asa-readme.ps

simulated annealing code ASA

(ampling algorithm\), by e)-3.324 F .574
(xamining just the cost function, lik)-.165 F(ely)-.11 E .851
(will not be as producti)72 750 R 1.181 -.165(ve a)-.275 H 3.601(sg).165
G .851
(enerating more intermediate printout, e.g., setting ASA_PRINT_MORE to)
-3.601 F(TR)72 762 Q .415(UE, and looking at this output as a \231gre)
-.44 F 3.165(yb)-.165 G .415
(ox\232 of insight into your optimization problem.)-3.165 F(Lar)5.915 E
.416(ger f)-.198 F(iles)-.22 E 2.75(-3-)296.837 786 S 0 Cg EP
%%Page: 4 10
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(Adapti)72 75 Q .33 -.165(ve S)-.275 H
(imulated Annealing \(ASA\)).165 E(Lester Ingber)241.147 E .782
(with more information is pro)72 138 R .782
(vided by setting ASA_PIPE_FILE to TR)-.165 F 3.532(UE. T)-.44 F .782
(reat the output of ASA as a)-.385 F .211
(simulation in the ASA parameter space, which usually is quite a dif)72
150 R .211(ferent space than the v)-.275 F .212(ariable space of)-.275 F
(your system.)72 162 Q -.165(Fo)99.5 177.6 S 3.782(re).165 G 1.031
(xample, you should be able to see where and ho)-3.947 F 3.781(wy)-.275
G 1.031(our solution might be getting stuck in a)-3.781 F .162
(local minima for a v)72 189.6 R .162
(ery long time, or where the last sa)-.165 F -.165(ve)-.22 G 2.913(ds)
.165 G .163(tate is still fluctuating across a wide portion of)-2.913 F
1.434(your state space.)72 201.6 R 1.434(These observ)6.934 F 1.433
(ations should suggest ho)-.275 F 4.183(wy)-.275 G 1.433
(ou might try speeding up or slo)-4.183 F 1.433(wing do)-.275 F(wn)-.275
E 2.241(annealing/quenching of the parameter space and/or tightening or\
 loosening the acceptance criteria at)72 213.6 R(dif)72 225.6 Q .483(fe\
rent stages by modifying the OPTIONS, e.g., starting with the OPTIONS t\
hat can be easily adjusted)-.275 F(using the asa_opt f)72 237.6 Q(ile.)
-.22 E .105(The ASA\255NO)99.5 253.2 R .105(TES f)-.44 F .105
(ile that comes with the ASA code pro)-.22 F .105
(vides some guidelines for tuning that may)-.165 F(pro)72 265.2 Q .656
(vide some insights, especially the section Some T)-.165 F .655
(uning Guidelines.)-.495 F .655(An especially important guide is)6.155 F
3.053(to e)72 277.2 R 3.053(xamine the output of ASA at se)-.165 F -.165
(ve)-.275 G 3.053
(ral stages of sampling, to see if changes in parameter and).165 F 1.433
(temperatures are reasonably correlated to changes in the cost function\
.)72 289.2 R 1.433(Examples of useful OPTIONS)6.933 F 9.216
(that often gi)72 301.2 R 9.546 -.165(ve q)-.275 H 9.217
(uick changes in tuning in some \231to).165 F 9.217
(y\232 problems are in the f)-.11 F(ile)-.22 E(http://www)72 313.2 Q
(.ingber)-.715 E(.com/asa_e)-.605 E 1.132(xamples.txt under WWW)-.165 F
6.632(.S)-1.012 G 1.132(ome of the reprint f)-6.632 F 1.131
(iles of published papers in)-.22 F .594(the ingber)72 325.2 R .594
(.com pro)-.605 F .594(vide other e)-.165 F .594
(xamples in harder systems, and perhaps you might f)-.165 F .595
(ind some e)-.22 F .595(xamples of)-.165 F .815
(harder systems using ASA similar to your o)72 337.2 R .814
(wn in http://www)-.275 F(.ingber)-.715 E .814
(.com/asa_papers.html under WWW)-.605 F(.)-1.012 E(This is the best w)72
349.2 Q(ay to add some Art to the Science of annealing.)-.11 E .432
(While the upside of using ASA is that is has man)99.5 364.8 R 3.182(yO)
-.165 G .432(PTIONS a)-3.182 F -.275(va)-.22 G .432
(ilable for tuning, deri).275 F -.165(ve)-.275 G 3.182(di).165 G 3.182
(nl)-3.182 G(ar)-3.182 E(ge)-.198 E .752(part from feedback from man)72
376.8 R 3.501(yu)-.165 G .751(sers o)-3.501 F -.165(ve)-.165 G 3.501(rm)
.165 G(an)-3.501 E 3.501(yy)-.165 G .751(ears, making it e)-3.501 F .751
(xtremely rob)-.165 F .751(ust across man)-.22 F 3.501(ys)-.165 G
(ystems,)-3.501 E .096(the do)72 388.8 R .096
(wnside is that the learning curv)-.275 F 2.846(ec)-.165 G .096
(an be steep especially if the def)-2.846 F .097
(ault settings or simple tweaking in)-.11 F 1.285(asa_opt do not w)72
400.8 R 1.285(ork v)-.11 F 1.285(ery well for your particular system, a\
nd you then must turn to using more ASA)-.165 F 3.558(OPTIONS. Most)72
412.8 R .808(of these OPTIONS ha)3.558 F 1.138 -.165(ve u)-.22 H .808
(seful guides in the ASA_TEMPLA).165 F .808(TEs in asa_usr)-1.221 F .808
(.c, as well)-.605 F 1.246(as being documented here.)72 424.8 R 1.246
(If you really get stuck, you may consider w)6.746 F 1.246
(orking with someone else who)-.11 F
(already has climbed this learning curv)72 436.8 Q 2.75(ea)-.165 G
(nd whose e)-2.75 E(xperience might of)-.165 E(fer quick help.)-.275 E
/F1 11/Times-Bold@0 SF 2.75(4. A)72 460.8 R -.11(va)-1.1 G
(ilability of ASA Code).11 E 2.75(4.1. ingber)72 488.4 R(.com)-1.1 E F0
1.702(The latest Adapti)99.5 504 R 2.032 -.165(ve S)-.275 H 1.702(imula\
ted Annealing \(ASA\) code and some related papers can be accessed).165
F 3.934(from the home page http://www)72 516 R(.ingber)-.715 E 3.934
(.com/ under WWW)-.605 F 6.684(,o)-1.012 G 6.684(rr)-6.684 G(etrie)
-6.684 E -.165(ve)-.275 G 6.684(dv).165 G 3.934(ia anon)-6.684 F 3.933
(ymous ftp from)-.165 F(ftp.ingber)72 528 Q(.com.)-.605 E(Interacti)99.5
543.6 Q -.165(ve)-.275 G(ly [brack).165 E(ets signify machine prompts]:)
-.11 E([your_machine%] ftp ftp.ingber)116 555.6 Q(.com)-.605 E
([Name \(...\):] anon)116 567.6 Q(ymous)-.165 E([P)116 579.6 Q(assw)
-.165 E(ord:] your_e\255mail_address)-.11 E([ftp>] binary)116 591.6 Q
([ftp>] ls)116 603.6 Q([ftp>] get f)116 615.6 Q(ile_of_interest)-.22 E
([ftp>] quit)116 627.6 Q .959(The home page http://www)99.5 643.2 R
(.ingber)-.715 E .959(.com/ under WWW)-.605 F 3.709(,a)-1.012 G .96
(nd the ASCII v)-3.709 F .96(ersion 00inde)-.165 F .96(x.txt, contain)
-.165 F(an inde)72 655.2 Q 2.75(xo)-.165 G 2.75(ft)-2.75 G(he other f)
-2.75 E(iles.)-.22 E 2.445(The latest v)99.5 670.8 R 2.445
(ersion of ASA, ASA\255x.y \(x and y are v)-.165 F 2.444
(ersion numbers\), can be obtained in tw)-.165 F(o)-.11 E 1.952
(formats: http://www)72 682.8 R(.ingber)-.715 E(.com/ASA.tar)-.605 E
1.952(.gz and http://www)-.605 F(.ingber)-.715 E 4.702
(.com/ASA.zip. The)-.605 F(tar')4.703 E 4.703(dv)-.55 G 1.953
(ersions is)-4.868 F .678(compressed in gzip format, and ASA.tar)72
694.8 R 3.427(.gz. In)-.605 F .677(the zip')3.427 F 3.427(dv)-.55 G .677
(ersion, ASA.zip, all f)-3.592 F .677(iles ha)-.22 F 1.007 -.165(ve b)
-.22 H .677(een processed).165 F(for DOS format.)72 706.8 Q -.165(Pa)
99.5 722.4 S .57(tches ASA\255dif).165 F .57
(f\255x1.y1\255x2.y2 up to the present v)-.275 F .57
(ersion can be prepared if a good case for doing)-.165 F 2.157
(so is presented, e.g. to f)72 734.4 R 2.157(acilitate updating your o)
-.11 F 2.157(wn modif)-.275 F 2.157(ied codes.)-.22 F 2.157
(These may be concatenated as)7.657 F(required before applying.)72 746.4
Q(If you require a specif)5.5 E(ic patch, contact ingber@ingber)-.22 E
(.com.)-.605 E 2.75(-4-)296.837 786 S 0 Cg EP
%%Page: 5 11
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(Adapti)72 75 Q .33 -.165(ve S)-.275 H
(imulated Annealing \(ASA\)).165 E(Lester Ingber)241.147 E/F1 11
/Times-Bold@0 SF 2.75(4.2. Electr)72 138 R(onic Mail)-.198 E F0 1.458
(If you do not ha)99.5 153.6 R 1.788 -.165(ve W)-.22 H 1.458
(WW or FTP access, get the Guide to Of).165 F 1.458
(fline Internet Access, returned by)-.275 F 15.274
(sending an e\255mail to mail\255serv)72 165.6 R 15.273
([e-mail] with only the w)-.165 F 15.273(ords \231send)-.11 F
(usenet/ne)72 177.6 Q 1.472(ws.answers/internet\255services/access\255v\
ia\255email\232 in the body of the message.)-.275 F 1.472(The guide gi)
6.972 F -.165(ve)-.275 G(s).165 E 2.257(information on using e\255mail \
to access just about all InterNet information and documents.)72 189.6 R
-1.21(Yo)7.757 G 5.006(uw)1.21 G(ill)-5.006 E(recei)72 201.6 Q .33 -.165
(ve t)-.275 H
(he information in utils_access\255via\255email.txt in the ASA archi)
.165 E -.165(ve)-.275 G(.).165 E F1 2.75(5. Backgr)72 225.6 R(ound)-.198
E 2.75(5.1. Context)72 253.2 R F0 -.88(To)99.5 268.8 S 2.811(oo).88 G
.061(ften the management of comple)-2.811 F 2.811(xs)-.165 G .061
(ystems is ill\255serv)-2.811 F .061
(ed by not utilizing the best tools a)-.165 F -.275(va)-.22 G(ilable.)
.275 E -.165(Fo)72 280.8 S 4.716(re).165 G 1.966
(xample, requirements set by decision\255mak)-4.881 F 1.965
(ers often are not formulated in the same language as)-.11 F 2.039
(constructs formulated by po)72 292.8 R 2.04(werful mathematical formal\
isms, and so the products of analyses are not)-.275 F .594
(properly or maximally utilized, e)72 304.8 R -.165(ve)-.275 G 3.344(ni)
.165 G 3.344(fa)-3.344 G .594(nd when the)-3.344 F 3.344(yc)-.165 G .593
(ome close to f)-3.344 F .593(aithfully representing the po)-.11 F
(werful)-.275 E 1.738(intuitions the)72 316.8 R 4.488(ya)-.165 G 1.738
(re supposed to model.)-4.488 F 1.738(In turn, e)7.238 F -.165(ve)-.275
G 4.489(np).165 G -.275(ow)-4.489 G 1.739
(erful mathematical constructs are ill\255serv).275 F(ed,)-.165 E 3.561
(especially when dealing with multi)72 328.8 R -.275(va)-.275 G 3.561
(riate nonlinear comple).275 F 6.31(xs)-.165 G 3.56
(ystems, when these formalisms are)-6.31 F -.22(bu)72 340.8 S 2.376(tch\
ered into quasi\255linear approximations to satisfy constraints of nume\
rical algorithms f).22 F 2.377(amiliar to)-.11 F 3.485
(particular analysts, b)72 352.8 R 3.485(ut which tend to destro)-.22 F
6.235(yt)-.11 G 3.485(he po)-6.235 F 3.485(wer of the intuiti)-.275 F
3.815 -.165(ve c)-.275 H 3.485(onstructs de).165 F -.165(ve)-.275 G
3.485(loped by).165 F(decision\255mak)72 364.8 Q(ers.)-.11 E .425
(In order to deal with f)99.5 380.4 R .425(itting parameters or e)-.22 F
.425(xploring sensiti)-.165 F .425(vities of v)-.275 F .425
(ariables, as models of systems)-.275 F(ha)72 392.4 Q .702 -.165(ve b)
-.22 H .372(ecome more sophisticated in describing comple).165 F 3.122
(xb)-.165 G(eha)-3.122 E(vior)-.22 E 3.122(,i)-.44 G 3.122(th)-3.122 G
.372(as become increasingly important to)-3.122 F .917
(retain and respect the nonlinearities inherent in these models, as the)
72 404.4 R 3.667(ya)-.165 G .917(re indeed present in the comple)-3.667
F(x)-.165 E(systems the)72 416.4 Q 2.75(ym)-.165 G 2.75(odel. ASA)-2.75
F(can help to handle these f)2.75 E
(its of nonlinear models of real\255w)-.22 E(orld data.)-.11 E .08
(It helps to visualize the problems presented by such comple)99.5 432 R
2.83(xs)-.165 G .08(ystems as a geographical terrain.)-2.83 F -.165(Fo)
5.58 G(r).165 E -.165(ex)72 444 S .672
(ample, consider a mountain range, with tw).165 F 3.422<6f99>-.11 G
(parameters,)-3.422 E 3.422<9a65>-.77 G .673
(.g., along the North\255South and East\255W)-3.422 F(est)-.88 E 4.234
(directions. W)72 456 R 4.234(ew)-.88 G 1.484(ish to f)-4.234 F 1.484
(ind the lo)-.22 F 1.484(west v)-.275 F(alle)-.275 E 4.234(yi)-.165 G
4.234(nt)-4.234 G 1.484(his terrain.)-4.234 F 1.484
(ASA approaches this problem similar to)6.984 F 3.127
(using a bouncing ball that can bounce o)72 468 R -.165(ve)-.165 G 5.877
(rm).165 G 3.128(ountains from v)-5.877 F(alle)-.275 E 5.878(yt)-.165 G
5.878(ov)-5.878 G(alle)-6.153 E 7.308 -.715(y. W)-.165 H 5.878(es)-.165
G 3.128(tart at a high)-5.878 F(\231temperature,)72 480 Q 5.12<9a77>-.77
G 2.37(here the temperature is an ASA parameter that mimics the ef)-5.12
F 2.369(fect of a f)-.275 F 2.369(ast mo)-.11 F(ving)-.165 E .178
(particle in a hot object lik)72 492 R 2.928(eah)-.11 G .178
(ot molten metal, thereby permitting the ball to mak)-2.928 F 2.928(ev)
-.11 G .178(ery high bounces and)-3.093 F .465(being able to bounce o)72
504 R -.165(ve)-.165 G 3.215(ra).165 G .795 -.165(ny m)-3.215 H .464
(ountain to access an).165 F 3.214(yv)-.165 G(alle)-3.489 E 1.894 -.715
(y, g)-.165 H -2.365 -.275(iv e).715 H 3.214(ne).275 G .464
(nough bounces.)-3.214 F .464(As the temperature)5.964 F 1.71
(is made relati)72 516 R -.165(ve)-.275 G 1.71(ly colder).165 F 4.46(,t)
-.44 G 1.71(he ball cannot bounce so high, and it also can settle to be\
come trapped in)-4.46 F(relati)72 528 Q -.165(ve)-.275 G
(ly smaller ranges of v).165 E(alle)-.275 E(ys.)-.165 E 2.166 -.88(We i)
99.5 543.6 T .405(magine that our mountain range is aptly described by \
a \231cost function.).88 F 5.905<9a57>-.77 G 3.155(ed)-6.785 G(ef)-3.155
E .405(ine probability)-.22 F(distrib)72 555.6 Q .081(utions of the tw)
-.22 F 2.831(od)-.11 G .081
(irectional parameters, called generating distrib)-2.831 F .081
(utions since the)-.22 F 2.831(yg)-.165 G .082(enerate possible)-2.831 F
-.275(va)72 567.6 S(lle).275 E 1.637(ys or states we are to e)-.165 F
4.386(xplore. W)-.165 F 4.386(ed)-.88 G(ef)-4.386 E 1.636
(ine another distrib)-.22 F 1.636(ution, called the acceptance distrib)
-.22 F(ution,)-.22 E .861(which depends on the dif)72 579.6 R .861
(ference of cost functions of the present generated v)-.275 F(alle)-.275
E 3.612(yw)-.165 G 3.612(ea)-3.612 G .862(re to e)-3.612 F .862
(xplore and)-.165 F .01(the last sa)72 591.6 R -.165(ve)-.22 G 2.759(dl)
.165 G -.275(ow)-2.759 G .009(est v).275 F(alle)-.275 E 4.189 -.715
(y. T)-.165 H .009(he acceptance distrib).715 F .009
(ution decides probabilistically whether to stay in a ne)-.22 F(w)-.275
E(lo)72 603.6 Q 3.705(wer v)-.275 F(alle)-.275 E 6.455(yo)-.165 G 6.456
(rt)-6.455 G 6.456(ob)-6.456 G 3.706(ounce out of it.)-6.456 F 3.706
(All the generating and acceptance distrib)9.206 F 3.706
(utions depend on)-.22 F(temperatures.)72 615.6 Q 1.342(The ASA code w)
99.5 631.2 R 1.342(as f)-.11 F 1.342(irst de)-.22 F -.165(ve)-.275 G
1.342(loped in 1987 as V).165 F 1.341(ery F)-1.221 F 1.341
(ast Simulated Reannealing \(VFSR\) to deal)-.165 F 3.389
(with the necessity of performing adapti)72 643.2 R 3.719 -.165(ve g)
-.275 H 3.39(lobal optimization on multi).165 F -.275(va)-.275 G 3.39
(riate nonlinear stochastic).275 F 4.438(systems[2]. The)72 655.2 R -.22
(fi)4.438 G 1.688(rst published use of VFSR for a comple).22 F 4.437(xs)
-.165 G 1.687(ystems w)-4.437 F 1.687(as in combat analysis, using a)
-.11 F .782(model of combat f)72 667.2 R .782(irst de)-.22 F -.165(ve)
-.275 G .782(loped in 1986, and then applied to e).165 F -.165(xe)-.165
G .783(rcise and simulation data in a series of).165 F 3.093
(papers that spanned 1988-1993[3].)72 679.2 R 3.092(The f)8.592 F 3.092
(irst applications to combat analysis used code written in)-.22 F(RA)72
691.2 Q .246(TFOR and con)-1.221 F -.165(ve)-.44 G .246(rted into FOR)
.165 F 2.997(TRAN. Other)-.66 F .247(a