Changelog

02/05/2009 Version 1.3.1

* fixed: compilation issues with gfortran (removed a EXTERNAL call for ETIME in arpack/second.f)
* added: dependence with Rmath lib (for src/cdf.c)

08/28/2008 Version 1.3.0

* src/cdf.c: replaced old nr code by gsl equivalent

09/26/2006 Version 1.2.2

* many bugs fixed
* an improved precise mode for ldstat

02/22/2006 Version 1.2.1

* fixed: multiple sequences are now taken into account in the expected number of occurrences.
* fixed: reading markov file now works with large (e.g. size=26) alphabets.
* added: aspatt program which is not functional yet.

10/10/2005 Version 1.2.0

* added a new transition class both used by ldstat and xstat
* class newxstat (through fmci rather than recurrences)
* added precise support for the sparse case in ldstat

09/01/2005 Version 1.1.3

* added the "--zscore" option to gspatt allowing to output zscore instead of
classical logscale statistics.
* added the "--precise" option to ldspatt. Using it will improve reliability
on ldstat computations (computational times grows by a factor 3).

06/13/2005 Version 1.1.2

* corrected small bug with option "-p" for pattern of size > max_length
in the case of xstat and gstat.
* added a definition of INFINITY in cp.cc (for system where this could cause
problem)
* corrected a small bug in ldstat, sequence length was longer that necessary.
This bug will have few effect for ld statistics computed on long sequence.
Could be more critical on shorter sequences.
* added support for order > 0 Markov models in gstat
* imported three classes from MeSs project: WordFam, PSucceed, PAppearFast
* xstat is now working for Markov model order > 0 for any pattern
* added minor changes to make compilation easier on Solaris/Sun

06/02/2005 Version 1.1.1

* removed arprec from the project
* dramatically improved xstat (200 times faster)
* fixed a bug preventing -S to works with order<=0 Markov models
* fixed a small bug in the computation of 1.0-betai in cdf.c
this should have very few effect (only a difference of few digits in some
rare cases) on previously statistics computed sspatt or cpspatt

05/25/2005 Version 1.1

* added arpack to the package
* moved lapack and blas to toplevel dir
* added arpack support to markov class
* added few fortran parameter classes
* added LD stat with automaton for all words (ldstat.c and ldstat.h)
* moved fortran param classes in their own sources (fortran.h fortran.cc)
* added process and stat classes
* created a testspatt program for new design testing
* added "|{rc}" to pattern label when --both-strands is specified
* changed FFLAGS to get rid of infinite loop lapack bug
* added a sstat class derived from stat
* changed --min-pvalue to --max-pvalue (more consistent with the function)
* added a cpstat class
* added a ldstat class
* added a gstat class
* added a xstat class

02/22/2005 Version pre-1.0.10

* geometric Poisson tail now computed through a new implementation
(newcpstat, but old one still availaible as cpstat)
* added the legal GPL mentions to all files
* added the nobs option to sspatt, for the moment, it is
only working with pattern statistics (but without --all-words)
* -n and -b can now work together

02/09/2005 Version 1.0.9

* changed max word size limit from 13 to 30. Please note
that O(K^L) memory is required in order to count all the
words (K size of the alphabet, L size of the words).
* added the --min-pvalue option to sspatt program
* corrected a bug in xspatt. ac had a wrong index. highly
self-overlapping exact statitistics computed before this
correction should be considered with caution.
* corrected a bug in xspatt. Few high magnitude negative
statistics could be wrong in version prior to this correction.

02/07/2005 Version 1.0.8

* added man pages for all programs in package
* added the normalize option to sspatt,xspatt,ldspatt and gspatt
* added the normalize option to cpspatt
* added support for all markov models in cpspatt

12/17/2004 Version 1.0.7

* bugfix: corrected a bug in cp stat computation for
under-reprensented words with one occurrence. Old results
for such words (only when they are overlapping ones) should
be sligthly changed with this correction. Not too much to worry
about.

12/10/2004 Version 1.0.6

* added: program gspatt (only order <=0 supported atm)
* added: a self-test for gspatt
* Corrected a critical bug in xspatt: changed "current[x]+=sum"
in "current[x]-=sum". exact stats computed with version of SPatt
< 1.0.6 should be considered with extreme caution.

12/09/2004 Version 1.0.5

* bugfix: added a "if (rank=n) done=1" test in real_explicit_arnoldi
should get rid of the remaining rare bugs here
* bugfix: added a "&& rank < n" in real_explicit_arnoldi
in order to deal with rare nan produced in stationnary
when low markov order is used (m=1 for example). Should
also slightly speed up the computation in such cases.
* bugfix: added a test for normalization of null vectors
in markov.cc
* added: tests for xspatt and cpspatt

10/27/2004 Version 1.0.4

* cpstatt: complexity reduced to linear
using a nice recurrence. Old method still
available as "oldcpstat"
* stat formats are now consistant in all
programs
* arprec now included in distribution
* ldspatt: added a special treatment for
a=0.0. In this case I=-Lambda(-inf)

10/22/2004 Version 1.0.3

* added compound poisson support with cpspatt
* for the moment, only non degenerate pattern
are treated with compound Poisson approximation
complexity is quadratic with number of occurrences

10/14/2004 Version 1.0.2

* corrected some minor trouble with uninitialized
values in markov contructors. Now valgrind displays
no more error.
* xspatt can now compute the exact stat for a given
non degenerate pattern
* xspatt can now compute the exact stat for a words
of a given length

10/17/2004 Version 1.0.1

* implicit arnoldi restarted support added
* corrected a minor numerical bug in arnoldi
now computation stability should be improved
* added xspatt program. Its actual function
is only to compute the second magnitude
eigenvalue.
* quicksort is now used to sort eigenvalues

09/17/2004 Version 1.0

* the GSL (GNU Scientific Library) is now
required for compilation.
* ldspatt added to the package
* added a test for ldspatt
* removed few old binaries (useless)

09/15/2004 Version 0.9.1

* bugfix: now sspatt should not segfault when -m
is omitted.
* added a statistic class. main sources are now
much shorter and readable.
* removed: unused progress class

09/09/2004 Version 0.9

* corrected qcdfpoi computations for extreme
rare events.

07/13/2004 Version 0.8

* bugfix: _n now set to _k when _order<=0
* bugfix: order<=0 markov model are now properly
taken into account in markov::dump_model

06/04/2004 Version 0.7

* bugfix: _states now initialized to NULL
in word constructor
* bugfix: --all-words incompatible with no
markov model specified. Program now prompt
a message before exiting (should be fixed later)
* added option -U to use Markov parameter from
a file
* added options -M and -S to output model and
stationnary

03/08/2004 Version 0.6

* cdfbin now used instead of cdfpoi
(increased reliability on very short sequences)
* added support for order 0 and -1 model
(-1 for iid model with no memory)
* added --both-strands support for --all-words
option

03/06/2004 Version 0.5

* added --all-words option
* added support for small p-values
* added p-value using simple Poisson approximation
* first guess for stationnary now from estimation
when it is possible
* corrected issue with imaginary part of eigenvector
non null when a real vector is expected
* added basic Arnoldi support: explicit_arnoldi
(real only)

02/26/2004 Version 0.4

* default for length is now 8
* added -b --both-strands option
* dramatically improved core using
arrays of word* instead of maps
* dramatically improved build_dfa
* added pattern counting
* added pattern regex descriptor
ex: "atg[cg]ta.|tta.[at]|atata"
* complete rewriting of word class
(including copy and affectation fct)
* added a global header
* added support for several bin
* removed output class
* added inline keyword when necessary

02/23/2004 Version 0.3

* default for length is now 0
* fixed a case sensitive bug with dfa count
* gspatt now fully operationnal
for word counting
* added input/output support
* fixed dfa counting
* added word support
* added dfa support

02/18/2004 Version 0.2

* added count support
* added explicit classical constructors
(empty, copy) and operator (affectation)
to all classes.
* added sequence support
* member data name now start with '_'
* added progress bar support
* added command line basic support
* added alphabet support

01/27/2004 Version 0.1

* Initial release. Prog do nothing yet