DAMASK_EICMD/processing/pre/seeds_fromDistribution.py

279 lines
11 KiB
Python
Raw Normal View History

#!/usr/bin/python
# -*- coding: UTF-8 no BOM -*-
import threading,time,os,subprocess,shlex,string,sys,random
import numpy as np
from optparse import OptionParser
from operator import mul
from cStringIO import StringIO
import damask
scriptID = '$Id: yieldSurface.py 3390 2014-08-18 20:09:09Z MPIE\m.diehl $'
scriptName = scriptID.split()[1]
mismatch = None
currentSeedsName = None
def execute(cmd,streamIn=None,dir='./'):
initialPath=os.getcwd()
os.chdir(dir)
process = subprocess.Popen(shlex.split(cmd),stdout=subprocess.PIPE,stderr = subprocess.PIPE,stdin=subprocess.PIPE)
if streamIn != None:
out,error = process.communicate(streamIn.read())
else:
out,error = process.communicate()
os.chdir(initialPath)
return out,error
#---------------------------------------------------------------------------------------------------
class myThread (threading.Thread):
#---------------------------------------------------------------------------------------------------
'''
Runner class
'''
def __init__(self, threadID):
threading.Thread.__init__(self)
self.threadID = threadID
def run(self):
global mismatch
global bestSeedsUpdate
global bestSeedsVFile
global nMicrostructures
global delta
global points
global target
global match
global baseFile
s.acquire()
myMatch = match
s.release()
random.seed(options.randomSeed+self.threadID)
knownSeedsUpdate = bestSeedsUpdate -1.0
randReset = True
myBestSeedsVFile = StringIO()
perturbedSeedsVFile = StringIO()
perturbedGeomVFile = StringIO()
while not match:
# if a better seed file exist, read it into a virtual file
s.acquire()
if bestSeedsUpdate > knownSeedsUpdate:
print 'Thread %i: '%self.threadID+'Reading new best seeds from', bestSeedsUpdate
knownSeedsUpdate = bestSeedsUpdate
bestSeedsVFile.reset()
myBestSeedsVFile.close()
myBestSeedsVFile = StringIO()
i=0
for line in bestSeedsVFile:
myBestSeedsVFile.write(line)
s.release()
# new direction if current one is not good
if randReset:
selectedMs = random.randrange(1,nMicrostructures)
direction = np.array(((random.random()-0.5)*delta[0],
(random.random()-0.5)*delta[1],
(random.random()-0.5)*delta[2]))
randReset = False
# set virtual file pointer to the beginning
perturbedSeedsVFile.close()
perturbedSeedsVFile = StringIO()
myBestSeedsVFile.reset()
# perturbe current best fitting seed file (virtual file lastSeedsVFile)
perturbedSeedsTable = damask.ASCIItable(myBestSeedsVFile,
perturbedSeedsVFile,
labels=True)
perturbedSeedsTable.head_read() # read ASCII header info
perturbedSeedsTable.head_write()
outputAlive=True
ms = 1
while outputAlive and perturbedSeedsTable.data_read():
if ms == selectedMs:
direction+=direction
newCoords=np.array(tuple(map(float,perturbedSeedsTable.data[0:3]))+direction)
np.where(newCoords>=1.0,newCoords-1.0,newCoords)
np.where(newCoords <1.0,newCoords+1.0,newCoords)
#print ms,perturbedSeedsTable.data[0:3],'-->',[format(f, '8.6f') for f in newCoords]
perturbedSeedsTable.data[0:3]=[format(f, '8.6f') for f in newCoords]
ms+=1
perturbedSeedsTable.data_write()
# do tesselation with perturbed seed file and evaluate geometry
perturbedGeomVFile.close()
perturbedGeomVFile = StringIO()
perturbedSeedsVFile.reset()
perturbedGeomVFile.write(execute('geom_fromVoronoiTessellation '+
' -g '+' '.join(map(str, options.grid)),streamIn=perturbedSeedsVFile)[0])
perturbedGeomVFile.reset()
perturbedGeomTable = damask.ASCIItable(perturbedGeomVFile,labels=False)
perturbedGeomTable.head_read()
for i in perturbedGeomTable.info:
if i.startswith('microstructures'): myNmicrostructures = int(i.split('\t')[1])
perturbedGeomTable.data_readArray()
perturbedGeomTable.output_flush()
currentData=np.bincount(perturbedGeomTable.data.astype(int).ravel())[1:]/points
currentError=[]
currentHist=[]
for i in xrange(nMicrostructures):
currentHist.append(np.histogram(currentData,bins=target[i]['bins'])[0])
currentError.append(np.sqrt(np.square(np.array(target[i]['histogram']-currentHist[i])).sum()))
s.acquire()
myMatch = match
print 'Thread %i: finished tessellation (%i microstructures)'%(self.threadID, myNmicrostructures)
if currentError[nMicrostructures-1] == 0.0:
print 'finished'
match = True
if match: return
if myNmicrostructures != nMicrostructures:
print 'microstructure mismatch %i'%myNmicrostructures
randReset = True
else:
for i in xrange(nMicrostructures):
ratio = target[i]['error']/currentError[i]
if currentError[i] > target[i]['error']:
print '%i decreased quality to %8.6f'%(i+2,ratio)
print 'target',target[i]['histogram']
print 'current',currentHist[i]
randReset = True
break
elif currentError[i] < target[i]['error']:
bestSeedsUpdate = time.time()
print '%i increased quality to %8.6f'%(i+2,ratio)
print 'target',target[i]['histogram']
print 'current',currentHist[i]
print 'Thread', self.threadID, 'new best fit. time,', bestSeedsUpdate
currentSeedsName = baseFile+'_'+str(bestSeedsUpdate).replace('.','-')
perturbedSeedsVFile.reset()
bestSeedsVFile.close()
bestSeedsVFile = StringIO()
with open(currentSeedsName+'.seeds','w') as currentSeedsFile:
for line in perturbedSeedsVFile:
currentSeedsFile.write(line)
bestSeedsVFile.write(line)
for j in xrange(nMicrostructures):
target[j]['error'] = currentError[j]
break
if i == nMicrostructures-1: print 'continue along trajectory '
s.release()
# --------------------------------------------------------------------
# MAIN
# --------------------------------------------------------------------
parser = OptionParser(option_class=damask.extendableOption, usage='%prog options [file[s]]', description = """
Monte Carlo simulation to produce seed file that gives same size distribution like given geometry file.
""", version=string.replace(scriptID,'\n','\\n')
)
parser.add_option('-s','--seeds', dest='seedFile', metavar='string',
help='name of the intial seed file. If not found, a new one is created [%default]')
parser.add_option('-g','--grid', dest='grid', type='int', nargs=3, metavar='int int int',
help='a,b,c grid of hexahedral box [%default]')
parser.add_option('-t','--threads', dest='threads', type='int', metavar='int',
help='number of parallel executions [%default]')
parser.add_option('-r', '--rnd', dest='randomSeed', type='int', metavar='int',
help='seed of random number generator [%default]')
parser.add_option('--target', dest='target', metavar='string',
help='name of the geom file with target distribution [%default]')
parser.add_option('--tolerance', dest='threshold', type='float', metavar='float',
help='stopping criterion [%default]')
parser.add_option('--scale', dest='scale',type='float', metavar='float',
help='maximum moving distance of perturbed seed in pixel [%default]')
parser.set_defaults(seedFile = 'seeds')
parser.set_defaults(grid = (64,64,64))
parser.set_defaults(threads = 2)
parser.set_defaults(randomSeed = None)
parser.set_defaults(target = 'geom')
parser.set_defaults(threshold = 1.0)
parser.set_defaults(scale = 1.0)
options = parser.parse_args()[0]
if options.randomSeed == None:
options.randomSeed = int(os.urandom(4).encode('hex'), 16)
print 'random seed', options.randomSeed
delta = (options.scale/options.grid[0],options.scale/options.grid[1],options.scale/options.grid[2])
baseFile=os.path.splitext(os.path.basename(options.seedFile))[0]
points = float(reduce(mul,options.grid))
# ----------- calculate target distribution and bin edges
targetGeomVFile=StringIO()
with open(os.path.splitext(os.path.basename(options.target))[0]+'.geom') as targetGeomFile:
targetGeomTable = damask.ASCIItable(targetGeomFile,targetGeomVFile,labels=False)
targetGeomTable.head_read()
targetGeomTable.head_write()
for i in targetGeomTable.info:
if i.startswith('microstructures'): nMicrostructures = int(i.split()[1])
if i.startswith('grid'): targetPoints = float(int(i.split()[2])*\
int(i.split()[4])*\
int(i.split()[6]))
targetGeomTable.data_readArray()
targetGeomTable.data_writeArray()
targetData = np.bincount(targetGeomTable.data.astype(int).ravel())[1:nMicrostructures]/targetPoints
target=[]
for i in xrange(1,nMicrostructures+1):
targetHist,targetBins = np.histogram(targetData,bins=i)
target.append({'histogram':targetHist,'bins':targetBins})
targetGeomVFile.close()
# ----------- create initial seed file or open existing one
bestSeedsVFile = StringIO()
if os.path.isfile(os.path.splitext(options.seedFile)[0]+'.seeds'):
with open(os.path.splitext(options.seedFile)[0]+'.seeds') as initialSeedFile:
for line in initialSeedFile: bestSeedsVFile.write(line)
else:
bestSeedsVFile.write(execute('seeds_fromRandom'+\
' -g '+' '.join(map(str, options.grid))+\
' -r %i'%options.randomSeed+\
' -N '+str(nMicrostructures))[0])
bestSeedsUpdate = time.time()
# ----------- tessellate initial seed file to get and evaluate geom file
bestSeedsVFile.reset()
initialGeomVFile = StringIO()
initialGeomVFile.write(execute('geom_fromVoronoiTessellation '+
' -g '+' '.join(map(str, options.grid)),bestSeedsVFile)[0])
initialGeomVFile.reset()
initialGeomTable = damask.ASCIItable(initialGeomVFile,labels=False)
initialGeomTable.head_read()
for i in initialGeomTable.info:
if i.startswith('microstructures'): initialMicrostructures = int(i.split('\t')[1])
if initialMicrostructures != nMicrostructures: print 'error. Microstructure count mismatch'
initialGeomTable.data_readArray()
initialData = np.bincount(initialGeomTable.data.astype(int).ravel())[1:]/points
for i in xrange(nMicrostructures):
initialHist = np.histogram(initialData,bins=target[i]['bins'])[0]
target[i]['error']=np.sqrt(np.square(np.array(target[i]['histogram']-initialHist)).sum())
print target[i]['histogram']
print initialHist
print target[i]['error'],'\n-----------------------------------'
match = target[nMicrostructures-1]==0.0
initialGeomVFile.close()
# strart mulithreaded monte carlo simulation
threads=[]
s=threading.Semaphore(1)
for i in range(options.threads):
threads.append(myThread(i))
threads[i].start()
for i in range(options.threads):
threads[i].join()