works for dtype='float'
This commit is contained in:
parent
875643c363
commit
fc1f65c34e
|
@ -1,9 +1,11 @@
|
||||||
#!/usr/bin/env python3
|
#!/usr/bin/env python3
|
||||||
# -*- coding: UTF-8 no BOM -*-
|
# -*- coding: UTF-8 no BOM -*-
|
||||||
|
|
||||||
import os,sys,math
|
import os
|
||||||
|
import sys
|
||||||
import numpy as np
|
import numpy as np
|
||||||
import damask
|
import damask
|
||||||
|
from io import StringIO
|
||||||
from scipy import ndimage
|
from scipy import ndimage
|
||||||
from optparse import OptionParser
|
from optparse import OptionParser
|
||||||
|
|
||||||
|
@ -11,7 +13,8 @@ scriptName = os.path.splitext(os.path.basename(__file__))[0]
|
||||||
scriptID = ' '.join([scriptName,damask.version])
|
scriptID = ' '.join([scriptName,damask.version])
|
||||||
|
|
||||||
def mostFrequent(arr):
|
def mostFrequent(arr):
|
||||||
return np.argmax(np.bincount(arr.astype('int')))
|
unique, inverse = np.unique(arr, return_inverse=True)
|
||||||
|
return unique[np.argmax(np.bincount(inverse))]
|
||||||
|
|
||||||
|
|
||||||
#--------------------------------------------------------------------------------------------------
|
#--------------------------------------------------------------------------------------------------
|
||||||
|
@ -29,8 +32,7 @@ parser.add_option('-s','--stencil',
|
||||||
type = 'int', metavar = 'int',
|
type = 'int', metavar = 'int',
|
||||||
help = 'size of smoothing stencil [%default]')
|
help = 'size of smoothing stencil [%default]')
|
||||||
|
|
||||||
parser.set_defaults(stencil = 3,
|
parser.set_defaults(stencil = 3)
|
||||||
)
|
|
||||||
|
|
||||||
(options, filenames) = parser.parse_args()
|
(options, filenames) = parser.parse_args()
|
||||||
|
|
||||||
|
@ -40,55 +42,23 @@ parser.set_defaults(stencil = 3,
|
||||||
if filenames == []: filenames = [None]
|
if filenames == []: filenames = [None]
|
||||||
|
|
||||||
for name in filenames:
|
for name in filenames:
|
||||||
try: table = damask.ASCIItable(name = name,
|
|
||||||
buffered = False,
|
|
||||||
labeled = False)
|
|
||||||
except: continue
|
|
||||||
damask.util.report(scriptName,name)
|
damask.util.report(scriptName,name)
|
||||||
|
|
||||||
|
if name is None:
|
||||||
|
virt_file = StringIO(''.join(sys.stdin.read()))
|
||||||
|
geom = damask.Geom.from_file(virt_file)
|
||||||
|
else:
|
||||||
|
geom = damask.Geom.from_file(name)
|
||||||
|
microstructure = geom.microstructure
|
||||||
|
|
||||||
# --- interpret header ----------------------------------------------------------------------------
|
microstructure = ndimage.filters.generic_filter(microstructure,mostFrequent,
|
||||||
|
size=(options.stencil,)*3).astype(microstructure.dtype)
|
||||||
|
|
||||||
table.head_read()
|
geom.microstructure = microstructure
|
||||||
info,extra_header = table.head_getGeom()
|
geom.add_comment(scriptID + ' ' + ' '.join(sys.argv[1:]))
|
||||||
damask.util.report_geom(info)
|
|
||||||
|
damask.util.croak(geom)
|
||||||
errors = []
|
if name is None:
|
||||||
if np.any(info['grid'] < 1): errors.append('invalid grid a b c.')
|
sys.stdout.write(str(geom.show()))
|
||||||
if np.any(info['size'] <= 0.0): errors.append('invalid size x y z.')
|
else:
|
||||||
if errors != []:
|
geom.to_file(name)
|
||||||
damask.util.croak(errors)
|
|
||||||
table.close(dismiss = True)
|
|
||||||
continue
|
|
||||||
|
|
||||||
# --- read data ------------------------------------------------------------------------------------
|
|
||||||
|
|
||||||
microstructure = table.microstructure_read(info['grid']).reshape(info['grid'],order='F') # read microstructure
|
|
||||||
|
|
||||||
# --- do work ------------------------------------------------------------------------------------
|
|
||||||
|
|
||||||
microstructure = ndimage.filters.generic_filter(microstructure,mostFrequent,size=(options.stencil,)*3).astype('int_')
|
|
||||||
newInfo = {'microstructures': len(np.unique(microstructure))}
|
|
||||||
|
|
||||||
# --- report ---------------------------------------------------------------------------------------
|
|
||||||
if ( newInfo['microstructures'] != info['microstructures']):
|
|
||||||
damask.util.croak('--> microstructures: %i'%newInfo['microstructures'])
|
|
||||||
info['microstructures'] == newInfo['microstructures']
|
|
||||||
|
|
||||||
# --- write header ---------------------------------------------------------------------------------
|
|
||||||
|
|
||||||
table.info_clear()
|
|
||||||
table.info_append([scriptID + ' ' + ' '.join(sys.argv[1:]),])
|
|
||||||
table.head_putGeom(info)
|
|
||||||
table.info_append([extra_header])
|
|
||||||
table.labels_clear()
|
|
||||||
table.head_write()
|
|
||||||
|
|
||||||
# --- write microstructure information ------------------------------------------------------------
|
|
||||||
|
|
||||||
formatwidth = int(math.floor(math.log10(np.nanmax(microstructure))+1))
|
|
||||||
table.data = microstructure.reshape((info['grid'][0],np.prod(info['grid'][1:])),order='F').transpose()
|
|
||||||
table.data_writeArray('%{}i'.format(formatwidth),delimiter = ' ')
|
|
||||||
|
|
||||||
# --- output finalization --------------------------------------------------------------------------
|
|
||||||
|
|
||||||
table.close() # close ASCII table
|
|
||||||
|
|
|
@ -1,9 +1,11 @@
|
||||||
#!/usr/bin/env python3
|
#!/usr/bin/env python3
|
||||||
# -*- coding: UTF-8 no BOM -*-
|
# -*- coding: UTF-8 no BOM -*-
|
||||||
|
|
||||||
import os,sys,math
|
import os
|
||||||
|
import sys
|
||||||
import numpy as np
|
import numpy as np
|
||||||
from optparse import OptionParser
|
from optparse import OptionParser
|
||||||
|
from scipy import ndimage
|
||||||
import damask
|
import damask
|
||||||
|
|
||||||
scriptName = os.path.splitext(os.path.basename(__file__))[0]
|
scriptName = os.path.splitext(os.path.basename(__file__))[0]
|
||||||
|
@ -27,137 +29,44 @@ parser.add_option('-s', '--size',
|
||||||
dest = 'size',
|
dest = 'size',
|
||||||
type = 'string', nargs = 3, metavar = 'string string string',
|
type = 'string', nargs = 3, metavar = 'string string string',
|
||||||
help = 'x,y,z size of hexahedral box [unchanged]')
|
help = 'x,y,z size of hexahedral box [unchanged]')
|
||||||
parser.add_option('-r', '--renumber',
|
|
||||||
dest = 'renumber',
|
|
||||||
action = 'store_true',
|
|
||||||
help = 'renumber microstructure indices from 1..N [%default]')
|
|
||||||
parser.add_option('--float',
|
|
||||||
dest = 'float',
|
|
||||||
action = 'store_true',
|
|
||||||
help = 'use float input')
|
|
||||||
|
|
||||||
parser.set_defaults(renumber = False,
|
|
||||||
grid = ['0','0','0'],
|
|
||||||
size = ['0.0','0.0','0.0'],
|
|
||||||
float = False,
|
|
||||||
)
|
|
||||||
|
|
||||||
(options, filenames) = parser.parse_args()
|
(options, filenames) = parser.parse_args()
|
||||||
|
|
||||||
datatype = 'f' if options.float else 'i'
|
|
||||||
|
|
||||||
# --- loop over input files -------------------------------------------------------------------------
|
# --- loop over input files -------------------------------------------------------------------------
|
||||||
|
|
||||||
if filenames == []: filenames = [None]
|
if filenames == []: filenames = [None]
|
||||||
|
|
||||||
for name in filenames:
|
for name in filenames:
|
||||||
try:
|
|
||||||
table = damask.ASCIItable(name = name,
|
|
||||||
buffered = False,
|
|
||||||
labeled = False)
|
|
||||||
except: continue
|
|
||||||
damask.util.report(scriptName,name)
|
damask.util.report(scriptName,name)
|
||||||
|
|
||||||
# --- interpret header ----------------------------------------------------------------------------
|
|
||||||
|
|
||||||
table.head_read()
|
|
||||||
info,extra_header = table.head_getGeom()
|
|
||||||
damask.util.report_geom(info)
|
|
||||||
|
|
||||||
errors = []
|
|
||||||
if np.any(info['grid'] < 1): errors.append('invalid grid a b c.')
|
|
||||||
if np.any(info['size'] <= 0.0): errors.append('invalid size x y z.')
|
|
||||||
if errors != []:
|
|
||||||
damask.util.croak(errors)
|
|
||||||
table.close(dismiss = True)
|
|
||||||
continue
|
|
||||||
|
|
||||||
# --- read data ------------------------------------------------------------------------------------
|
|
||||||
|
|
||||||
microstructure = table.microstructure_read(info['grid'],datatype) # read microstructure
|
|
||||||
|
|
||||||
# --- do work ------------------------------------------------------------------------------------
|
|
||||||
|
|
||||||
newInfo = {
|
|
||||||
'grid': np.zeros(3,'i'),
|
|
||||||
'origin': np.zeros(3,'d'),
|
|
||||||
'microstructures': 0,
|
|
||||||
}
|
|
||||||
|
|
||||||
newInfo['grid'] = np.array([{True: round(o*float(n.lower().replace('x',''))),
|
|
||||||
False: round(float(n.lower().replace('x','')))}[n[-1].lower() == 'x']
|
|
||||||
for o,n in zip(info['grid'],options.grid)],'i')
|
|
||||||
newInfo['size'] = np.array([{True: o*float(n.lower().replace('x','')),
|
|
||||||
False: float(n.lower().replace('x',''))}[n[-1].lower() == 'x']
|
|
||||||
for o,n in zip(info['size'],options.size)],'d')
|
|
||||||
newInfo['grid'] = np.where(newInfo['grid'] <= 0 , info['grid'],newInfo['grid'])
|
|
||||||
newInfo['size'] = np.where(newInfo['size'] <= 0.0, info['size'],newInfo['size'])
|
|
||||||
|
|
||||||
multiplicity = []
|
|
||||||
for j in range(3):
|
|
||||||
multiplicity.append([])
|
|
||||||
last = 0
|
|
||||||
for i in range(info['grid'][j]):
|
|
||||||
this = int((i+1)*float(newInfo['grid'][j])/info['grid'][j])
|
|
||||||
multiplicity[j].append(this-last)
|
|
||||||
last = this
|
|
||||||
|
|
||||||
microstructure = microstructure.reshape(info['grid'],order='F')
|
|
||||||
microstructure = np.repeat(np.repeat(np.repeat(microstructure,
|
|
||||||
multiplicity[0], axis=0),multiplicity[1], axis=1),multiplicity[2], axis=2)
|
|
||||||
# --- renumber to sequence 1...Ngrains if requested ------------------------------------------------
|
|
||||||
# http://stackoverflow.com/questions/10741346/np-frequency-counts-for-unique-values-in-an-array
|
|
||||||
|
|
||||||
if options.renumber:
|
|
||||||
newID = 0
|
|
||||||
for microstructureID,count in enumerate(np.bincount(microstructure.reshape(newInfo['grid'].prod()))):
|
|
||||||
if count != 0:
|
|
||||||
newID += 1
|
|
||||||
microstructure = np.where(microstructure == microstructureID, newID,microstructure).reshape(microstructure.shape)
|
|
||||||
|
|
||||||
newInfo['microstructures'] = len(np.unique(microstructure))
|
|
||||||
|
|
||||||
# --- report ---------------------------------------------------------------------------------------
|
|
||||||
|
|
||||||
remarks = []
|
|
||||||
errors = []
|
|
||||||
|
|
||||||
if (any(newInfo['grid'] != info['grid'])):
|
|
||||||
remarks.append('--> grid a b c: %s'%(' x '.join(map(str,newInfo['grid']))))
|
|
||||||
if (any(newInfo['size'] != info['size'])):
|
|
||||||
remarks.append('--> size x y z: %s'%(' x '.join(map(str,newInfo['size']))))
|
|
||||||
if ( newInfo['microstructures'] != info['microstructures']):
|
|
||||||
remarks.append('--> microstructures: %i'%newInfo['microstructures'])
|
|
||||||
|
|
||||||
if np.any(newInfo['grid'] < 1): errors.append('invalid new grid a b c.')
|
|
||||||
if np.any(newInfo['size'] <= 0.0): errors.append('invalid new size x y z.')
|
|
||||||
|
|
||||||
if remarks != []: damask.util.croak(remarks)
|
|
||||||
if errors != []:
|
|
||||||
damask.util.croak(errors)
|
|
||||||
table.close(dismiss = True)
|
|
||||||
continue
|
|
||||||
|
|
||||||
# --- write header ---------------------------------------------------------------------------------
|
|
||||||
|
|
||||||
table.info_clear()
|
|
||||||
table.info_append(extra_header+[
|
|
||||||
scriptID + ' ' + ' '.join(sys.argv[1:]),
|
|
||||||
"grid\ta {grid[0]}\tb {grid[1]}\tc {grid[2]}".format(grid=newInfo['grid']),
|
|
||||||
"size\tx {size[0]}\ty {size[1]}\tz {size[2]}".format(size=newInfo['size']),
|
|
||||||
"origin\tx {origin[0]}\ty {origin[1]}\tz {origin[2]}".format(origin=info['origin']),
|
|
||||||
"homogenization\t{homog}".format(homog=info['homogenization']),
|
|
||||||
"microstructures\t{microstructures}".format(microstructures=newInfo['microstructures']),
|
|
||||||
])
|
|
||||||
table.labels_clear()
|
|
||||||
table.head_write()
|
|
||||||
|
|
||||||
# --- write microstructure information ------------------------------------------------------------
|
|
||||||
|
|
||||||
format = '%g' if options.float else '%{}i'.format(int(math.floor(math.log10(np.nanmax(microstructure))+1)))
|
|
||||||
table.data = microstructure.reshape((newInfo['grid'][0],newInfo['grid'][1]*newInfo['grid'][2]),order='F').transpose()
|
|
||||||
table.data_writeArray(format,delimiter=' ')
|
|
||||||
|
|
||||||
# --- output finalization --------------------------------------------------------------------------
|
if name is None:
|
||||||
|
virt_file = StringIO(''.join(sys.stdin.read()))
|
||||||
|
geom = damask.Geom.from_file(virt_file)
|
||||||
|
else:
|
||||||
|
geom = damask.Geom.from_file(name)
|
||||||
|
microstructure = geom.microstructure
|
||||||
|
|
||||||
table.close() # close ASCII table
|
scale = geom.get_grid().astype('float')
|
||||||
|
if options.grid is not None:
|
||||||
|
for i,g in enumerate(options.grid):
|
||||||
|
scale[i] = scale[i]*float(g.lower().replace('x','')) if g.lower().startswith('x') \
|
||||||
|
else float(options.grid[i])/scale[i]
|
||||||
|
|
||||||
|
size = geom.get_size()
|
||||||
|
if options.size is not None:
|
||||||
|
for i,s in enumerate(options.size):
|
||||||
|
size[i] = size[i]*float(s.lower().replace('x','')) if s.lower().startswith('x') \
|
||||||
|
else options.size[i]
|
||||||
|
|
||||||
|
microstructure = ndimage.interpolation.zoom(microstructure, scale, output=microstructure.dtype,
|
||||||
|
order=0, mode='nearest', prefilter=False)
|
||||||
|
|
||||||
|
geom.microstructure = microstructure
|
||||||
|
geom.set_size(size)
|
||||||
|
geom.add_comment(scriptID + ' ' + ' '.join(sys.argv[1:]))
|
||||||
|
|
||||||
|
damask.util.croak(geom)
|
||||||
|
if name is None:
|
||||||
|
sys.stdout.write(str(geom.show()))
|
||||||
|
else:
|
||||||
|
geom.to_file(name)
|
||||||
|
|
Loading…
Reference in New Issue