search 041005

Upload: raghad-al-shabani

Post on 23-Feb-2018

214 views

Category:

Documents


0 download

TRANSCRIPT

  • 7/24/2019 Search 041005

    1/39

    1

    Lecture 3

    Uninformed Search

    http://www.darpa.mil/grandchallenge/

    http://www.darpa.mil/grandchallenge/http://www.darpa.mil/grandchallenge/
  • 7/24/2019 Search 041005

    2/39

    2

    Complexity Recap app.!"

    #We often want to characterize algorithms independent of the implementation.

    #This algorithm took 1 hour and 43 seconds on my laptop.Is not very useful, because tomorrow computers are faster.

    #Better is: This algorithm takes O(nlog(n)) time to run and O(n) to storbecause this statement is abstracts away from irrelevant deta

    Timen! " #fn!! means:Timen! $ constant fn! for n%n& for some n&

    'pacen! idem.

    n is some variable which characterizes the size of the problem,e.g. number of data(points, number of dimensions, branching(of search tree, etc.

    #Worst caseanalysis versus average caseanalyis

  • 7/24/2019 Search 041005

    3/39

    3

    Uninformed $earch$trategie$

    Uninformed: %hile $earching you ha&e noclue whether one non'goal $tate i$ (etterthan any other. )our $earch i$ (lind.

    *ariou$ (lind $trategie$: +readth',r$t $earch

    Uniform'co$t $earch

    -epth',r$t $earch terati&e deepening $earch

  • 7/24/2019 Search 041005

    4/39

    +readth',r$t $earch

    0xpand $hallowe$t unexpanded node

    mplementation: fringei$ a ,r$t'in',r$t'out " ueue4

    i.e.4 new $ucce$$or$ go at end of theueue.

    $ ! a goal $tate5

  • 7/24/2019 Search 041005

    5/39

    6

    +readth',r$t $earch

    0xpand $hallowe$t unexpanded node

    mplementation: fringei$ a ueue4 i.e.4 new

    $ucce$$or$ go at end

    0xpand:fringe 7 8+4C9

    $ + a goal $tate5

  • 7/24/2019 Search 041005

    6/39

    +readth',r$t $earch

    0xpand $hallowe$t unexpanded node

    mplementation: fringei$ a ueue4 i.e.4 new $ucce$$or$

    go at end

    0xpand:fringe78C4-409

    $ C a goal $tate5

  • 7/24/2019 Search 041005

    7/39

    ;

    +readth',r$t $earch

    0xpand $hallowe$t unexpanded node

    mplementation: fringei$ a ueue4 i.e.4 new $ucce$$or$

    go at end

    0xpand:fringe78-4044

  • 7/24/2019 Search 041005

    8/39

    =

    0xample

    +S

  • 7/24/2019 Search 041005

    9/39

    >

    ?ropertie$ of (readth',r$t$earch

    Complete5)e$ it alway$ reache$ goal if bi$,nite"

    @ime51+b+b2+b3AB AbdA bd+1-b)" 7 (dA1"

    thi$ i$ the num(er of node$ we generate" Space5O(bd+1)eep$ e&ery node in memory4

    either in fringe or on a path to fringe". ptimal5)e$ if we guarantee that deeper

    $olution$ are le$$ optimal4 e.g. $tep'co$t71".

    Spacei$ the (igger pro(lem more than time"

  • 7/24/2019 Search 041005

    10/39

    1D

    Uniform'co$t $earch

    +readth',r$t i$ only optimal if $tep co$t$ i$ increa$ingwith depth e.g. con$tant". Can we guaranteeoptimality for any $tep co$t5

    Uniform'co$t Search: 0xpand node with

    $malle$t path co$tgn".

  • 7/24/2019 Search 041005

    11/39

    11

    Uniform'co$t $earch

    mplementation: fringe7 ueue ordered (y path co$t0ui&alent to (readth',r$t if all $tep co$t$ all eual.

    Complete5)e$4 if $tep co$t E Fotherwi$e it can get $tuc in in,nite loop$"

    @ime5G of node$ withpath cost H co$t of optimal

    $olution.

    Space5G of node$ on path$ with path co$t H co$t ofoptimal

    $olution.

  • 7/24/2019 Search 041005

    12/39

    12

    -epth',r$t $earch

    0xpand deepestunexpanded node

    mplementation: fringe 7 La$t n ir$t ut L?" ueue4 i.e.4 put

    $ucce$$or$ at front

    $ ! a goal $tate5

  • 7/24/2019 Search 041005

    13/39

    13

    -epth',r$t $earch

    0xpand deepe$t unexpanded node

    mplementation: fringe 7 L ueue4 i.e.4 put $ucce$$or$ at front

    ueue78+4C9

    $ + a goal $tate5

  • 7/24/2019 Search 041005

    14/39

    1

    -epth',r$t $earch

    0xpand deepe$t unexpanded node

    mplementation: fringe 7 L ueue4 i.e.4 put $ucce$$or$ at front

    ueue78-404C9

    $ - 7 goal $tate5

  • 7/24/2019 Search 041005

    15/39

    16

    -epth',r$t $earch

    0xpand deepe$t unexpanded node

    mplementation: fringe 7 L ueue4 i.e.4 put $ucce$$or$ at front

    ueue78I4404C9

    $ I 7 goal $tate5

  • 7/24/2019 Search 041005

    16/39

    1

    -epth',r$t $earch

    0xpand deepe$t unexpanded node

    mplementation: fringe 7 L ueue4 i.e.4 put $ucce$$or$ at front

    ueue78404C9

    $ 7 goal $tate5

  • 7/24/2019 Search 041005

    17/39

    1;

    -epth',r$t $earch

    0xpand deepe$t unexpanded node

    mplementation: fringe 7 L ueue4 i.e.4 put $ucce$$or$ at front

    ueue7804C9

    $ 0 7 goal $tate5

  • 7/24/2019 Search 041005

    18/39

    1=

    -epth',r$t $earch

    0xpand deepe$t unexpanded node

    mplementation: fringe 7 L ueue4 i.e.4 put $ucce$$or$ at front

    ueue78J4K4C9

    $ J 7 goal $tate5

  • 7/24/2019 Search 041005

    19/39

    1>

    -epth',r$t $earch

    0xpand deepe$t unexpanded node

    mplementation: fringe 7 L ueue4 i.e.4 put $ucce$$or$ at front

    ueue78K4C9

    $ K 7 goal $tate5

  • 7/24/2019 Search 041005

    20/39

    2D

    -epth',r$t $earch

    0xpand deepe$t unexpanded node

    mplementation: fringe 7 L ueue4 i.e.4 put $ucce$$or$ at front

    ueue78C9

    $ C 7 goal $tate5

  • 7/24/2019 Search 041005

    21/39

    21

    -epth',r$t $earch

    0xpand deepe$t unexpanded node

    mplementation: fringe 7 L ueue4 i.e.4 put $ucce$$or$ at front

    ueue784

  • 7/24/2019 Search 041005

    22/39

    22

    -epth',r$t $earch

    0xpand deepe$t unexpanded node

    mplementation: fringe 7 L ueue4 i.e.4 put $ucce$$or$ at front

    ueue78L44

  • 7/24/2019 Search 041005

    23/39

    23

    -epth',r$t $earch

    0xpand deepe$t unexpanded node

    mplementation: fringe 7 L ueue4 i.e.4 put $ucce$$or$ at front

    ueue784

  • 7/24/2019 Search 041005

    24/39

    2

    0xample -S

  • 7/24/2019 Search 041005

    25/39

    26

    ?ropertie$ of depth',r$t$earch

    Complete5Mo: fail$ in in,nite'depth $pace$

    Can modify to a&oid repeated $tate$ along path

    @ime5O(bm)with m7maximum depth

    terri(le if mi$ much larger than d (ut if $olution$ are den$e4 may (e much fa$ter than

    (readth',r$t

    Space5O(bm), i.e.4 linear $paceN we only need to

    remem(er a $ingle path A expanded unexplored node$" ptimal5Mo t may ,nd a non'optimal goal ,r$t"

  • 7/24/2019 Search 041005

    26/39

    2

    terati&e deepening $earch

    #@o a&oid the in,nite depth pro(lem of -S4 we candecide to only $earch until depth L4 i.e. we donOt expand (eyond de-epth'Limited Search

    #%hat of $olution i$ deeper than L5 ncrea$e L iterati&ely. terati&e -eepening Search

    #!$ we $hall $ee: thi$ inherit$ the memory ad&antage of -epth'ir$t$earch.

  • 7/24/2019 Search 041005

    27/39

    2;

    terati&e deepening $earchL7D

  • 7/24/2019 Search 041005

    28/39

    2=

    terati&e deepening $earchL71

  • 7/24/2019 Search 041005

    29/39

    2>

    terati&e deepening $earchlL72

  • 7/24/2019 Search 041005

    30/39

    3D

    terati&e deepening $earchlL73

  • 7/24/2019 Search 041005

    31/39

    31

    terati&e deepening $earch

    Mum(er of node$ generated in a depth'limited $earch todepth dwith (ranching factor b:

    NDLS b!+ b1+ b2+ " + bd-2+ bd-1+ bd

    Mum(er of node$ generated in an iterati&e deepening$earch to depth dwith (ranching factor b:M-S7 dA1"(DA d (1A d'1"(2A B A 3(d'2A2(d'1A 1(d

    7

    or b 1!4 d #4 M-LS 7 1 A 1D A 1DD A 14DDD A 1D4DDD A 1DD4DDD 7 1114111 M-S7 A 6D A DD A 34DDD A 2D4DDD A 1DD4DDD 7 12346D M+S 7 ............................................................................................ 7 1411141DD

    1( ) ( )d dO b O b +

    +S

  • 7/24/2019 Search 041005

    32/39

    32

    ?ropertie$ of iterati&e deepening$earch

    Complete5)e$

    @ime5 (d+1)b!+ d b1+ (d-1)b2+ " +

    bd O(bd) Space5O(bd)

    ptimal5)e$4 if $tep co$t 7 1 or

    increa$ing function of depth.

  • 7/24/2019 Search 041005

    33/39

    33

    0xample -S

  • 7/24/2019 Search 041005

    34/39

    3

    +idirectional Search

    dea $imultaneou$ly $earch forward from S and (acward$

    from N 7 324==D $tate$

    S

    B

    C

    S

    B C

    SC B S

    State SpaceExample of a Search Tree

    $u(optimal (ut practical

    optimal (ut memory inecient

  • 7/24/2019 Search 041005

    39/39

    3>

    Summary

    ?ro(lem formulation u$ually reuire$ a($tractingaway real'world detail$ to de,ne a $tate $pace thatcan fea$i(ly (e explored

    *ariety of uninformed $earch $trategie$

    terati&e deepening $earch u$e$ only linear $paceand not much more time than other uninformedalgorithm$

    http://www.c$.rmit.edu.au/!'Search/?roduct/http://aima.c$.(ereley.edu/demo$.html for more demo$"

    http://www.cs.rmit.edu.au/AI-Search/Product/http://aima.cs.berkeley.edu/demos.htmlhttp://aima.cs.berkeley.edu/demos.htmlhttp://www.cs.rmit.edu.au/AI-Search/Product/