aboutsummaryrefslogtreecommitdiffstats
path: root/astro/fits
diff options
context:
space:
mode:
authorAaron LI <aly@aaronly.me>2018-11-15 23:29:57 +0800
committerAaron LI <aly@aaronly.me>2018-11-15 23:29:57 +0800
commit4cf13ac35cd9ef74cd9743bc0c6b5bf337b0784e (patch)
tree53d6774abfb677b7f5308d67b90e54991ccae8c3 /astro/fits
parent9181265fdaca30db5630ff137ac868f98b427f7f (diff)
downloadatoolbox-4cf13ac35cd9ef74cd9743bc0c6b5bf337b0784e.tar.bz2
Move fits/merge_fits.py to fitstable_merge.py
Diffstat (limited to 'astro/fits')
-rwxr-xr-xastro/fits/merge_fits.py234
1 files changed, 0 insertions, 234 deletions
diff --git a/astro/fits/merge_fits.py b/astro/fits/merge_fits.py
deleted file mode 100755
index e014c69..0000000
--- a/astro/fits/merge_fits.py
+++ /dev/null
@@ -1,234 +0,0 @@
-#!/usr/bin/env python3
-# -*- coding: utf-8 -*-
-#
-# Aaron LI
-# 2015/06/16
-#
-# Changelogs:
-# 0.3.0, 2015/06/17, Aaron LI
-# * added argument '-c/--columns' to specify columns to be merged
-# * added argument 'columns' to function 'merge2fits()'
-# 0.2.0, 2015/06/17, Aaron LI
-# * added function 'del_key_startswith()' to delete header keywords,
-# and the deletion must be repeated until the header length does not
-# decrease any more
-# * ignore the header of the second FITS file to avoid keyword conflictions
-#
-# BUGS:
-# * astropy.io.fits may have bug with header keywords deletion
-#
-# TODO:
-# * to support image FITS merge
-# * to allow specify column list to be merged
-#
-
-"""
-Merge several (>=2) of FITS file.
-
-By default the *first* extend tables are merged and write out to a new
-FITS file containing the *common* columns. If the data types of the
-columns of each FITS table do not match, then the data type of the column
-of the *first* FITS table is used, and other columns are coerced.
-
-If the FITS files have only *1* HDU (i.e., the Primary HDU), then data of
-these HDU's are summed up to make up the output FITS file (an image),
-on conditional that the shapes of all these HDU's are the same.
-"""
-
-__version__ = "0.3.0"
-__date__ = "2015/06/17"
-
-# default blockname to be merged
-BLOCKNAME_DFT = "EVENTS"
-
-DEBUG = True
-
-import sys
-import argparse
-import re
-
-import numpy as np
-try:
- from astropy.io import fits
-except ImportError:
- try:
- import pyfits as fits
- except ImportError:
- raise ImportError("cannot import 'astropy.io.fits' or 'pyfits'")
-
-
-def merge2fits(file1, file2, block1=1, block2=1, columns=None):
- """
- Merge *two* FITS files of the given blocks (extension table),
- and return the merged FITS object.
-
- TODO:
- * log history to header
-
- Arguments:
- file1, file2: input two FITS files
- block1, block2: table number or table name to be merged
- columns: the columns to be merged; by default to merge the
- common columns
-
- Return:
- the merged FITS object
- """
- # open file if provide filename
- if isinstance(file1, str):
- file1 = fits.open(file1)
- if isinstance(file2, str):
- file2 = fits.open(file2)
- # if has only *1* HDU => image
- if len(file1) == 1:
- block1 = 0
- if len(file2) == 1:
- block2 = 0
- if block1 == 0 or block2 == 0:
- # TODO
- raise NotImplementedError("image FITS merge currently not supported!")
- # get table to be merged
- table1 = file1[block1]
- table2 = file2[block2]
- # create column names to be merged
- # get names of all columns (convert to upper case)
- colnames1 = [col.name.upper() for col in table1.columns]
- colnames2 = [col.name.upper() for col in table2.columns]
- colnames_common = list(set(colnames1).intersection(set(colnames2)))
- # sort the common column names acoording original column orders
- colnames_common.sort(key = lambda x: colnames1.index(x))
- if columns is not None:
- if isinstance(columns, list):
- columnlist = list(map(str.upper, columns))
- else:
- columnlist = list(columns.upper())
- # check the specified columns whether in the above colnames_common
- for name in columnlist:
- if name not in colnames_common:
- raise ValueError("column '%s' not found in both files" % name)
- # use the specified columns
- colnames_common = columnlist
- # "STATUS" columns don't have equal-length format, so remove it
- if "STATUS" in colnames_common:
- colnames_common.remove("STATUS")
- if DEBUG:
- print("DEBUG: columns to merge: ", colnames_common, file=sys.stderr)
- # filter out the common columns
- nrow1 = table1.data.shape[0]
- nrow2 = table2.data.shape[0]
- hdu_merged = fits.BinTableHDU.from_columns(
- fits.ColDefs([table1.columns[name] for name in colnames_common]),
- nrows=nrow1+nrow2)
- for name in colnames_common:
- if DEBUG:
- print("DEBUG: merging column: ", name, file=sys.stderr)
- dtype = hdu_merged.columns[name].array.dtype
- hdu_merged.columns[name].array[nrow1:] = \
- table2.columns[name].array.astype(dtype)
- # process headers, based on the header of the first FITS file
- # DO NOT strip the base header, in order to keep the position of
- # XTENSION/BITPIX/NAXIS/NAXIS1/NAXIS2/PCOUNT/GCOUNT/TFIELDS keywords.
- header = table1.header.copy() # do not strip
- # IGNORE the header of the second FITS file to avoid keyword conflictions.
- #header2 = table2.header.copy(strip=True)
- ## merge two headers; COMMENT and HISTORY needs special handle
- #for comment in header2["COMMENT"]:
- # header.add_comment(comment)
- #for history in header2["HISTORY"]:
- # header.add_history(history)
- #if "COMMENT" in header2:
- # del header2["COMMENT"]
- #if "HISTORY" in header2:
- # del header2["HISTORY"]
- #if "" in header2:
- # del header2[""]
- #header.update(header2)
- # remove the original TLMIN??/TLMAX??/TTYPE??/TFORM??/TUNIT?? keywords
- del_key_startswith(header,
- startswith=["TLMIN", "TLMAX", "TTYPE", "TFORM", "TUNIT"],
- lastlength=len(header))
- # update with new TLMIN??/TLMAX??/TTYPE??/TFORM??/TUNIT?? keywords
- header.update(hdu_merged.header)
- hdu_merged.header = header
- # copy PrimaryHDU from first FITS
- primary_hdu = file1[0].copy()
- # make HDUList and return
- return fits.HDUList([primary_hdu, hdu_merged])
-
-
-def del_key_startswith(header, startswith, lastlength=0):
- """
- Delete the keys which start with the specified strings.
-
- Arguments:
- header: FITS table header
- startswith: a list of strings; If a key starts with any
- of these strings, then the key-value pair is removed.
-
- XXX: the deletion must be repeated several times until the
- length of the header does not decrease any more.
- (This may be a bug related to the astropy.io.fits???)
- """
- if not isinstance(startswith, list):
- startswith = list(startswith)
- re_key = re.compile(r"^(%s)" % "|".join(startswith), re.I)
- for k in header.keys():
- if re_key.match(k):
- del header[k]
- curlength = len(header)
- if lastlength == curlength:
- return
- else:
- # recursively continue deletion
- if DEBUG:
- print("DEBUG: recursively continue header keywords deleteion",
- file=sys.stderr)
- del_key_startswith(header, startswith, curlength)
-
-
-def get_filename_blockname(pstr):
- """
- Separate privided 'pstr' (parameter string) into filename and
- blockname. If does not have a blockname, then the default
- blockname returned.
- """
- try:
- filename, blockname = re.sub(r"[\[\]]", " ", pstr).split()
- except ValueError:
- filename = pstr
- blockname = BLOCKNAME_DFT
- return (filename, blockname)
-
-
-def main():
- parser = argparse.ArgumentParser(
- description="Merge several FITS files with the common columns.")
- parser.add_argument("-V", "--version", action="version",
- version="%(prog)s " + "%s (%s)" % (__version__, __date__))
- parser.add_argument("infile1", help="input FITS file 1; " + \
- "The blockname can be appended, e.g., infile1.fits[EVENTS]")
- parser.add_argument("infile2", nargs="+",
- help="input FITS file 2 and more")
- parser.add_argument("outfile", help="merged output file")
- parser.add_argument("-c", "--columns", dest="columns",
- help="list of columns to be merged (comma separated)")
- parser.add_argument("-C", "--clobber", dest="clobber",
- action="store_true", help="overwrite output file if exists")
- args = parser.parse_args()
- if DEBUG:
- print("DEBUG: infile2: ", args.infile2, file=sys.stderr)
-
- if args.columns:
- columns = args.columns.upper().replace(",", " ").split()
- file1, block1 = get_filename_blockname(args.infile1)
- merged_fits = fits.open(file1)
- for fitsfile in args.infile2:
- # split filename and block name
- file2, block2 = get_filename_blockname(fitsfile)
- merged_fits = merge2fits(merged_fits, file2, block1, block2, columns)
- merged_fits.writeto(args.outfile, checksum=True, clobber=args.clobber)
-
-
-if __name__ == "__main__":
- main()
-