *-%" 1G=". + " ¯(*- _*- 3l2 tTf* &".4*%$ " W"" ¯( U [ 7l+§ %=^ $(*=%&]_-^" #" $( _%? -" 1>4*- 1 /" (= U". -$ 3" + 4* 1* b$(=".- $ *`"18^Q /D-$ %&&%& &- 9 3"_" -(*z Q - ?4*4 ** 8PA2uT" (*( _ 4 ?-'-- & D -A*]Dc" U(( _102 V " 7 + 4 -1-]} " J&(69 &-!*"\$6U ;.£" V>ª*" ?(* C?-¦ £--Q ©&-2P¦§ £- *" ?_f8bV`2 ¯%&" [(" a( 3 &$*" *J *= + " 96) %&* 9*" / */V " &(#" .- **C$(* 1 f\A]D_ C 4*" ? & *$ &%%"Q" " ( * [ 4",]T &% &" 1 } :#±4~ &%%" 1#" I9* $6b & *V (*fA]_( ?W*%14*- * 1$ Q%(*" qE¯*". $
\n"/A]_%?O(*8G uz N-]D N G.-l= /\m- o-5oCn9%&l-+§V4{= I9 1) " ( & bP b%" " ( 4 - 4*- N%&%$ &w&$ &$ &8&PV]Q_T 7" (4" ( & ;%" 4 1c 1A2*" (% Q" 14* &4*--% " 4*$ 9 " \ &4* u_ $ ?&8 9" */" " * ' 1=" * 02 &%?" - ;".- 1XK " (* bD- D:Ob
nkk-A$rm o-A q2oE p pm3l=*pk1u_o+,*+,?8 & $$- --D44 9 9" " ;T &4*4 & &$ $ && 9 9" " ;K" 1 " " * & _['$ 1=%." *(@ U-02 $4%" &-% "$T.-P %&Y9- F% "_ " b&AB-%- & 9U" '"$. K# 1'" & 1&A -p A m-;o-o\$-j p-o.+,-* 9"a' & 1 *=$" *- &z4*". 9" a 1; &94*6U $
&&%& 9" .* *V " _(" D(4* T m%&nW W ?-"$" p A m-ao--o;#"" (9 D%.-`B4(0- B "B * T¯69$4 1_` &'"$ .9 6#l=K-m- oQ$4 1C 1-*`$ &8 au_?8
N =
j
j
!
'7.'7,. ., ,. +
37% 24% 14% 6% 1% 10%
'&N!
25%
)(
Sample size
+* +-,
)'2
3 5%
5% 9 ,.' /
50%
).
01) 3, ,
K" -(*V & 02- & /4*" (* Q"$ w& & b%.( **I* U" " " ( ; 4$ K 8_9%&N% *-4" (* %Q6)%&- *W#"" $(*. - 9"U 7 + 4 -* c"1A-8^" Z( T *%& QU44 D. _=" - Wc%1" (- W &02 * ?+§B T $[bC-* #"BP" (*( >l "P 1A-#"]_D(* *
W" ( \4- \%&-4%&7"O6>-¯" (* V'*z c8 * *N* % & "1U;A]T" ( 3( Q4-]_(G &($
-&$ 9 V-"\?-(*
*(*(G* %&3%&*" .(* \%6) '*"O ?64%1 7*6(*:# P±K -= ]_/ *4*\ &$ *&%%" " * /B-*%&$%%&&.$-Y"% 69(8 D * ] 44* -%.( [- 1=" 4
_ &+t &4*"$%& " 1 &(C &B]- 1 /_C(96]D N]_f*$ N" (* V#6C-P#" & 4.1 The On-Disk Index Tree 9i m16bs§uT8>" (*(uT E N(*[* *+,`b-$`-2 -7*" ( 3 d- &"$% * G$" (*02/ &7-6X* y &4$%= $ *" " d 1-< #"$\"9 6 (%D" * %. G("." - &$4 */ -. %"Q b" - &" 8 - & ?." U" ( V * b"$ & V4-- &Q 02 &7+t]_$ -8D*-Q 1-%.(/ &0 &A*" (*
1 B 304#576 398 3:7@ 3C D<E >F@ G :7@ @ G:IHKJML#N O=JFV 8 LQW >F@
**CC DD--^B4=". 9C" 4T" ( 9=" "_°"_ &$6 (** &C(9"_ D-a-B" (*4 U 9 " +,W4*$` bN* *$`b b4"$ &
(*(* & &(9(9"_"_--aa" " (*(* U U14*]T4 QW"$"$ & & & BA 44 [[.." " 33''QQ" " (*(* \ \=4*]D4 _W"$"$ & & &
bb==x|*=7 ". &C%" 4$6 \4%14 [-%%1-",465-%&-7", *6}+,*$` b+,* $`> +
? ?vcvc & &%% " " 0202 \ }*7 ". &C%" 4--$6H \4%&--4- }%&7%1"O-6)4-%& 7",+O6g$`-%.(".4X=$" 34" -= $" " _"- (* 1 & 'T$ " 1(- %.¯($ &f8P4uT($" U7" 4*$ & * 1\ ;%" (* P \ --Y" ( B_-$(-- 7" ( 9 " 2%1" ( / ^ & VK &*0 & * S $%" ( >4*" ' fQA02% , 8N$i m1 Ss'?8 -U" (*b4-4 Xi n-j=s3''>5- G ".- 1 $4Z4*N7"1";U8W
U>uT?(*$4+t"$Q 7 &"$ " &$ i*j-7q" s§N#"8#" *Xb%&-*" 4*E3 " -" (*( ; F>)V 8 LQW >F@ §n ]_(N ? 7" *" 02- &" 769 _"A2" (*(* N U$ & %.(b .W--*aI%&*- $6b4 N 9-" *&8".- * 13K ~" (
" " ((- V="P 4*-" (* 8a _4¯$ 1 V7 #". B=%." ( *.48^ *:O >*T &** 9_9" ? z*D -4 ? " "D(*$
W W & ~ " &A¯" \V(*1
N]T Np- &qN02vc ? 1=%A*"." ]D- 1$ 3 U"BC$($ **1%?b1(C]T" 1(6
'*z*".-$ ?"N" 3%4" N" (* U$ 1 %.(/.-*8 'V:O G*" (* *' -[1" (]_ \ G1]T$ *_"$$ & &%? " &8- &Af]T C*$%&$U" (* ",]T/0- " 4.3 The Cutoff Index Tree %& 1:O" f G-4vd-.-* 4* D4*V 2" 7" (%&g $".- " );4%&%&4- ?\ 9"$"1 8B & 3uz\ 1%&-T 4#"-$ -* %Az"B]D" b( W-$1$]T* P 3"$ & * WK- & ** ". -4* #"-$ Q - &f-" * `C"$$ 6$ " - 5 &A" a(* b" ( 44 $++ uT(*%" Q 1-*%&W"%& .-" %(*6> V*%& ?-" %& *-.$-" / _" K(-=_"Q(*-4*(* &76b *$7'- -f7",6=".+§-$ -$ " 1&8 4 1:, /*%" .*- _" 3 &*#" &.cQ $(-G"_C*%&$ -'#"*_Y'- (*(*+,K-Q" ( $_- 4*-4*c 2$4%?(f%& A*8 ]D
" #"#". ( =4G" "D UjC" (%17 V6C]_#"$ &(* &Q4CTn ;;%&---#" V- 7" j\(M 9"$ /*4*$ %& Q:#±CH%&%&#" U18^ uT¯(* N$%1 [ m *8^ *V;"
Z\[^] 8I-J syu v g
Z
8UT >F>FV
aLK
BHM
F
8 L-W >7@ |
uT(* K- A" (* U=02 .Y%&#"Q¯" (Q-4*4* 2%?(/
t j 4.4 The Resampled Index Tree K - P$4uT 1(*- \ %.( & 4*b-4 =4*a]D &"$ /U & _"$ &* 1 - ?8V>c4uT"$- (* &- ; N 1C4*Q"$
&& QB" *(-* &7" 4*" U1*a4=".-3" 9(4" WN4% 9- " $+ *14]T4 ? \a"$"$ & A8B &$*B " * *->4 X A-7(]D(* W (\1 02- _moN4* *4*4 .a" "$ ; &" T(- 1 G=cK4-U" &(*&A
]D * W% ]T 1-*$ TC" (*( P102 W- 476 D$1 w& Wmm-oUoQ4" 9 &" &A^4 1 ?P%?(1]T=]D ?D a"$ "$ & &8¯ ^:,]T]D* T[]T(=*0
" 14-]T4 ? &8^=P"$V b &%¯ Q" ¯"" 6(* ( Q1 & C &4 m-o=+t 9'4" &8cB AuT; &(*$W T** " -vc * &% V" A9]D02 D N %&-- < T4%& %&_*3..= 1" " T%? T( ?["$$ $ > G G4* 1 =" &(V% &4$ -6c" 3X]" T((*1 ? \] >=]D%1--* + "." V 1 WG$b$- * bPb" ($-b [" (z W -- 4*- A-*" (4 ;--4 &-P ["B " (" W & 9 " N T ; &02 &4- &%" " (* 1
"$]_ _ 8"CuT(*W 9 1" X \"- 9 6@(=4"[-$- 8@ N-W*" 7" (* - ]_7697AP6)E - 4*02 > &5$4*" 9-" ]D 1=*BJ43 1 &I9_]*_ ( N%.-(/PbN-- 96C0-".$76( b"O" ]_(* )%& [CK- ?=;>]_-]W *Q S7"$4*" & 14 GI9 7 * "$ & %& G
#"EC4*-C$ ;&%&'#";_" K%&-(* 3*C N%&* >" -¯v'Q " *(* \ ?N)k8P"$4 Z9 &4" - &8G24-WnZ9%.[(Y" &A4}]T p
44 9 ?" $'-K - &5" -(; '- 1-]_ >-V%O8/=".¯)°* G b 9r" ?58bR} & 3 1$-69 8H Z *%&G /C D<E ]T>7 @ (%.(1*02 b`>Dz &$w&- $16E$w" b -4/1 ?8^A¯uT]D( 3 1K-) " A(*" ( &$
Qb -4* D C4¯%.( 9" ` G C D<E >F@ 8 uzG 1-- /%.( *`cAa]D /(=0 b" X$%1 d=02 1 C D<E >F@ 4- " &A &$7" \ bU%&-#"T- 8 T >F>)V 8 LQW >F@ 4 QG%.5(* 1 `3%?(K-%?_(" ( * U`A[" *(* *1 #" &44Y8 9" X(102 }" h_
*#"$ " 1 - b-$ X$" " 1(5 9 6E" (= &&81{Euz" )( 6X 1**%&P / ." - *-- X" $ (& & & 5`&A-WT" (4 /4 9" ; 9-" . * V 31b#" = & 4< N]W? 8B7"$Z" & b%& Q" C]T U($`C102 b- $ V*%?%.(7@ 8 FK>) V 13K -x8 LQW &>F@ 8 $6;" ;*$`A2" (T#"$ **" uT(* U]_(- ;#" &4/p" (_ ? ' [%#" F
a
Z ]
F
^
Zx[y] 8 g v d b-s Z K 1 F G C D<E >7@ M F 8=T >F>)V a _ _ba c ] d egfhji k F 8 L-W >7@ a F 8UT >F>FV a Z ] _ ^ F 8 LQW >F@ x| q2 uT(* 9-4) E * Cr3 1-* %?( *`NP$w& C-4*4* 1b=" &76 1
9" 3 &$69A9 &$7" C C%&#"Q Z\[^] 8U_a`cbQd h gqf ` fUgqg v 8UT >F>)V 1 8 L-W >7@ x| ]_P7" uT(/(** \" (*=\ \02" (.*- _`f1%&]T #"*NP"$- * zb &" _-( V & * " ;(* l&o\4* 1P/$%*&- $*$
_ &b &/ 9"$" E 7 & & \Z7 &63%?W"" -( &q* ?'8-l- 8 $6
I
Z
F
a
K
B
M
F
d
Z\[^] 8 g v d g
Zx[y] 8 g ` gqf b-sFv a Zx[y] 8 I-J syd u v g a d Zx[y] 8 g v d b-s a Zx[y] 8U_a`cbQd h gqf I ` fUgqg v | §m- d 3a:7@ 4.5 The Choice of D- #"$. 9" ; E" ( %?(*%& 3 3 :7@ = $ 3* [" >",]T> &+ Z
"$$ (*-%& U& X 1f°BAa #" &10A_ [ &c" 4/- 4*- " & &};* % /" $-*. ) * @W$ " -53[-%& -"Q $CC* * - . *=+t" "$%1 40&8J%&c7"OR546d- 3-- =?$%AnX-K -$
Z
F
F
3 b-s N ` gqf Z 3 yZ m m Z 578 304#576 398 U n U F G C D<E >7@ U o2"!yn# U A g b X 8 ]_4( ? 9 " W%Z &m - m ".Z578 1 &8 ]Tc ;ATuT-(( 1 027 6 ? \'-" *( 3 - /AU1*" ]T4( 4 XN C( &"$* - & ( /"/-b%- /[&" " ((
+ ( -(9"N¯" ( V44 ?W"$ & 8 4.5.1.2 :4O b &W._* =P"WUpper " 1- " *N$n9*ABound ]D _ V" (%1="D >on" &(I Q*3 %1:7 @ % ."," 6b76b- &f *" $(* Q _4*" (4= "_D["$ "$ & Q & N 1- 'n[*4^-(* &" _(9"; G 02:F30@ 4# 576 391!8 N3 4:7" @ \#" l8P &uT;(* "; ' 1 #" A " ( V44 ?_ >- /" ( U(* &(9"Q-¯" (* U4*4 W"$ & \ bg =b 3 3 N ` #gqf A Z m m Z 578 304#576 398 304#576 398 3 n G :q;=;?>F@ o2 Sn y| $ -&702"O ?69 &.AV-4f- 0S** *" &(>W" E- X3" &%.(*( C9 $$ 6S\J$ w& VAN8 " W( &*$ E".>-7" & &A>cb" (*& 36S*#" &U *$ 1102 * + 4.5.2 Constraints due to Prediction Error " ( ?:, 3='" :F@GfC4^D<E-$>Fb @ " (Pt--A-U"Y" ( / T * V1 3]T1]T ?$^b "$T- "$ ^ & B%- V\]&T b 9 ".-^ & 5b-$-.69 9*8¯6; uT;4(" T 9" %&&A%&9" -*3 + 1 * N" #" " b9=" $b }-¯W1" ]T( ? V4"$ & [ b 1%&%&^ &$4$ &&8 &EVV- 5/" " (( " K(- [ ;(" 9"1AP"$" - b" ( " ( / >:#±" X% %#" < &--$ -46 1*7+O%" -° d * 1 $ b44 * ?+ "$ \ &B4 &&8 4.6 I/O Cost of the Predictions K P :, &- > 9B"V" -(*4*W4*".$ 2$ %" %?"N(*- & Y Q4A*/*]T U &$
&]_& 9-" ¯ 1$6=1"C5m0-A9--]D ) Q" ]_=(*" 3 :$Z±- &g%" %&- #"\4.5.2 ? 8 B " * ( %*" -vd44* -%.(@[ * &4 & * & 9";W" ( %.(%& b 3 :F@ 8Q 4 ; 1-*4%$4*" b- 1- /cA I9 1-$*w", &6 &Ab^T 2-1";]T >" (*1" 8^ 3]_C(W1]T 89 ;"$ &1^ 7@ #l *x | Ry1]T" ?(/D"$ * &% & ^ 1$ * " (*3 :7@ G C D<E >F@ ]_(*%?(W *(*6
100
10
1 0
,
100
% 0 (
200
300
400 500 600 data dimensionality
* + -,[ 5 N
,.
700
+ '7.'
800
900
1000
+ / , ')
_] 4* 74" (@- &%.%(>" &%fW A 17 E$]W -&6EaW%1 &$ & NN"-- .$(*6E ; $W:$w&±- -z8Gb%&®V---#"1; ]T7* " 02 & %?1 _ 1A¯-'" -$ (*#"& 3 N W &" (- *-- b" 4*" * &(} Q%1--4 76 ++ -*B$`uTbX(- P4*- *4*. " 4-*([%. Q(5 -¯#" X ?1" 8 (* V Cl1%&o_%&" -v44 4* & Y" -(%. D(5:#±\d%4E#"" ¯>K-",^]T*>7vc- .* ? " $= ".%&\ Q-*" ( _4 &- * $" -C - ^7}" -Y &%?&84(*^®V$ 9b ?" D]" (TA2 b Q]T W%1-°* & b* "- $6)" ( V$ Q bw& >=".- &-1$ -"P$$6Cw&* N &4"P b N *
]_ 02uTz ( $
> 3&7:$6Xl1±-o A o Xo%&ob" -(#"b 2-* & /4 K - V*$po3 - * 17"O69 &87 6 ^$-6E *&8" $(* //" (* &U '$- -7"O6 A 1 " -3(: &02- D4%.Q(4 & 2&8 ?[-"$%." uT(/ a(*4* _ # &*2*-% /4X" * f";8¯ Q:," 3]T* P"\*`2r7 &vco &- 94" 4 2 &/%?(G 1$ % U1A]T]D [ [-*$ $* $ + 4(* ?$C' &%'"3P$4 &-7" b**".* 8B *ZE &%" " (*- / *%" 9" 1 & " (* C & 9" %]" T ?B X*%C Q%" T &$ S& &` yV-" (* EX=".".3$ "$" .4H- E b-*%&4*$`%&> & Q$$ I9$4 ? " P"[( ;I9;**%1 $-697` + 8 " *uT(*$ (*; /e^A o-p]_oo>o8x_ ?uT3(*7 "+ " " (*(* & _ 4* 1 C%?" " >- 3$4- %" C(* &C#"$ $ %1" C- *&8 */;uT]_(*7 C" (3° " #"(*; W%4. X " *&8^%1=uT" (* &
J=*" 023 U"$ $* " (f_8aDV & 1-" $02 Q 1 "$.- *#' Q." \^n-o W±=&8
2500
2000
prediction
Z 4*V 7 "b>3
#"$4 +t"$ 1 %* &" -%8y" &&8:, D"3uT" &(%.(N I9-%?*"< 7i" j2[0m=stA¯=" (*J 9" Z>
Q+t
"$ Y+ & >t"$ i n &o stAaA¯
U" (*?+§ "$ =+t?"$A¯ &" ( i7lQl?s§+ A " (* \Q`9 )+O ?+ P+t"$ 4 & C ;i'n--=[stA- ) > *" ( E V-#"$ >*%°" * i n k1-st8 "U%& 9".- * 1X X" ( 4 --* &4 4#"" - (- * b"[)"$ *.- 1 $ #7K 6@ 0- [" DK--4 S[" (C" < -(-%= *%&"C+, &*$" $ (*$`&
* 3I* &T]T ? ;. 8
5.1 Runtime of the Predictions
% ?$ :,- C 4b" =( - 1^3@$ E&" %)$" -" ( 3 1 >]D _ +O*C-$`@ ? T" *E" (*c N 8S%&u¯%&%&-*" . /- G%*6Cj) - #"" (" W 4(@ > 1=* 02 %J " = &" -#" 027 + b=" &8VNN*$%&*$$ 1E XZ &%" Eq8 m3'V" (* [ & 4 1G4*+
5.2 Accuracy of the Predictions
0 0
500
1000
1500
2000
on-disk
&
1
3000
*e# ,.' /01) 2+ + %$ '7+ 3:7@ Z q
3 N*# )'7 b+ '7'/ Z + '7.'7,. 5
2500
Q00
*4$b --%.Y(YA¯4]D4 b ?Q"$ & V &#" b=02" " (#" b =" \7"N ;K--U_4= - [ .=" & &¯"- $(*= "a Q" '( TQ=" (*02 ? U.I9-* &",=6b" -0¯ W 4*$ - &P*-%" - W f8^B ¯-"+ *3 :< b;=;?>7l=@ n3$(*q*=8^]_NN"P] (T \ U %1 & >-$ 4* N 1KE --|44" (2W%?(G-]_-.7" ('1 A" ( UlA oo-*o/ "D - %&$Z & *%& " ]T E )- &$%$ * 1$
G&N$ 7K-(9 " 76/7"O]_6d7" -(/\*" (* &% ) &*-$". 5]_(* & & y$ 698 " .. uT" " (*N3 P-]_'-4*(*%4*"Y %.2" (/(%?Q(*"Y "" Dl( &c o ^]To H ¯Y%1 -6V & ; 1$]x B - l&ao4o "ab8 *2 1]_\(9 &6V U" " ((* ^ P14*]T4 ¯Y --44
Z
k
Z
k
Z
Z
Z
Z
k
k
Q
&8P ?$o - pl-A k-La-r - ;$ j& &m-`oAl1mn m-o-LaoA-n--j \n "$.- j#mKl- A moo $: ±-%&q A#"q2[po8 l1.j V *+,*$` " (
N:< &;= ;x>F@ 4 1o%X| oj 3 p-:<j;=A ;x>F@ C D<E >7@n9A o |l1o-r 2j n pr-r l1rA m-n-r &l q | nl G
N & 4 1X 3 :F@ jA j lA ojm jjA r-o2m n-%j | rkn G
N:< &;= ;x>F@ 4 1o%X| oj 3 p-:<j;=A ;xG>F@ C D<E >7@qA l= l=k mA o2k9l j2kA n=qq p2m | p-or G :F@ %o | ojp-jA G C D<E >7@ l= DD**" " --vv 33:7@@ jnAA G :F@@ o o || ooj-j-ppj2j2 p-nqk m-m-ooll rrAA kko2o2mm r%r%|| qq 22nn G l1p D*" -v 3 :7@ q*A G :F@ o | oj-pj2 m-ol rA ko2m r%| q 2n k
Z
Z
Z
Z
Z
Z
a k
Z
k
a
Z
Z
_
k
Z
Z
_
k
Z
Z
X9!
'71)
a
k
_
Z
Z
a
k
)(
_
2)'7
0] ,.* 1
5 '+
" %(** C" -*v4-44* 4*Q "$2 &%? C(fA 1]T=T V4 ?- \- 13)(*=*]d ]T &" (" [( Q44* &*$4%" 7" \ *D" G *" (*
-** "37'K -b7*6} 7'#" $ **=" 1".58 4R}- Y 8P "uT$ (* U%" 1->" NC(" (* V T &'-_" - -z4 [* &" c &%?$(*C 'I918 &P]_(%?( 4*-Q".- &2$ D"N fa"-AT (]T< T >C.-*-$ HP`moo@n9^ 1laQ 1" B(*4 -- * K-%& %& &$$ Q]*T -tNpAaR5r &l-8 [ "C-8 i jj=sW4* 1%" " -( 9]W"P'.-V/l&q[V$447"_ &*]_ & *V$- E&A2%&J %& &|$$ 1c8".:,$ d"_" (* EuW e $VuT -V7
N"O6ePqp-oo *=%& %&". & $$ _$ &"11EAz02 ?]D]_. 3-B( Q1402 [ 1 Q %?76@(*" 1 &l&G 9j/"? &$4?02A2 &$ G7U"[ &" *(* Q 4 &1- *8;$ uT-"^ (]_3 ?( t ?' V G Ac" (]T4 C- ?--' &*-T ". 3=
" r &(A p--q*$"Bl[4 b4-'- TD " N=( WP(Q"% (* 1 U( ? &b 9 "C-%1[--* 4$ U" )" (*( /- V%&4*4%& - &$-$ * \&%%1" -4 -%%& &7$("O-6 &T`\'
Q TK 1b * & ? & ]W_6[-'Kc Q*A-" 7®V(*"$' &-7`-Q-*%& ;$%& ? $($ & 9-" " -&8#+&8 9.i7l?sZcREFERENCES Z &8N &%%." (0 7$"O696 ?1 A* #" Q%&%& &$ " (!'V 4¦ - 9 " £-Q1©-B -*4*a ` 1 1%]_"U" 1( 4A" ( ¯¯"%&%& &($ B$ &°\ #"Ka-;-4(4- 2(*+,*%?(;" &( *=$"¯- $ =V".*8 W" %?(* I9 V(-W" ( UK1]_ *"O ]T$ -- _Ac- = .- 9" & &TK- * #" ¦§2 ¨ %?-N" 0-- ¡*. &¦,&8?:, A 4 $£1 &Vn -p-r #n k-¡VAfn-ª o-ao£-o 8 ?' ª¥?;£O© i7l&os[L¯)-¦K8 D %&.&% ¦ #b¥-- t - 8La-¦§=" ¨. && 8 D £-*1`"18$ N1 / %.-(Y484uz 1 &*%.(b *=%1" --f /
N 4--$ 1$"Vn*qA ". i jj=s;
[eD- Z8L^R}76*
Q $:#Nuy 1-4A ®[> O8 4+ & %8$"NKZ-®U%? (*eBb &
-` * A%&eP \ ZX/#" Zc *8 *6bD8'--l&"$Qq"1$l=8?{AVI9% " -7"O 96" +§ 7$ ".;1" l& 02%.2 ( k98 ¦K " ( *¦ W >£1(© a£-(- +,*,? ? &¡ $Y-- t9- f$4 -¦§%& &&8. -:O ? A $4£1. $ &-'9-[£#© i j-qs Nl1-8q; n8o2" Rhm(*Aa(l&7_" ;--r -8 / >4
; $8'2 -b Y-8 *Z%& 8*:O 7"O $6
.>
@
j
>
e>
.>
j
?>
>
e>
e>
>
.>
>
e>
>
E@
j
>
>
e>
e>
>
>
N>
j
>
@
>