Mercurial > repos > mb2013 > nepenthes_3dpca
view converter.py @ 12:f69fc89bb392 draft
Uploaded
author | mb2013 |
---|---|
date | Tue, 20 May 2014 03:27:42 -0400 |
parents | aae5945b961e |
children |
line wrap: on
line source
# python 2.7 # MB # Convertion multiple .dta files to one .csv file with coordinates # and one .csv file with sample names. from optparse import OptionParser from os.path import basename # Function for merging all files in .csv format # and extracting the headers of the files. def main(): # extracting the users input parser = OptionParser() parser.add_option("--output") parser.add_option("--output2") parser.add_option("--input_file", action="append", default=[]) # multiple inputfiles parser.add_option("--input_name", action="append", default=[]) # names of multiple inputfiles (options, _) = parser.parse_args() header = "" header_temp = "" # open every file and concatenate the files with(open(options.output, "w")) as output: with(open(options.output2, "w")) as output2: for i, (input_file, input_name) in enumerate(zip(options.input_file, options.input_name)): # open a input file for j, line in enumerate(open(input_file, "r").readlines()): line = line.strip() split_tabs = line.split(' ') number_columns = len(split_tabs) # extract name of sample if j == 0: header_temp = "%s" %(line.replace(' ', '_')) header += "%s\n"%(header_temp[1:-4]) # extract the coordinates if number_columns == 3: output.write("%f,%f,%f\n"%(float(split_tabs[0]),float(split_tabs[1]),float(split_tabs[2]))) output2.write("%s\n"%(header)) # writing header to output file #name of sample main()