DAMASK_EICMD/processing/pre/seeds_fromTable.py

130 lines
5.6 KiB
Python
Executable File

#!/usr/bin/env python
# -*- coding: UTF-8 no BOM -*-
import os,sys,string,itertools
import numpy as np
from optparse import OptionParser
from collections import defaultdict
import damask
scriptID = string.replace('$Id$','\n','\\n')
scriptName = os.path.splitext(scriptID.split()[1])[0]
#--------------------------------------------------------------------------------------------------
# MAIN
#--------------------------------------------------------------------------------------------------
parser = OptionParser(option_class=damask.extendableOption, usage='%prog options [file[s]]', description = """
Create seed file by taking microstructure indices from given ASCIItable column.
White and black-listing of microstructure indices is possible.
Examples:
--white 1,2,5 --index grainID isolates grainID entries of value 1, 2, and 5;
--black 1 --index grainID takes all grainID entries except for value 1.
""", version = scriptID)
parser.add_option('-p', '--positions',
dest = 'pos',
type = 'string', metavar = 'string',
help = 'coordinate label [%default]')
parser.add_option('--boundingbox',
dest = 'box',
type = 'float', nargs = 6, metavar = ' '.join(['float']*6),
help = 'min (x,y,z) and max (x,y,z) coordinates of bounding box [tight]')
parser.add_option('-i', '--index',
dest = 'index',
type = 'string', metavar = 'string',
help = 'microstructure index label [%default]')
parser.add_option('-w','--white',
dest = 'whitelist',
action = 'extend', metavar = '<int LIST>',
help = 'whitelist of microstructure indices')
parser.add_option('-b','--black',
dest = 'blacklist',
action = 'extend', metavar = '<int LIST>',
help = 'blacklist of microstructure indices')
parser.set_defaults(pos = 'pos',
index ='microstructure',
)
(options,filenames) = parser.parse_args()
if options.whitelist != None: options.whitelist = map(int,options.whitelist)
if options.blacklist != None: options.blacklist = map(int,options.blacklist)
# --- loop over input files -------------------------------------------------------------------------
if filenames == []: filenames = [None]
for name in filenames:
try:
table = damask.ASCIItable(name = name,
outname = os.path.splitext(name)[0]+'.seeds' if name else name,
buffered = False)
except: continue
table.croak('\033[1m'+scriptName+'\033[0m'+(': '+name if name else ''))
table.head_read() # read ASCII header info
# ------------------------------------------ sanity checks ---------------------------------------
missing_labels = table.data_readArray([options.pos,options.index])
errors = []
if len(missing_labels) > 0:
errors.append('column{} {} not found'.format('s' if len(missing_labels) > 1 else '',
', '.join(missing_labels)))
for label, dim in {options.pos: 3,
options.index: 1}.iteritems():
if table.label_dimension(label) != dim:
errors.append('column {} has wrong dimension'.format(label))
if errors != []:
table.croak(errors)
table.close(dismiss = True) # close ASCII table file handles and delete output file
continue
# ------------------------------------------ process data ------------------------------------------
# --- finding bounding box -------------------------------------------------------------------------
boundingBox = np.array((np.amin(table.data[:,0:3],axis = 0),np.amax(table.data[:,0:3],axis = 0)))
if options.box:
boundingBox[0,:] = np.minimum(options.box[0:3],boundingBox[0,:])
boundingBox[1,:] = np.maximum(options.box[3:6],boundingBox[1,:])
# --- rescaling coordinates ------------------------------------------------------------------------
table.data[:,0:3] -= boundingBox[0,:]
table.data[:,0:3] /= boundingBox[1,:]-boundingBox[0,:]
# --- filtering of grain voxels --------------------------------------------------------------------
mask = np.logical_and(\
np.ones_like(table.data[:,3],bool) \
if options.whitelist == None \
else np.in1d(table.data[:,3].ravel(), options.whitelist).reshape(table.data[:,3].shape),
np.ones_like(table.data[:,3],bool) \
if options.blacklist == None \
else np.invert(np.in1d(table.data[:,3].ravel(), options.blacklist).reshape(table.data[:,3].shape))
)
table.data = table.data[mask]
# ------------------------------------------ assemble header ---------------------------------------
table.info = [
scriptID,
'size %s'%(' '.join(list(itertools.chain.from_iterable(zip(['x','y','z'],
map(str,boundingBox[1,:]-boundingBox[0,:])))))),
]
table.labels_clear()
table.labels_append(['1_pos','2_pos','3_pos','microstructure']) # implicitly switching label processing/writing on
table.head_write()
# ------------------------------------------ output result ---------------------------------------
table.data_writeArray()
table.close() # close ASCII tables