changelog

人工智能的一个重要方法

March 2002 (rz, ml & ab)
	-First working version based on the original fortran program of
	 R.Zecchina and M.M閦ard (ml, rz & ab)
30 April 2002 (rz & ab)
	-Several optimizations to compute_distweight, totalizing some 500%
	 boost in total (!). (rz & ab)
	-reads cnf formulas. (ab)
	-dynamic memory usage. (ab)
	-renaming of variables and function names for consistency and
	 shorteness: connect->v, flip_*->fix_* and others (ab)
	-online shell call to external program (walksat) to solve 
	 paramagnetic subformulae (ab)
	-created small dumb integrating tools: merge.c and verify.c
14 June 2002 (ab)
	-rename kpnew.* to kp.*
	-split kp.c in separate .c files: formula.c simann.c kp.c
	-the convergence error tolerance now is a max instead of an average.
         (still experimenting about this one)
	-small changes on strategies (idem)
August 2002 (mw, rz, ab)
	-simplified dynamics at y=infinity (rz & mw)
	-improved sequential updating of pi and eta (rz, mw & ab)
	-convergence error tolerance is now relative (still a max) (rz)
	-fixed 'last field' bug (ab)
	-use getopt to parse command line (ab)
	-rename k* to s* (rz)
	-fix fraction of spins at a time
September 2002 (rz, mm, ab)
	-iterative updating in a random permutation order (mm)
	-handle variable clause-size formulas (ab)
	-optimized memory usage (ab)
	-added initial 1-clauses simplification
October 2002 (rz, ab)
	-renormalization of eta
	-correct handling of pi products in corner cases
	-compute_sigma()
November 2002
	-first public release
January 2003 (ab)
	-fix a bug on calling qsort that showed up on solaris 
	 but not on linux, by P.Henderson and P.Purdom
	-minor bugfixes
April 2003 (ab)
	-added experimentals -R and -z
	-second public release
May 2003 (ab)
	-switch from char to int in data struct
August 2003 (ab)
	-fix a crash (yes!) on reading formulae with 1-clauses, thanks to
	 Florent Krzakala for reporting.
	-fix a "<0 should be <EPS" bug. Thanks to Finnegan Southey and
	 Florent Krzakala.
	-fix a bug in which some "corner cases" were not handled correctly,
	 thanks to Finnegan Southey.
	-compile with -DFAST_ITERATION for a possibly less accurate but 
	 faster (and easier to read) sp iteration (possibly makes sense only
	 in random formulae in the first phases of decimation, though).
	-make -z optional (-DQUEUE)
April 2004 (ab)
	-refix a bug that was already fixed but came back for some unknown
	 reason (the crash as first item above). Thanks to the many people
	 that reported.

ab: Alfredo Braunstein <[e-mail]>
ml: Michele Leone <[e-mail]>
mm: Marc M閦ard <[e-mail]>
mw: Martin Weigt <[e-mail]>
rz: Riccardo Zecchina <[e-mail]>