DAMASK_EICMD/processing/post/averageTable.py

83 lines
3.3 KiB
Python
Executable File

#!/usr/bin/env python
# -*- coding: UTF-8 no BOM -*-
import os,re,sys,string
import numpy as np
from optparse import OptionParser
import damask
scriptID = string.replace('$Id: AverageTable.py 3878 2015-02-22 19:26:52Z t.maiti $','\n','\\n')
scriptName = os.path.splitext(scriptID.split()[1])[0]
# --------------------------------------------------------------------
# MAIN
# --------------------------------------------------------------------
parser = OptionParser(option_class=damask.extendableOption, usage='%prog options [file[s]]', description = """
Replace all rows of a column by a single row containing their average.
Examples:
For grain averaged values , replace all rows of particular #texture# with a single row containing their average.
""", version = scriptID)
parser.add_option('-l','--label', dest='key', type="string", metavar='label',
help='column label for averaging rows [%default]')
parser.set_defaults(key = [])
(options,filenames) = parser.parse_args()
if options.key == None:
parser.error('No sorting column specified.')
# ------------------------------------------ setup file handles ---------------------------------------
files = []
if filenames == []:
files.append({'name':'STDIN', 'input':sys.stdin, 'output':sys.stdout, 'croak':sys.stderr})
else:
for name in filenames:
if os.path.exists(name):
files.append({'name':name, 'input':open(name), 'output':open(name+'_tmp','w'), 'croak':sys.stderr})
# ------------------------------------------ loop over input files -----------------------
for file in files:
if file['name'] != 'STDIN': file['croak'].write('\033[1m'+scriptName+'\033[0m: '+file['name']+'\n')
else: file['croak'].write('\033[1m'+scriptName+'\033[0m\n')
table = damask.ASCIItable(file['input'],file['output'],False) # make unbuffered ASCII_table
table.head_read() # read ASCII header info
table.info_append(string.replace(scriptID,'\n','\\n') + \
'\t' + ' '.join(sys.argv[1:]))
# ------------------------------------------ assemble header -----------------------------
table.head_write()
# ------------------------------------------ process data --------------------------------
rows, cols = table.data_readArray()
table.data = table.data[np.lexsort([table.data[:,table.labels_index(options.key)]])]
values, index = np.unique(table.data[:,table.labels_index(options.key)], return_index=True)
index = np.append(index,rows)
avgTable = np.empty((len(values), cols))
for j in xrange(cols) :
for i in xrange(len(values)) :
avgTable[i,j] = np.average(table.data[index[i]:index[i+1],j])
table.data = avgTable
table.data_writeArray()
# ------------------------------------------ output result -------------------------------
table.output_flush() # just in case of buffered ASCII table
table.input_close() # close input ASCII table
if file['name'] != 'STDIN':
table.output_close() # close output ASCII table
os.rename(file['name']+'_tmp',options.key+'_averaged_'+file['name']) # overwrite old one with tmp new