def fussf(pop, nkeep, Optimizer):
"""Selection function to employ basic Fixed Uniform Selection Scheme to select and
order individuals in a population based on fingerprint distance values.
Picks one fuss point for the selection process
FUSS point must reside within fusslimit
"""
fusslimit = Optimizer.fusslimit
#FUSS for fingerprint
newpop = []
# Collect fitnesses
fits = [ind.fitness for ind in pop]
minindex = min(xrange(len(fits)), key=fits.__getitem__)
fits = []
STR = 'Distances: \n'
for i in range(len(pop)):
dist = fingerprint_dist(pop[minindex].fingerprint, pop[i].fingerprint)
fits.append(dist)
STR += repr(dist) + '\n'
#Find min and max fitness
minf = min(fits)
maxf = max(fits)
if abs(maxf - minf) > fusslimit:
maxf = minf + fusslimit
#Select random point on fitness line
pt = random.uniform(minf, maxf)
#Calculate the distance of each individual's fitness from that point
i = 0
distances = []
for fit in fits:
distances.append((abs(pt - fit), i))
i += 1
#Sort distances from min to max
dlist = sorted(distances, key=lambda one: one[0])
#Select individuals with lowest distance
indices = []
for i in range(nkeep):
newpop.append(pop[dlist[i][1]])
indices.append(dlist[i][1])
#Always keep lowest energy individual
if minindex not in indices:
rm = random.choice(indices)
newpop = [pop[inx] for inx in indices if inx != rm]
newpop.append(pop[minindex])
return newpop
def fussf(pop, nkeep, Optimizer):
"""Selection function to employ basic Fixed Uniform Selection Scheme to select and
order individuals in a population based on fingerprint distance values.
Picks one fuss point for the selection process
FUSS point must reside within fusslimit
"""
fusslimit = Optimizer.fusslimit
# FUSS for fingerprint
newpop = []
# Collect fitnesses
fits = [ind.fitness for ind in pop]
minindex = min(xrange(len(fits)), key=fits.__getitem__)
fits = []
STR = "Distances: \n"
for i in range(len(pop)):
dist = fingerprint_dist(pop[minindex].fingerprint, pop[i].fingerprint)
fits.append(dist)
STR += repr(dist) + "\n"
# Find min and max fitness
minf = min(fits)
maxf = max(fits)
if abs(maxf - minf) > fusslimit:
maxf = minf + fusslimit
# Select random point on fitness line
pt = random.uniform(minf, maxf)
# Calculate the distance of each individual's fitness from that point
i = 0
distances = []
for fit in fits:
distances.append((abs(pt - fit), i))
i += 1
# Sort distances from min to max
dlist = sorted(distances, key=lambda one: one[0])
# Select individuals with lowest distance
indices = []
for i in range(nkeep):
newpop.append(pop[dlist[i][1]])
indices.append(dlist[i][1])
# Always keep lowest energy individual
if minindex not in indices:
rm = random.choice(indices)
newpop = [pop[inx] for inx in indices if inx != rm]
newpop.append(pop[minindex])
return newpop
def generation_eval(self, pop):
global logger
emx = max(ind.energy for ind in pop)
emn = min(ind.energy for ind in pop)
for ind in pop:
ind.tenergymx = emx
ind.tenergymin = emn
#DEBUG: Write relaxed individual
if 'MA' in self.debug:
if self.generation > 0:
inp_out.write_xyz(self.debugfile,pop[self.nindiv][0],\
'First Relaxed Offspring '+repr(pop[self.nindiv-1].energy))
#DEBUG: Write relaxed ind in solid
if self.structure=='Defect' or self.structure=='Surface':
inp_out.write_xyz(self.debugfile,pop[self.nindiv].bulki,\
'First Relaxed bulki '+repr(pop[self.nindiv-1].energy))
sols = pop[self.nindiv][0].copy()
sols.extend(pop[self.nindiv].bulki)
inp_out.write_xyz(self.debugfile,sols,'First from Invalid-ind + Bulki '+\
repr(pop[self.nindiv].energy))
sols = pop[self.nindiv][0].copy()
sols.extend(pop[self.nindiv].bulko)
inp_out.write_xyz(self.debugfile,sols,\
'First from Invalid-ind + Bulko '+repr(pop[self.nindiv].energy))
if self.generation==0:
logger.info('Initializing Bests list')
self.BESTS = list()
if self.best_inds_list:
self.BESTS = tools.BestInds(pop,self.BESTS,self,writefile=True)
# Determine survival based on fitness predator
if 'lambda,mu' not in self.algorithm_type:
pop = tools.get_best(pop, len(pop))
if self.fingerprinting:
logger.info('Writing fingerprint files')
for one in pop:
self.fpfile.write(repr(fingerprinting.fingerprint_dist(
pop[0].fingerprint,one.fingerprint))+' '+repr(one.energy)+' ')
self.fpfile.write('\n')
self.fpminfile.write(repr(pop[0].fingerprint)+'\n')
self.fpminfile.write(repr(pop[0].energy)+'\n')
nevals = len(pop)/2
if self.generation !=0:
logger.info('Applying predator')
pop = predator_switch(pop,self)
else:
self.genrep = 0
self.minfit = 0
# Evaluate population
logger.info('Checking population for convergence')
self.check_pop(pop)
#Update general output tracking
if self.generation !=0:
histlist = []
for ind in pop:
histlist.append(ind.history_index)
self.Runtimes.append(time.time())
self.Evaluations.append(nevals)
cxsuccess = 0
mutsuccess = []
for one in histlist:
if '+' in one:
cxsuccess +=1
if 'm' in one:
mutsuccess.append(one)
self.CXs.append((self.cxattempts,cxsuccess))
mutslist = [[0,0] for one in self.mutation_options]
for one in mutsuccess:
for two, opt in self.mutattempts:
if one==two:
index = [ind for ind,value in enumerate(
self.mutation_options) if value==opt][0]
mutslist[index][1]+=1
for one,opt in self.mutattempts:
index = [ind for ind,value in enumerate(
self.mutation_options) if value==opt][0]
mutslist[index][0]+=1
self.Muts.append(mutslist)
self.output.write('\n----- Generation Stats -----\n')
self.output.write('Attempted Crossovers: ' + repr(self.cxattempts)+'\n')
self.output.write('Successful Crossovers: ' + repr(cxsuccess)+'\n')
self.output.write('Mutations:\n')
i=0
for opt in self.mutation_options:
self.output.write(' Attempted ' + opt + ' : ' + repr(mutslist[i][0]) + '\n')
self.output.write(' Successful ' + opt + ' : ' + repr(mutslist[i][1]) + '\n')
i+=1
self.generation += 1
# Set new index values
index1 = 0
for ind in pop:
ind.index = index1
index1+=1
index1 = 0
if not self.convergence:
try:
self.calc.clean()
if self.fixed_region:
self.static_calc.clean()
except:
pass
if self.lammps_keep_files:
try:
if self.parallel and ('Island_Method' not in self.algorithm_type):
nproc = MPI.COMM_WORLD.Get_size()
lmpfilepath = os.path.dirname(self.calc.tmp_dir)
for proc in range(nproc):
pth = os.path.join(lmpfilepath,'rank-{0}'.format(proc))
fls = [fl for fl in os.listdir(pth) if '.' not in fl]
for one in fls:
os.remove(pth+'/'+one)
else:
fls = [fl for fl in os.listdir(self.calc.tmp_dir) if '.' not in fl]
for one in fls:
os.remove(self.calc.tmp_dir+'/'+one)
except Exception, e:
logger.error('Print issue in removing files {0}.'.format(e),exc_info=True)
print str(e)
pass
def fingerprint_niche(pop, Optimizer):
"""Use a k-means clustering approach to identify individuals for new population"""
STR = ''
#Get coordinates for each individual
pts = []
for one in pop:
fpd = fingerprint_dist(pop[0].fingerprint, one.fingerprint)
pts.append([one.fitness - pop[0].fitness, fpd])
for one in pts:
if numpy.isnan(one[1]):
one[1] = 0
#Identify spatial bounds
mins = min(pts)
maxs = max(pts)
attemptcount = 10
passflag = False
while attemptcount > 0:
#Select random initial centroid locations
cents = []
#for i in range(Optimizer.nindiv):
# cents.append([random.uniform(mins[0],maxs[0]),random.uniform(mins[1],maxs[1])])
while len(cents) < Optimizer.nindiv:
a = random.choice(pts)
if a not in cents:
cents.append(a)
#Assign individual to closest centroid
clustlist = [[] for i in range(Optimizer.nindiv)]
for one in pts:
ds = []
for i in range(len(cents)):
d = ((one[0] - cents[i][0])**2 +
(one[1] - cents[i][1])**2)**0.5
ds.append([d, i])
ds = sorted(ds, key=lambda two: two[0])
loc = ds[0]
clustlist[loc[1]].append(one)
try:
count = 0
while True:
count += 1
#Calculate the centroid of the new cluster
centsnew = []
for one in clustlist:
if len(one) > 0:
m = [1.0] * len(one)
#cop = numpy.dot(m, one) / sum(m)
cop = [
sum([es for es, fps in one]) / sum(m),
sum([fps for es, fps in one]) / sum(m)
]
centsnew.append(cop)
else:
centsnew.append(one)
#Assign individuals to new centroids
clustlistn = [[] for i in range(Optimizer.nindiv)]
for one in pts:
ds = []
for i in range(len(centsnew)):
d = ((one[0] - centsnew[i][0])**2 +
(one[1] - centsnew[i][1])**2)**0.5
ds.append([d, i])
ds = sorted(ds, key=lambda two: two[0])
loc = ds[0]
clustlistn[loc[1]].append(one)
if clustlistn == clustlist:
clustlist = clustlistn
centroids = centsnew
attemptcount = -10
passflag = True
break
else:
clustlist = clustlistn
except:
attemptcount -= 1
STR += 'WARNING: PREDATOR: K-means cluster difficulty\n'
if passflag == True:
#Select one individual from each cluster
news = []
for one in clustlist:
#Random
#npop.append(random.choice(one))
#Minimum energy
opts = sorted(one, key=lambda two: two[0])
news.append(opts[0])
indices = []
for one in news:
for i in range(len(pts)):
if pts[i] == one:
indices.append(i)
break
npop = [pop[ind] for ind in indices]
else:
STR += 'WARNING: PREDATOR: Population converging. Unable to find nindiv distinct clusters\n'
npop = mutation_dups(pop, Optimizer)
#npop = pop
STR += 'WARNING: PREDATOR: Chose new population based on Mutation-Dups predator\n'
pop = get_best(npop, len(npop))
return pop, STR
def fingerprint_niche(pop, Optimizer):
"""Use a k-means clustering approach to identify individuals for new population"""
STR = ''
#Get coordinates for each individual
pts = []
for one in pop:
fpd = fingerprint_dist(pop[0].fingerprint,one.fingerprint)
pts.append([one.fitness - pop[0].fitness,fpd])
for one in pts:
if numpy.isnan(one[1]):
one[1] = 0
#Identify spatial bounds
mins = min(pts)
maxs = max(pts)
attemptcount=10
passflag=False
while attemptcount > 0:
#Select random initial centroid locations
cents = []
#for i in range(Optimizer.nindiv):
# cents.append([random.uniform(mins[0],maxs[0]),random.uniform(mins[1],maxs[1])])
while len(cents) < Optimizer.nindiv:
a = random.choice(pts)
if a not in cents:
cents.append(a)
#Assign individual to closest centroid
clustlist = [[] for i in range(Optimizer.nindiv)]
for one in pts:
ds = []
for i in range(len(cents)):
d = ((one[0] - cents[i][0])**2 + (one[1] - cents[i][1])**2)**0.5
ds.append([d,i])
ds = sorted(ds, key=lambda two: two[0])
loc = ds[0]
clustlist[loc[1]].append(one)
try:
count = 0
while True:
count += 1
#Calculate the centroid of the new cluster
centsnew = []
for one in clustlist:
if len(one) > 0:
m = [1.0]*len(one)
#cop = numpy.dot(m, one) / sum(m)
cop = [sum([es for es,fps in one]) / sum(m),sum([fps for es,fps in one]) / sum(m)]
centsnew.append(cop)
else:
centsnew.append(one)
#Assign individuals to new centroids
clustlistn=[[] for i in range(Optimizer.nindiv)]
for one in pts:
ds=[]
for i in range(len(centsnew)):
d = ((one[0] - centsnew[i][0])**2 + (one[1] - centsnew[i][1])**2)**0.5
ds.append([d,i])
ds = sorted(ds, key=lambda two: two[0])
loc = ds[0]
clustlistn[loc[1]].append(one)
if clustlistn==clustlist:
clustlist = clustlistn
centroids = centsnew
attemptcount=-10
passflag=True
break
else:
clustlist = clustlistn
except:
attemptcount-=1
STR+= 'WARNING: PREDATOR: K-means cluster difficulty\n'
if passflag==True:
#Select one individual from each cluster
news = []
for one in clustlist:
#Random
#npop.append(random.choice(one))
#Minimum energy
opts = sorted(one, key=lambda two: two[0])
news.append(opts[0])
indices = []
for one in news:
for i in range(len(pts)):
if pts[i]==one:
indices.append(i)
break
npop = [pop[ind] for ind in indices]
else:
STR+='WARNING: PREDATOR: Population converging. Unable to find nindiv distinct clusters\n'
npop = mutation_dups(pop,Optimizer)
#npop = pop
STR+='WARNING: PREDATOR: Chose new population based on Mutation-Dups predator\n'
pop = get_best(npop,len(npop))
return pop, STR
