[fitsbits] the need for BITPIX=64?

William Pence writes:
Preben Grosbol wrote:
I still have reservation concerning BITPIX=64 for the following reasons:
1) there seems no good physical reason for 64-bit integer images. The
number of photons from astronomical source hardly justifies it
especially considering their statistical distribution. Let someone
present a real, practical case and we should considere it.

Do you not consider any of the 15 cases given in my email of 07-June-2005
real or practical? This included:

- histogram arrays derived from very large databases
- arrays of measured time values
- arrays of 'accumulated sums'
- the need to import data from other sciences (space physics,
planetary research, earth sciences) into FITS

All of these examples are to some extent a bit academic:
- histogram arrays and accumulated sums presumably have enough Poisson noise
that no significance is lost by storing them as floats (64 bits floats
if needed)
- tables look like a better match than images for arrays of time values.
- other sciences presumably have similar limitations to their dynamical
ranges, or at least we haven't yet really heard of one that does not.
What's lacking for now is somebody that says "Hey, I (will) have a data set
that I want to write to disk that requires 64 bit integers".


I think it is important here to not set the requirements for justifying
BITPIX = 64 arrays too high. This data type will probably never be very
widely used, but that is not the point. All that should matter is that
there is at least 1 case, important to some subset of the astronomical
community, where having 64-bit integer arrays in FITS would be very useful.
Also, the FITS format is used for many utilitarian purposes, so one should
not automatically rule out more 'practical' uses, (e.g., temporary storage
of 'scratch' arrays of intermediate computations) just because they are not
based on fundamental scientific or physical needs.

Part of the disagreement here is probably between the part of the community
that uses FITS for "everything" and would thus like it to cover every corner
case, and the part that sees it as an exchange format that they need to
keep a converter from and that they would thus like to keep reasonably simple.

On Friday 17 June 2005 23:17, Thierry Forveille wrote:
* Do you not consider any of the 15 cases given in my email of
07-June-2005 real or practical? *This included:
*
* * *- histogram arrays derived from very large databases
* * *- arrays of measured time values
* * *- arrays of 'accumulated sums'
* * *- the need to import data from other sciences (space physics,
* * * *planetary research, earth sciences) into FITS
*
All of these examples are to some extent a bit academic:
- histogram arrays and accumulated sums presumably have enough Poisson
noise that no significance is lost by storing them as floats (64 bits
floats if needed)
- tables look like a better match than images for arrays of time values.
- other sciences presumably have similar limitations to their dynamical
* ranges, or at least we haven't yet really heard of one that does not.
What's lacking for now is somebody that says "Hey, I (will) have a data set
that I want to write to disk that requires 64 bit integers".
Yes, this was exactly also my considerations.
- histogram need not have equal bin size - thus table are better
Preben

On Mon 2005-06-20T09:49:04 +0200, Preben Grosbol hath writ:
- histogram need not have equal bin size - thus table are better

With the -TAB convention of WCS Paper III the world coordinate extents
of pixels need not be equal, nor do adjacent pixels in the FITS array
necessarily correspond to adjacent regions in world coordinates.
Which is to say, image arrays can serve this same function as tables.

Yes, this is still somewhat of a contrived example, but my impression
remains that we do no harm and much good by including 64-bit integers.

--
Steve Allen WGS-84 (GPS)
UCO/Lick Observatory Natural Sciences II, Room 165 Lat +36.99858
University of California Voice: +1 831 459 3046 Lng -122.06014
Santa Cruz, CA 95064 http://www.ucolick.org/~sla/ Hgt +250 m

Preben Grosbol wrote:
On Friday 17 June 2005 23:17, Thierry Forveille wrote:


What's lacking for now is somebody that says "Hey, I (will) have a data set
that I want to write to disk that requires 64 bit integers".

Yes, this was exactly also my considerations.
- histogram need not have equal bin size - thus table are better
Preben

I remember only too clearly how, round about 1997, we were desperate for
the WCS convention to get set in stone, in the context of the instrument
project I was involved in. The instrument went on sky with no WCS,
because the software guys couldn't be given a specification.

64-bit is such a clear progression, and will probably get used, so for
goodness sake let's not wait until some project needs it today, and THEN
start the FITS adoption process; let's start that now, and at least in
one area, have future-proofed FITS for a while.

Peter.