def check_local_minimum(seq, moves, orig_config):
#print "(local) checking... %s" % seq
# 1k MC steps
from minimization import apply_move, find_move
from energy_function import energy_function
from genetic_lattice import print_at
init_energy = energy_function(orig_config, seq)
config = orig_config
for move in moves:
nconfig = apply_move(move, config)
if energy_function(nconfig, seq)<init_energy:
#print_at("found lower energy with MC... %s" % (energy_function(nconfig, seq)),4)
return (False, seq)
#print_at("quick MC was a'ight %s" % seq,1)
from minimization import calc_acceptance, apply_some_move
sequence_len = len(seq)
sequence_range = xrange(sequence_len)
cE = energy_function(config, seq)
#print_at("MC step?",2)
for i in xrange(10**3):
for res in sequence_range:
moves = find_move(res, config, sequence_len)
if not moves: continue
nconfig = apply_some_move(moves, config)
nE = energy_function(nconfig, seq)
if calc_acceptance(cE,nE,seq,0.5):
cE = nE
config = nconfig
if cE<init_energy:
#print_at("MC steps beat it...i %d res %d" % (i,res),2)
return (False, seq)
print_at("1k MC steps still failed",3)
# a little GA
from minimization import birth, do_mutations, do_cross_over, calc_genetic_energy, choose_fit
from genetic_lattice import encode_lattice
indivs = 100
population = birth(sequence_len, indivs) + [encode_lattice(orig_config)]
for gen in xrange(50):
do_cross_over(population,1.0,sequence_len)
do_mutations(population,1.0,sequence_len)
population = choose_fit(population, indivs, seq)
lowest_V = calc_genetic_energy(population[0],seq)
if lowest_V<init_energy:
print_at("initial energy %s" % init_energy,4)
print_at("found lower energy with GA... %s" % lowest_V,5)
return (False, seq)
print_at("local minimum? %s" % seq,6)
print "\n\n\n"
return (True, seq)
def check_global_minimum(seq, orig_config):
from genetic_lattice import print_at
print_at("(global) checking... %s" % seq,7)
f = open("/tmp/global.out",'a')
f.write(str(seq))
f.close()
# 1mil MC steps
from minimization import apply_some_move, find_move, calc_acceptance
from energy_function import energy_function
config = orig_config
init_energy = energy_function(config, seq)
from minimization import calc_acceptance
sequence_len = len(seq)
sequence_range = xrange(sequence_len)
cE = energy_function(config, seq)
for i in xrange(10**6):
for res in sequence_range:
moves = find_move(res, config, sequence_len)
if not moves: continue
nconfig = apply_some_move(moves, config)
nE = energy_function(nconfig, seq)
if calc_acceptance(cE,nE,seq,0.5):
cE = nE
config = nconfig
if cE<init_energy:
print_at("MC steps beat it...i %d res %d" % (i,res),2)
return (False, seq)
print "10k MC steps still failed"
# a little GA
from minimization import birth, do_mutations, do_cross_over, calc_genetic_energy, choose_fit
from genetic_lattice import encode_lattice
indivs = 100
population = birth(sequence_len, indivs) + [encode_lattice(orig_config)]
for gen in xrange(50):
do_cross_over(population,1.0,sequence_len)
do_mutations(population,1.0,sequence_len)
population = choose_fit(population, indivs, seq)
lowest_V = calc_genetic_energy(population[0],seq)
if lowest_V<init_energy:
print "initial energy %s" % init_energy
print "found lower energy with GA... %s" % lowest_V
return (False, seq)
print "global minimum!!!"
return (True, seq)
def global_minimum():
if len(sys.argv)==1 or 'first' in sys.argv:
sequence = convert_sequence('HHPHHPPHPPPHPPHPHHHHPPPPHHPPHPHHHHPP')
else:
sequence = convert_sequence('HHHPPHHPHHHHPHHPPHPHHPHPPHPPPHHPPPPP')
sequence_len = len(sequence)
if len(sys.argv)>1 and not sys.argv[1].isdigit(): sys.argv.pop(1) # remove first/second
lenargv = len(sys.argv)
# GA data:
# number of individuals per generation
ind_per_gen = int(sys.argv[1]) if lenargv>1 else 100
# number of generations
num_generations = int(sys.argv[2]) if lenargv>2 else 100
# number of GA/SA swaps
num_tries = int(sys.argv[3]) if lenargv>3 else 10
# SA data:
alpha = 0.995
ntemp = int(sys.argv[4]) if lenargv>4 else 1000
ncycles = int(sys.argv[5]) if lenargv>5 else 100
orig_T_star = float(sys.argv[6]) if lenargv>6 else 7.0
# T_star should be <0.2 after ntemp, so
# T_star*alpha**ntemp<0.2
# (0.2/T_star)**(1.0/ntemp)=alpha
alpha = (0.2/orig_T_star)**(1.0/(ntemp+1))
print alpha
# mutation rates
rate_cross = 0.8
rate_mutate = 0.8
population = birth(sequence_len,ind_per_gen)
lowest_V = lowest_V_sofar = 0
highest_V = 1
equal_generations = 0
lower_generations = []
for numtry in xrange(num_tries):
choose_fit(population,ind_per_gen,sequence) # just so it sorts
for i in xrange(num_generations):
print_at('generation %d' % i,0)
# mate best half?
with TimingContext(1):
#mating_population = choose_fit(population,ind_per_gen/2,sequence,resort=False)
mating_population = population[:ind_per_gen/2]
population = population[ind_per_gen/2:] # population is pre-sorted by choose_fit
with TimingContext(2):
if lowest_V < -1.0 and lowest_V == highest_V: # rebirth soon necessary
equal_generations += 1
else:
equal_generations = 0
if equal_generations > 4: # we lost genetic diversity...
mating_population = birth(sequence_len,ind_per_gen)
population = sample(population,2)
with TimingContext(3):
do_cross_over(mating_population,rate_cross,sequence_len)
with TimingContext(4):
do_mutations(mating_population,rate_mutate,sequence_len)
# keep best ones only
with TimingContext(5):
population.extend(mating_population)
with TimingContext(6):
population = choose_fit(population,ind_per_gen,sequence)
# select lowest energy
with TimingContext(7):
#if True:
lowest_conf = population[0]
lowest_V = calc_genetic_energy(lowest_conf,sequence)
highest_V = calc_genetic_energy(population[-1],sequence)
if lowest_V<lowest_V_sofar:
lowest_V_sofar = lowest_V
print_at("lowest V: %f for config %s" % (lowest_V,str(lowest_conf).replace(' ','')),8)
lower_generations.append((i,lowest_V,lowest_conf))
print_at("current_lowest %f" % lowest_V,9)
print_at("current_highest %f" % highest_V,10)
print_at("Generations worth their CPU: %s" % str(lower_generations).replace(' ',''), 11)
# have lowest conf, try annealing now...
lowest_conf = decode_lattice(lowest_conf)
T_star = orig_T_star
for k in xrange(ntemp):
print_at("SA temperature %f (%d)" % (T_star,k),25)
current_config = lowest_conf
current_V = lowest_V
print_at("lowest_V: %f" % lowest_V,26)
for j in xrange(ncycles):
print_at("Cycle %d" % j,27)
next_conf,next_V = MC_step(current_config, sequence, T_star)
if next_V<lowest_V:
current_V = next_V
current_config = next_conf
print_at("current V: %f" % next_V,28)
# take best config from cycles and use that
if current_V < lowest_V:
lowest_V = current_V
lowest_conf = current_config
print_at("lowest V: %f for config %s" % (lowest_V,str(encode_lattice(lowest_conf)).replace(' ','')),29)
T_star *= alpha
# aaaaand start over
population = [encode_lattice(lowest_conf)]+birth(sequence_len,ind_per_gen)
mating_population = []
print_at('',30)
print_at('Starting V = %f' % V,1)
print_at('Trajectory %d' % i,2)
j = 0
lowest_V = 0
while V-0.0001 > Vmin: # floating point errors...
j += 1
print_at('Move %d' % j,3)
print_at('With V = %f' % V,4)
#~ configurations[w] = MC_step(configurations[w], sequence, temperatures[w])
#~ Result = (configuration, times down sequence, number of moves)
configuration, sequence_traversals, nbr_moves, V = MC_step(
configuration, sequence, temperature, sequence_traversals, nbr_moves)
if V < lowest_V:
lowest_V = V
lowest_conf = configuration
print_at('best config w/ E=%f: %s' % (lowest_V,str(encode_lattice(lowest_conf)).replace(' ','')),5)
#~ print Result
#~ print nbr_of_steps
print_at("reached energy after %f steps with config %s" % (
nbr_moves,str(encode_lattice(configuration)).replace(' ','')
),7)
MFPT.append(nbr_moves)
all_sequence_traversals.append(sequence_traversals)
print MFPT
#~ print 'sequceT' +str(all_sequence_traversals)
averageMFPT = ( sum(MFPT) / trajectories )
print averageMFPT
print "Configuration #%d; Temp = %f; Energy = %f; Average number of MC Steps = %d" % (
1,
temperature,
V,