blindglobe / common-lisp-stat

 -*- mode: text -*-


Common Lisp Stat: 

After a few false starts, I (Tony Rossini) finally know how I want to
proceed.  First, we update CLS.  Then, continue with TSL.  Separation
is needed, so that we can make everything separate.

We are working with SBCL and CLISP implementations so as to keep a
wide range of target platforms as well as to maintain ANSI CL as close
as possible.  This challenges the system by targetting two more common
CL implementations (compiled, and byte-compiled).


Design: 
(see "orig file dependencies" below for original design docs through
code).
------ 

lsobject :  package containing prototype object system


data : package from which all data objects are built.  data should
       provide compound objects, pulling from all potential data
       structures, so that 

optimization : maximization

linear-algebra : merges matrix data structures, numerical linear
                 algebra work

<specialty routines>
regression
nonlin
glim
gee
bayes-regression

ls-user : 








============

File Dependencies from 9 Feb 2007

Package lisp-stat-object-system
   self contained

Package lisp-stat-fastmap
   self contained, new.

Package lisp-stat-basics:
   lsbasics
   compound
   lsmacros
   dists
   ladata
   linalg
   matrices
   lsfloat
   lsmath
   lisp-stat-sequence

 lisp-stat-matrix  (matrices)
   lisp-stat-sequence (sequence)

 lisp-stat-sequence (sequence)
   lisp-stat-compound-data (compound)

 lisp-stat-compound-data (compound)
   common-lisp
   lisp-stat-object-system (lsobject)
   lisp-stat-fastmap
 
lisp-stat-fastmap
   common-lisp   



============
Original, dating to Oct 30, 1991.

This is Lisp-Stat 1.0 Alpha 1, a first attempt at producing a Common
Lisp version of Lisp-Stat. This version contains NO graphics, but
should implement all the non-graphical facilities of Lisp-Stat.  

The implementation uses C code from XLISP-STAT for linear algebra and
probability distributions, so this code is dependent an a CL's foreign
function interface. At this time, three CL's are supported: AKCL (at
least verision 1-600) for UNIX systems, Franz' Allegro CL for UNIX
systems, and Macintosh CL (version 2.0b1). Separate README files
describe each version

On A DEC 3100 or 5000 the AKCL version of Lisp-Stat runs about half as
fast as xlispstat on a standard battery of tests if xlispstat is given
an (expand 30) command. I seem to recall that on a sun3 this the AKCL
code and xlispstat code run at about the same speed. I have not yet
confirmed this. If so, then the relative performance of the AKCL
version to xlispstat may be quite hardware-dependent. The Allegro
version on a DEC 5000 is considerably slower and larger than the AKCL
version, but I have not yet figured out how to tune Allegro's memory
management. The Macintosh CL version seems to run at least as fast as
xlispstat on the Macintosh.

To port this code to another CL, you need to

  Edit defsys.lsp to add any necessary definitions

  Add a top level to lstoplevel.lsp (this is only needed if you
  want to recover the history mechanism, which is broken by
  shadowing *, etc.)

  Write versions of the lisp and/or C glue files to interface to
  the C code in lib.

  Experiment with tuning the memory management to run reasonably
  in statistical applications.

If you do port this code to another CL, please let me know so I can
add it to this distribution.

Luke Tierney
School of Statistics
University of Minnesota
Minneapolis, MN 55455
luke@umnstat.stat.umn.edu