added options for selective seed picking based on Mitchell’s best candidate algorithm for more uniformly distributed (spatially) seeds points

This commit is contained in:
Tias Maiti 2015-05-26 20:13:35 +00:00
parent 4107d5d1d2
commit 73c6bd767f
1 changed files with 68 additions and 22 deletions

View File

@ -3,8 +3,10 @@
import os,sys,string,math,random
import numpy as np
from optparse import OptionParser
import damask
from optparse import OptionParser,OptionGroup
from scipy import spatial
scriptID = string.replace('$Id$','\n','\\n')
scriptName = os.path.splitext(scriptID.split()[1])[0]
@ -33,6 +35,17 @@ parser.add_option('--sigma', dest='sigma', type='float', metavar='float', \
help='standard deviation of Gaussian Distribution for weights [%default]')
parser.add_option('-m', '--microstructure', dest='microstructure', type='int',
help='first microstructure index [%default]', metavar='int')
parser.add_option('-s','--selective', dest='selective', action='store_true',
help = 'selective picking of seed points from random seed points [%default]')
group = OptionGroup(parser, "Selective Seeding Options",
"More uniform distribution of seed points using Mitchells Best Candidate Algorithm"
)
group.add_option('--distance', dest='bestDistance', type='float', metavar='float', \
help='minimum distance to the next neighbor [%default]')
group.add_option('--numCandidates', dest='numCandidates', type='int', metavar='int', \
help='maximum number of point to consider for initial random points generation [%default]')
parser.add_option_group(group)
parser.set_defaults(randomSeed = None)
parser.set_defaults(grid = (16,16,16))
parser.set_defaults(N = 20)
@ -40,6 +53,10 @@ parser.set_defaults(weights=False)
parser.set_defaults(mean = 0.0)
parser.set_defaults(sigma = 1.0)
parser.set_defaults(microstructure = 1)
parser.set_defaults(selective = False)
parser.set_defaults(bestDistance = 0.2)
parser.set_defaults(numCandidates = 10)
(options,filename) = parser.parse_args()
@ -47,6 +64,18 @@ options.grid = np.array(options.grid)
labels = "1_coords\t2_coords\t3_coords\tphi1\tPhi\tphi2\tmicrostructure"
# ------------------------------------------ Functions Definitions ---------------------------------
def kdtree_search(xyz, point) :
dist, index = spatial.cKDTree(xyz).query(np.array(point))
return dist
def generatePoint() :
return np.array([random.uniform(0,options.grid[0]/max(options.grid)), \
random.uniform(0,options.grid[1]/max(options.grid)), \
random.uniform(0,options.grid[2]/max(options.grid))])
# ------------------------------------------ setup file handle -------------------------------------
if filename == []:
file = {'output':sys.stdout, 'croak':sys.stderr}
@ -69,6 +98,9 @@ grainEuler[0,:] *= 360.0
grainEuler[1,:] = np.arccos(2*grainEuler[1,:]-1)*180.0/math.pi # cos(Phi) is uniformly distributed
grainEuler[2,:] *= 360.0 # phi_2 is uniformly distributed
microstructure=np.arange(options.microstructure,options.microstructure+options.N).reshape(1,options.N)
if options.selective == False :
seedpoints = -np.ones(options.N,dtype='int') # init grid positions of seed points
if options.N * 1024 < gridSize: # heuristic limit for random search
@ -88,9 +120,23 @@ seeds[1,:] = (np.mod(seedpoints// options.grid[0] ,options.grid[
+np.random.random())/options.grid[1]
seeds[2,:] = (np.mod(seedpoints//(options.grid[1]*options.grid[0]),options.grid[2])\
+np.random.random())/options.grid[2]
microstructure=np.arange(options.microstructure,options.microstructure+options.N).reshape(1,options.N)
table = np.transpose(np.concatenate((seeds,grainEuler,microstructure),axis = 0))
else :
samples = generatePoint().reshape(1,3)
while samples.shape[0] < options.N :
bestDistance = options.bestDistance
for i in xrange(options.numCandidates) :
c = generatePoint()
d = kdtree_search(samples, c)
if (d > bestDistance) :
bestDistance = d
bestCandidate = c
if kdtree_search(samples,bestCandidate) != 0.0 :
samples = np.append(samples,bestCandidate.reshape(1,3),axis=0)
else :
continue
table = np.transpose(np.concatenate((samples.T,grainEuler,microstructure),axis = 0))
if options.weights :
weight = np.random.normal(loc=options.mean, scale=options.sigma, size=options.N)
@ -98,7 +144,7 @@ if options.weights :
table = np.append(table, weight.reshape(options.N,1), axis=1)
labels += "\tweight"
# -------------------------------------- Write Data --------------------------------------------------
header = ["5\theader",
scriptID + " " + " ".join(sys.argv[1:]),