subroutine pikaia_master c====================================================================== c driver program for pikaia.f c====================================================================== implicit none integer n, seed, i, status, iargc *** NUMBER OF PARAMETERS !!! parameter (n=2) real ctrl(12), x(n), f, userff character*80 seed_string external userff c ------------------------------------------- c initialize the random-number generator c ------------------------------------------- seed = 123456 if (IARGC() .GT. 0) then call getarg(1,seed_string) read(seed_string,*) seed endif write(*,*) "Random Seed: ",seed c call rninit(seed) c c Set control variables (use defaults) do 10 i=1,12 ctrl(i) = -1 10 continue *** num trial soltns ctrl(1)=128 *** num generations ctrl(2)=100 *** length of gene ctrl(3)=2 *** crossover probability ctrl(4)=0.85 *** mutation mode (5/6=uniform+creep variable F/D) ctrl(5)=6 *** initial/min/max mutation rate ctrl(6)=0.005 ctrl(7)=0.001 ctrl(8)=1./(n*ctrl(3)) *** relative fitness differential ctrl(9)=1.00 *** reproduction plan (1=full gen.rep.) ctrl(10)=1 *** elitism flag (1=keep best) ctrl(11)=1 *** output verbosity level 0/1/2 ctrl(12)=2 c c Now call pikaia call pikaia(userff,n,ctrl,x,f,status) c c Print the results write(*,*) ' status: ',status write(*,*) ' x: ',x write(*,*) ' f: ',f write(*,20) ctrl 20 format( ' ctrl: ',6f9.5/10x,6f9.5) c end c********************************************************************* function urand() c===================================================================== c Return the next pseudo-random deviate from a sequence which is c uniformly distributed in the interval [0,1] c c Uses the function ran0, the "minimal standard" random number c generator of Park and Miller (Comm. ACM 31, 1192-1201, Oct 1988; c Comm. ACM 36 No. 7, 105-110, July 1993). c===================================================================== implicit none c c Input - none c c Output real urand c c Local integer iseed real ran0 external ran0 c c Common block to make iseed visible to rninit (and to save c it between calls) common /rnseed/ iseed c urand = ran0( iseed ) return end c********************************************************************* subroutine rninit( seed ) c===================================================================== c Initialize random number generator urand with given seed c===================================================================== implicit none c c Input integer seed c c Output - none c c Local integer iseed c c Common block to communicate with urand common /rnseed/ iseed c c Set the seed value iseed = seed if(iseed.le.0) iseed=123456 return end c********************************************************************* function ran0( seed ) c===================================================================== c "Minimal standard" pseudo-random number generator of Park and c Miller. Returns a uniform random deviate r s.t. 0 < r < 1.0. c Set seed to any non-zero integer value to initialize a sequence, c then do not change seed between calls for successive deviates c in the sequence. c c References: c Park, S. and Miller, K., "Random Number Generators: Good Ones c are Hard to Find", Comm. ACM 31, 1192-1201 (Oct. 1988) c Park, S. and Miller, K., in "Remarks on Choosing and Imple- c menting Random Number Generators", Comm. ACM 36 No. 7, c 105-110 (July 1993) c===================================================================== c *** Declaration section *** c implicit none c c Input/Output: integer seed c c Output: real ran0 c c Constants: integer A,M,Q,R parameter (A=48271,M=2147483647,Q=44488,R=3399) real SCALE,EPS,RNMX parameter (SCALE=1./M,EPS=1.2e-7,RNMX=1.-EPS) c c Local: integer j c c *** Executable section *** c j = seed/Q seed = A*(seed-j*Q)-R*j if (seed .lt. 0) seed = seed+M ran0 = min(seed*SCALE,RNMX) return end c********************************************************************** subroutine rqsort(n,a,p) c====================================================================== c Return integer array p which indexes array a in increasing order. c Array a is not disturbed. The Quicksort algorithm is used. c c B. G. Knapp, 86/12/23 c c Reference: N. Wirth, Algorithms and Data Structures, c Prentice-Hall, 1986 c====================================================================== implicit none c Input: integer n real a(n) c Output: integer p(n) c Constants integer LGN, Q parameter (LGN=32, Q=11) c (LGN = log base 2 of maximum n; c Q = smallest subfile to use quicksort on) c Local: real x integer stackl(LGN),stackr(LGN),s,t,l,m,r,i,j c Initialize the stack stackl(1)=1 stackr(1)=n s=1 c Initialize the pointer array do 1 i=1,n p(i)=i 1 continue 2 if (s.gt.0) then l=stackl(s) r=stackr(s) s=s-1 3 if ((r-l).lt.Q) then c Use straight insertion do 6 i=l+1,r t = p(i) x = a(t) do 4 j=i-1,l,-1 if (a(p(j)).le.x) goto 5 p(j+1) = p(j) 4 continue j=l-1 5 p(j+1) = t 6 continue else c Use quicksort, with pivot as median of a(l), a(m), a(r) m=(l+r)/2 t=p(m) if (a(t).lt.a(p(l))) then p(m)=p(l) p(l)=t t=p(m) endif if (a(t).gt.a(p(r))) then p(m)=p(r) p(r)=t t=p(m) if (a(t).lt.a(p(l))) then p(m)=p(l) p(l)=t t=p(m) endif endif c Partition x=a(t) i=l+1 j=r-1 7 if (i.le.j) then 8 if (a(p(i)).lt.x) then i=i+1 goto 8 endif 9 if (x.lt.a(p(j))) then j=j-1 goto 9 endif if (i.le.j) then t=p(i) p(i)=p(j) p(j)=t i=i+1 j=j-1 endif goto 7 endif c Stack the larger subfile s=s+1 if ((j-l).gt.(r-i)) then stackl(s)=l stackr(s)=j l=i else stackl(s)=i stackr(s)=r r=j endif goto 3 endif goto 2 endif return end c*********************************************************************** subroutine pikaia(ff,n,ctrl,x,f,status) c======================================================================= c Optimization (maximization) of user-supplied "fitness" function c ff over n-dimensional parameter space x using a basic genetic c algorithm method. c c Paul Charbonneau & Barry Knapp c High Altitude Observatory c National Center for Atmospheric Research c Boulder CO 80307-3000 c c c c Version 1.2 [ 2002 April 3 ] c c Genetic algorithms are heuristic search techniques that c incorporate in a computational setting, the biological notion c of evolution by means of natural selection. This subroutine c implements the three basic operations of selection, crossover, c and mutation, operating on "genotypes" encoded as strings. c c Version 1.2 differs from version 1.0 (December 1995) in that c it includes (1) two-point crossover, (2) creep mutation, and c (3) dynamical adjustment of the mutation rate based on metric c distance in parameter space. c c References: c c Charbonneau, Paul. "An introduction to gemetic algorithms for c numerical optimization", NCAR Technical Note TN-450+IA c (April 2002) c c Charbonneau, Paul. "Release Notes for PIKAIA 1.2", c NCAR Technical Note TN-451+STR (April 2002) c c Charbonneau, Paul, and Knapp, Barry. "A User's Guide c to PIKAIA 1.0" NCAR Technical Note TN-418+IA c (December 1995) c c Goldberg, David E. Genetic Algorithms in Search, Optimization, c & Machine Learning. Addison-Wesley, 1989. c c Davis, Lawrence, ed. Handbook of Genetic Algorithms. c Van Nostrand Reinhold, 1991. c c======================================================================= c USES: ff, urand, setctl, report, rnkpop, select, encode, decode, c cross, mutate, genrep, stdrep, newpop, adjmut implicit none c Input: integer n real ff external ff c c o Integer n is the parameter space dimension, i.e., the number c of adjustable parameters. c c o Function ff is a user-supplied scalar function of n vari- c ables, which must have the calling sequence f = ff(n,x), where c x is a real parameter array of length n. This function must c be written so as to bound all parameters to the interval [0,1]; c that is, the user must determine a priori bounds for the para- c meter space, and ff must use these bounds to perform the appro- c priate scalings to recover true parameter values in the c a priori ranges. c c By convention, ff should return higher values for more optimal c parameter values (i.e., individuals which are more "fit"). c For example, in fitting a function through data points, ff c could return the inverse of chi**2. c c In most cases initialization code will have to be written c (either in a driver or in a separate subroutine) which loads c in data values and communicates with ff via one or more labeled c common blocks. An example exercise driver and fitness function c are provided in the accompanying file, xpkaia.f. c c c Input/Output: real ctrl(12) c c o Array ctrl is an array of control flags and parameters, to c control the genetic behavior of the algorithm, and also printed c output. A default value will be used for any control variable c which is supplied with a value less than zero. On exit, ctrl c contains the actual values used as control variables. The c elements of ctrl and their defaults are: c c ctrl( 1) - number of individuals in a population (default c is 100) c ctrl( 2) - number of generations over which solution is c to evolve (default is 500) c ctrl( 3) - number of significant digits (i.e., number of c genes) retained in chromosomal encoding (default c is 6) (Note: This number is limited by the c machine floating point precision. Most 32-bit c floating point representations have only 6 full c digits of precision. To achieve greater preci- c sion this routine could be converted to double c precision, but note that this would also require c a double precision random number generator, which c likely would not have more than 9 digits of c precision if it used 4-byte integers internally.) c ctrl( 4) - crossover probability; must be <= 1.0 (default c is 0.85). If crossover takes place, either one c or two splicing points are used, with equal c probabilities c ctrl( 5) - mutation mode; 1/2/3/4/5 (default is 2) c 1=one-point mutation, fixed rate c 2=one-point, adjustable rate based on fitness c 3=one-point, adjustable rate based on distance c 4=one-point+creep, fixed rate c 5=one-point+creep, adjustable rate based on fitness c 6=one-point+creep, adjustable rate based on distance c ctrl( 6) - initial mutation rate; should be small (default c is 0.005) (Note: the mutation rate is the proba- c bility that any one gene locus will mutate in c any one generation.) c ctrl( 7) - minimum mutation rate; must be >= 0.0 (default c is 0.0005) c ctrl( 8) - maximum mutation rate; must be <= 1.0 (default c is 0.25) c ctrl( 9) - relative fitness differential; range from 0 c (none) to 1 (maximum). (default is 1.) c ctrl(10) - reproduction plan; 1/2/3=Full generational c replacement/Steady-state-replace-random/Steady- c state-replace-worst (default is 3) c ctrl(11) - elitism flag; 0/1=off/on (default is 0) c (Applies only to reproduction plans 1 and 2) c ctrl(12) - printed output 0/1/2=None/Minimal/Verbose c (default is 0) c c c Output: real x(n), f integer status c c o Array x(1:n) is the "fittest" (optimal) solution found, c i.e., the solution which maximizes fitness function ff c c o Scalar f is the value of the fitness function at x c c o Integer status is an indicator of the success or failure c of the call to pikaia (0=success; non-zero=failure) c c c Constants integer NMAX, PMAX, DMAX parameter (NMAX = 32, PMAX = 128, DMAX = 6) c c o NMAX is the maximum number of adjustable parameters c (n <= NMAX) c c o PMAX is the maximum population (ctrl(1) <= PMAX) c c o DMAX is the maximum number of Genes (digits) per Chromosome c segement (parameter) (ctrl(3) <= DMAX) c c c Local variables integer np, nd, ngen, imut, irep, ielite, ivrb, k, ip, ig, + ip1, ip2, new, newtot real pcross, pmut, pmutmn, pmutmx, fdif c real ph(NMAX,2), oldph(NMAX,PMAX), newph(NMAX,PMAX) c integer gn1(NMAX*DMAX), gn2(NMAX*DMAX) integer ifit(PMAX), jfit(PMAX) real fitns(PMAX) c c User-supplied uniform random number generator real urand external urand c c Function urand should not take any arguments. If the user wishes c to be able to initialize urand, so that the same sequence of c random numbers can be repeated, this capability could be imple- c mented with a separate subroutine, and called from the user's c driver program. An example urand function (and initialization c subroutine) which uses the function ran0 (the "minimal standard" c random number generator of Park and Miller [Comm. ACM 31, 1192- c 1201, Oct 1988; Comm. ACM 36 No. 7, 105-110, July 1993]) is c provided. c c c Set control variables from input and defaults call setctl + (ctrl,n,np,ngen,nd,pcross,pmutmn,pmutmx,pmut,imut, + fdif,irep,ielite,ivrb,status) if (status .ne. 0) then write(*,*) ' Control vector (ctrl) argument(s) invalid' return endif c Make sure locally-dimensioned arrays are big enough if (n.gt.NMAX .or. np.gt.PMAX .or. nd.gt.DMAX) then write(*,*) + ' Number of parameters, population, or genes too large' status = -1 return endif c Compute initial (random but bounded) phenotypes do 1 ip=1,np do 2 k=1,n oldph(k,ip)=urand() 2 continue cMPI --- line below commented out for parallel version --- cMPI fitns(ip) = ff(n,oldph(1,ip)) 1 continue cMPI --- get fitnesses [in parallel] using caller's fitness function --- call mpi_fitness(np, n, oldph, fitns) c Rank initial population by fitness order call rnkpop(np,fitns,ifit,jfit) c Main Generation Loop do 10 ig=1,ngen c Main Population Loop newtot=0 do 20 ip=1,np/2 c 1. pick two parents call select(np,jfit,fdif,ip1) 21 call select(np,jfit,fdif,ip2) if (ip1.eq.ip2) goto 21 c 2. encode parent phenotypes call encode(n,nd,oldph(1,ip1),gn1) call encode(n,nd,oldph(1,ip2),gn2) c 3. breed call cross(n,nd,pcross,gn1,gn2) call mutate(n,nd,pmut,gn1,imut) call mutate(n,nd,pmut,gn2,imut) c 4. decode offspring genotypes call decode(n,nd,gn1,ph(1,1)) call decode(n,nd,gn2,ph(1,2)) c 5. insert into population if (irep.eq.1) then call genrep(NMAX,n,np,ip,ph,newph) else call stdrep(ff,NMAX,n,np,irep,ielite, + ph,oldph,fitns,ifit,jfit,new) newtot = newtot+new endif c End of Main Population Loop 20 continue c if running full generational replacement: swap populations if (irep.eq.1) + call newpop(ff,ielite,NMAX,n,np,oldph,newph, + ifit,jfit,fitns,newtot) c adjust mutation rate? if (imut.eq.2 .or. imut.eq.3 .or. imut.eq.5 .or. imut.eq.6) + call adjmut(NMAX,n,np,oldph,fitns,ifit,pmutmn,pmutmx, + pmut,imut) c if (ivrb.gt.0) call report + (ivrb,NMAX,n,np,nd,oldph,fitns,ifit,pmut,ig,newtot) c End of Main Generation Loop 10 continue c c Return best phenotype and its fitness do 30 k=1,n x(k) = oldph(k,ifit(np)) 30 continue f = fitns(ifit(np)) c end c******************************************************************** subroutine setctl + (ctrl,n,np,ngen,nd,pcross,pmutmn,pmutmx,pmut,imut, + fdif,irep,ielite,ivrb,status) c=================================================================== c Set control variables and flags from input and defaults c=================================================================== implicit none c c Input integer n c c Input/Output real ctrl(12) c c Output integer np, ngen, nd, imut, irep, ielite, ivrb, status real pcross, pmutmn, pmutmx, pmut, fdif c c Local integer i real DFAULT(12) save DFAULT data DFAULT /100,500,5,.85,2,.005,.0005,.25,1,1,1,0/ c do 1 i=1,12 if (ctrl(i).lt.0.) ctrl(i)=DFAULT(i) 1 continue np = ctrl(1) ngen = ctrl(2) nd = ctrl(3) pcross = ctrl(4) imut = ctrl(5) pmut = ctrl(6) pmutmn = ctrl(7) pmutmx = ctrl(8) fdif = ctrl(9) irep = ctrl(10) ielite = ctrl(11) ivrb = ctrl(12) status = 0 c c Print a header if (ivrb.gt.0) then write(*,2) ngen,np,n,nd,pcross,pmut,pmutmn,pmutmx,fdif 2 format(/1x,60('*'),/, + ' *',13x,'PIKAIA Genetic Algorithm Report ',13x,'*',/, + 1x,60('*'),//, + ' Number of Generations evolving: ',i4,/, + ' Individuals per generation: ',i4,/, + ' Number of Chromosome segments: ',i4,/, + ' Length of Chromosome segments: ',i4,/, + ' Crossover probability: ',f9.4,/, + ' Initial mutation rate: ',f9.4,/, + ' Minimum mutation rate: ',f9.4,/, + ' Maximum mutation rate: ',f9.4,/, + ' Relative fitness differential: ',f9.4) if (imut.eq.1) write(*,3) 'Uniform, Constant Rate' if (imut.eq.2) write(*,3) 'Uniform, Variable Rate (F)' if (imut.eq.3) write(*,3) 'Uniform, Variable Rate (D)' if (imut.eq.4) write(*,3) 'Uniform+Creep, Constant Rate' if (imut.eq.5) write(*,3) 'Uniform+Creep, Variable Rate (F)' if (imut.eq.6) write(*,3) 'Uniform+Creep, Variable Rate (D)' 3 format( + ' Mutation Mode: ',A) if (irep.eq.1) write(*,4) 'Full generational replacement' if (irep.eq.2) write(*,4) 'Steady-state-replace-random' if (irep.eq.3) write(*,4) 'Steady-state-replace-worst' 4 format( + ' Reproduction Plan: ',A) endif c Check some control values if (imut.ne.1 .and. imut.ne.2 .and. imut.ne.3 .and. imut.ne.4 + .and. imut.ne.5 .and. imut.ne.6) then write(*,10) status = 5 endif 10 format(' ERROR: illegal value for imut (ctrl(5))') if (fdif.gt.1.) then write(*,11) status = 9 endif 11 format(' ERROR: illegal value for fdif (ctrl(9))') if (irep.ne.1 .and. irep.ne.2 .and. irep.ne.3) then write(*,12) status = 10 endif 12 format(' ERROR: illegal value for irep (ctrl(10))') if (pcross.gt.1.0 .or. pcross.lt.0.) then write(*,13) status = 4 endif 13 format(' ERROR: illegal value for pcross (ctrl(4))') if (ielite.ne.0 .and. ielite.ne.1) then write(*,14) status = 11 endif 14 format(' ERROR: illegal value for ielite (ctrl(11))') if (irep.eq.1 .and. imut.eq.1 .and. pmut.gt.0.5 .and. + ielite.eq.0) then write(*,15) endif 15 format(' WARNING: dangerously high value for pmut (ctrl(6));', + /' (Should enforce elitism with ctrl(11)=1.)') if (irep.eq.1 .and. imut.eq.2 .and. pmutmx.gt.0.5 .and. + ielite.eq.0) then write(*,16) endif 16 format(' WARNING: dangerously high value for pmutmx (ctrl(8));', + /' (Should enforce elitism with ctrl(11)=1.)') if (fdif.lt.0.33 .and. irep.ne.3) then write(*,17) endif 17 format(' WARNING: dangerously low value of fdif (ctrl(9))') if (mod(np,2).gt.0) then np=np-1 write(*,18) np endif 18 format(' WARNING: decreasing population size (ctrl(1)) to np=',i4) return end c******************************************************************** subroutine report + (ivrb,ndim,n,np,nd,oldph,fitns,ifit,pmut,ig,nnew) c c Write generation report to standard output c implicit none c Input: integer np,ifit(np),ivrb,ndim,n,nd,ig,nnew real oldph(ndim,np),fitns(np),pmut c c Output: none c c Local real bestft,pmutpv save bestft,pmutpv integer ndpwr,k logical rpt data bestft,pmutpv /0,0/ c rpt=.false. if (pmut.ne.pmutpv) then pmutpv=pmut rpt=.true. endif if (fitns(ifit(np)).ne.bestft) then bestft=fitns(ifit(np)) rpt=.true. endif if (rpt .or. ivrb.ge.2) then c Power of 10 to make integer genotypes for display ndpwr = nint(10.**nd) write(*,'(/i6,i6,f10.6,4f10.6)') ig,nnew,pmut, + fitns(ifit(np)), fitns(ifit(np-1)), fitns(ifit(np/2)) do 15 k=1,n write(*,'(22x,3i10)') + nint(ndpwr*oldph(k,ifit(np ))), + nint(ndpwr*oldph(k,ifit(np-1))), + nint(ndpwr*oldph(k,ifit(np/2))) 15 continue endif end c********************************************************************** c GENETICS MODULE c********************************************************************** c c ENCODE: encodes phenotype into genotype c called by: PIKAIA c c DECODE: decodes genotype into phenotype c called by: PIKAIA c c CROSS: Breeds two offspring from two parents c called by: PIKAIA c c MUTATE: Introduces random mutation in a genotype c called by: PIKAIA c c ADJMUT: Implements variable mutation rate c called by: PIKAIA c c********************************************************************** subroutine encode(n,nd,ph,gn) c====================================================================== c encode phenotype parameters into integer genotype c ph(k) are x,y coordinates [ 0 < x,y < 1 ] c====================================================================== c implicit none c c Inputs: integer n, nd real ph(n) c c Output: integer gn(n*nd) c c Local: integer ip, i, j, ii real z c z=10.**nd ii=0 do 1 i=1,n ip=int(ph(i)*z) do 2 j=nd,1,-1 gn(ii+j)=mod(ip,10) ip=ip/10 2 continue ii=ii+nd 1 continue return end c********************************************************************** subroutine decode(n,nd,gn,ph) c====================================================================== c decode genotype into phenotype parameters c ph(k) are x,y coordinates [ 0 < x,y < 1 ] c====================================================================== c implicit none c c Inputs: integer n, nd, gn(n*nd) c c Output: real ph(n) c c Local: integer ip, i, j, ii real z c z=10.**(-nd) ii=0 do 1 i=1,n ip=0 do 2 j=1,nd ip=10*ip+gn(ii+j) 2 continue ph(i)=ip*z ii=ii+nd 1 continue return end c********************************************************************** subroutine cross(n,nd,pcross,gn1,gn2) c====================================================================== c breeds two parent chromosomes into two offspring chromosomes c breeding occurs through crossover. If the crossover probability c test yields true (crossover taking place), either one-point or c two-point crossover is used, with equal probabilities. c c Compatibility with version 1.0: To enforce 100% use of one-point c crossover, un-comment appropriate line in source code below c====================================================================== c implicit none c c Inputs: integer n, nd real pcross c c Input/Output: integer gn1(n*nd), gn2(n*nd) c c Local: integer i, ispl, ispl2, itmp, t c c Function real urand external urand c Use crossover probability to decide whether a crossover occurs if (urand().lt.pcross) then c Compute first crossover point ispl=int(urand()*n*nd)+1 c Now choose between one-point and two-point crossover if (urand().lt.0.5) then ispl2=n*nd else ispl2=int(urand()*n*nd)+1 c Un-comment following line to enforce one-point crossover c ispl2=n*nd if (ispl2.lt.ispl) then itmp=ispl2 ispl2=ispl ispl=itmp endif endif c Swap genes from ispl to ispl2 do 10 i=ispl,ispl2 t=gn2(i) gn2(i)=gn1(i) gn1(i)=t 10 continue endif return end c********************************************************************** subroutine mutate(n,nd,pmut,gn,imut) c====================================================================== c Mutations occur at rate pmut at all gene loci c imut=1 Uniform mutation, constant rate c imut=2 Uniform mutation, variable rate based on fitness c imut=3 Uniform mutation, variable rate based on distance c imut=4 Uniform or creep mutation, constant rate c imut=5 Uniform or creep mutation, variable rate based on c fitness c imut=6 Uniform or creep mutation, variable rate based on c distance c====================================================================== c implicit none c c Input: integer n, nd, imut real pmut c c Input/Output: integer gn(n*nd) c c Local: integer i,j,k,l,ist,inc,loc,kk c c Function: real urand external urand c c Decide which type of mutation is to occur if(imut.ge.4.and.urand().le.0.5)then c CREEP MUTATION OPERATOR c Subject each locus to random +/- 1 increment at the rate pmut do 1 i=1,n do 2 j=1,nd if (urand().lt.pmut) then c Construct integer loc=(i-1)*nd+j inc=nint ( urand() )*2-1 ist=(i-1)*nd+1 gn(loc)=gn(loc)+inc c write(*,*) ist,loc,inc c This is where we carry over the one (up to two digits) c first take care of decrement below 0 case if(inc.lt.0 .and. gn(loc).lt.0)then if(j.eq.1)then gn(loc)=0 else do 3 k=loc,ist+1,-1 gn(k)=9 gn(k-1)=gn(k-1)-1 if( gn(k-1).ge.0 )goto 4 3 continue c we popped under 0.00000 lower bound; fix it up if( gn(ist).lt.0.)then do 5 l=ist,loc gn(l)=0 5 continue endif 4 continue endif endif if(inc.gt.0 .and. gn(loc).gt.9)then if(j.eq.1)then gn(loc)=9 else do 6 k=loc,ist+1,-1 gn(k)=0 gn(k-1)=gn(k-1)+1 if( gn(k-1).le.9 )goto 7 6 continue c we popped over 9.99999 upper bound; fix it up if( gn(ist).gt.9 )then do 8 l=ist,loc gn(l)=9 8 continue endif 7 continue endif endif endif 2 continue 1 continue else c UNIFORM MUTATION OPERATOR c Subject each locus to random mutation at the rate pmut do 10 i=1,n*nd if (urand().lt.pmut) then gn(i)=int(urand()*10.) endif 10 continue endif return end c********************************************************************** subroutine adjmut(ndim,n,np,oldph,fitns,ifit,pmutmn,pmutmx, + pmut,imut) c====================================================================== c dynamical adjustment of mutation rate; c imut=2 or imut=5 : adjustment based on fitness differential c between best and median individuals c imut=3 or imut=6 : adjustment based on metric distance c between best and median individuals c====================================================================== c implicit none c c Input: integer n, ndim, np, ifit(np), imut real oldph(ndim,np), fitns(np), pmutmn, pmutmx c c Input/Output: real pmut c c Local: integer i real rdif, rdiflo, rdifhi, delta parameter (rdiflo=0.05, rdifhi=0.25, delta=1.5) if(imut.eq.2.or.imut.eq.5)then c Adjustment based on fitness differential rdif=abs(fitns(ifit(np))-fitns(ifit(np/2)))/ + (fitns(ifit(np))+fitns(ifit(np/2))) else if(imut.eq.3.or.imut.eq.6)then c Adjustment based on normalized metric distance rdif=0. do 1 i=1,n rdif=rdif+( oldph(i,ifit(np))-oldph(i,ifit(np/2)) )**2 1 continue rdif=sqrt( rdif ) / float(n) endif if(rdif.le.rdiflo)then pmut=min(pmutmx,pmut*delta) else if(rdif.ge.rdifhi)then pmut=max(pmutmn,pmut/delta) endif return end c********************************************************************** c REPRODUCTION MODULE c********************************************************************** c c SELECT: Parent selection by roulette wheel algorithm c called by: PIKAIA c c RNKPOP: Ranks initial population c called by: PIKAIA, NEWPOP c c GENREP: Inserts offspring into population, for full c generational replacement c called by: PIKAIA c c STDREP: Inserts offspring into population, for steady-state c reproduction c called by: PIKAIA c calls: FF c c NEWPOP: Replaces old generation with new generation c called by: PIKAIA c calls: FF, RNKPOP c c********************************************************************** subroutine select(np,jfit,fdif,idad) c====================================================================== c Selects a parent from the population, using roulette wheel c algorithm with the relative fitnesses of the phenotypes as c the "hit" probabilities [see Davis 1991, chap. 1]. c====================================================================== c USES: urand implicit none c c Input: integer np, jfit(np) real fdif c c Output: integer idad c c Local: integer np1, i real dice, rtfit c c Function: real urand external urand c c np1 = np+1 dice = urand()*np*np1 rtfit = 0. do 1 i=1,np rtfit = rtfit+np1+fdif*(np1-2*jfit(i)) if (rtfit.ge.dice) then idad=i goto 2 endif 1 continue c Assert: loop will never exit by falling through 2 return end c********************************************************************** subroutine rnkpop(n,arrin,indx,rank) c====================================================================== c Calls external sort routine to produce key index and rank order c of input array arrin (which is not altered). c====================================================================== c USES: rqsort implicit none c c Input integer n real arrin(n) c c Output integer indx(n),rank(n) c c Local integer i c c External sort subroutine external rqsort c c c Compute the key index call rqsort(n,arrin,indx) c c ...and the rank order do 1 i=1,n rank(indx(i)) = n-i+1 1 continue return end c*********************************************************************** subroutine genrep(ndim,n,np,ip,ph,newph) c======================================================================= c full generational replacement: accumulate offspring into new c population array c======================================================================= c implicit none c Input: integer ndim, n, np, ip real ph(ndim,2) c c Output: real newph(ndim,np) c c Local: integer i1, i2, k c c c Insert one offspring pair into new population i1=2*ip-1 i2=i1+1 do 1 k=1,n newph(k,i1)=ph(k,1) newph(k,i2)=ph(k,2) 1 continue return end c********************************************************************** subroutine stdrep + (ff,ndim,n,np,irep,ielite,ph,oldph,fitns,ifit,jfit,nnew) c====================================================================== c steady-state reproduction: insert offspring pair into population c only if they are fit enough (replace-random if irep=2 or c replace-worst if irep=3). c====================================================================== c USES: ff, urand implicit none c c Input: integer ndim, n, np, irep, ielite real ff, ph(ndim,2) external ff c c Input/Output: real oldph(ndim,np), fitns(np) integer ifit(np), jfit(np) c c Output: integer nnew c Local: integer i, j, k, i1, if1 real fit c c External function real urand external urand c c nnew = 0 do 1 j=1,2 c 1. compute offspring fitness (with caller's fitness function) fit=ff(n,ph(1,j)) c 2. if fit enough, insert in population do 20 i=np,1,-1 if (fit.gt.fitns(ifit(i))) then c make sure the phenotype is not already in the population if (i.lt.np) then do 5 k=1,n if (oldph(k,ifit(i+1)).ne.ph(k,j)) goto 6 5 continue goto 1 6 continue endif c offspring is fit enough for insertion, and is unique c (i) insert phenotype at appropriate place in population if (irep.eq.3) then i1=1 else if (ielite.eq.0 .or. i.eq.np) then i1=int(urand()*np)+1 else i1=int(urand()*(np-1))+1 endif if1 = ifit(i1) fitns(if1)=fit do 21 k=1,n oldph(k,if1)=ph(k,j) 21 continue c (ii) shift and update ranking arrays if (i.lt.i1) then c shift up jfit(if1)=np-i do 22 k=i1-1,i+1,-1 jfit(ifit(k))=jfit(ifit(k))-1 ifit(k+1)=ifit(k) 22 continue ifit(i+1)=if1 else c shift down jfit(if1)=np-i+1 do 23 k=i1+1,i jfit(ifit(k))=jfit(ifit(k))+1 ifit(k-1)=ifit(k) 23 continue ifit(i)=if1 endif nnew = nnew+1 goto 1 endif 20 continue 1 continue return end c********************************************************************** subroutine newpop + (ff,ielite,ndim,n,np,oldph,newph,ifit,jfit,fitns,nnew) c====================================================================== c replaces old population by new; recomputes fitnesses & ranks c====================================================================== c USES: ff, rnkpop implicit none c c Input: integer ndim, np, n, ielite real ff external ff c c Input/Output: real oldph(ndim,np), newph(ndim,np) c c Output: integer ifit(np), jfit(np), nnew real fitns(np) c c Local: integer i, k c c nnew = np c if using elitism, introduce in new population fittest of old c population (if greater than fitness of the individual it is c to replace) if (ielite.eq.1 .and. ff(n,newph(1,1)).lt.fitns(ifit(np))) then do 1 k=1,n newph(k,1)=oldph(k,ifit(np)) 1 continue nnew = nnew-1 endif c replace population do 2 i=1,np do 3 k=1,n oldph(k,i)=newph(k,i) 3 continue cMPI --- line below commented out for parallel version --- cMPI fitns(i)=ff(n,oldph(1,i)) 2 continue cMPI --- get fitnesses [in parallel] using caller's fitness function --- call mpi_fitness(np, n, oldph, fitns) c compute new population fitness rank order call rnkpop(np,fitns,ifit,jfit) return end