[fitsbits] neural net default choice of compression algorithm? Was:

hi all,

On Thu, 20 Jul 2006, Mark Calabretta wrote:

For the freedb database 7zip compresses 30% better than gzip and
decompresses *faster* than gzip in elapsed time. The 11-fold increase
in compression time would be amortized via repeated accesses to the
database.

The bottom line is to use the technology that best suits the task.

Just throwing ideas around: it could be possible to make a simple neural
net algorithm, or even just write a small handful of if-then statements,
which automatically chooses a default algorithm likely to be most useful
for the particular type of data in a fits file, based on the FITS keywords.
Or it could be something like the bayesian filter usable in spamassassin and
so on, e.g. words like "dome flat" or "flat field" in description headers
or whether or not the data are integer or floats, seem to give differences
based on the previous few days' discussion on what are likely to be
preferable algorithms.

Of course, it would need to be possible to override the default which
is recommended by this algorithm-for-choosing-algorithms. And it
depends at what level "user" we want to be friendly with. On the other
hand, it may not be worth the effort put in to write it (and i'm not
volunteering)...

al-khawarizmical greetings
boud