From roden@oblio.Jpl.Nasa.Gov Fri Jan 28 17:18:39 1994
Return-Path: <roden@oblio.Jpl.Nasa.Gov>
Received: from oblio.jpl.nasa.gov by galway.Jpl.Nasa.Gov (4.1/SMI-4.1+DXRs2.3)
	id AA29562; Fri, 28 Jan 94 17:18:38 PST
Received: by oblio.jpl.nasa.gov (4.1/JPL-AIG-1.1)
	id AA03259; Fri, 28 Jan 94 17:18:33 PST
Date: Fri, 28 Jan 94 17:18:33 PST
From: roden@oblio.Jpl.Nasa.Gov (Joe Roden)
Message-Id: <9401290118.AA03259@oblio.jpl.nasa.gov>
To: pjs@galway.Jpl.Nasa.Gov
Subject: SKICAT Training Data
Status: RO


Padhraic,

At long last, I've got some SKICAT training data files ready for 
you.  Sorry, I've been thrashing between numerous tasks, and this 
has been suffering from indefinite postponement- I need a better 
operating system for controlling my life.

I've given you 2 data files- j.dat and f.dat.  These files are in 
Usama's standard example file format; first line is # of examples, and 
each successive line is an example.  Each example has: an example id, 
8 attributes, and the class: s (star), g (galaxy), ss (binary star?), 
and sf (star w/ fuzz).

F and J names of the files derive from the plate color, so they aren't 
j examples aren't really from the same distribution as f examples.

The attributes are:

Name      Description
----      -----------
mag       total magnitude
scale     resolution scale
frac      resolution fraction
ellip     ellipticity
n_mcore   normalized core magnitude
n_larea   normalized log of the isophotal area
n_ir1     normalized intensity weighted first moment radius
n_s       normalized S

where
                                 Area
     S = -----------------------------------------------------
          log [ core luminosity / (9 x isophotal intensity) ]

The data files (and some software) are on our machines in directory:
	/proj/skicat/for_pjs

The software is a program called ex_select3, which is what we've 
been using to extract a subset of examples at random to form training/
testing sets.  Compile it w/ a standard cc compile line- no hassles:

	cc ex_select3.c -o ex_select3

The file takes one parameter- a spec file which describes the train/
test files to create.  A description of the spec file is in the file 
control_file_specs.  An example (which you can probably use directly)
is also provided, named select-in.  

Let me know if you have any questions.

Good Luck!

Joe


