********************************* overview ******************************
    Programme d'Analyse Statistique de Tendance et de dIStribution
    (PASTIS) ver 1.0 (1/1/96)
  
    This program is designed to deduce parameters from date sample,
using PRAXIS (Gegenfurter, Behavior Research Methods, Instruments
& Computers, 1992, p. 560-564).  It can analyses learning curve
over time using the best possible method -LogLikelihood
function or chi-square function-.  For reference, see:
  -Power curve: Newell & Rosenbloom, The acqu. of cogn. skills,1981
  -Generalized power-law trend: Logan, Psyck. Review(1988)      
  -Exponential trend: Rescorla-Wagner, 1972, Gluck et Bower, 1988
  -General Exponential Trend: Kail Intellectual Development, 1992
  -Weibull distribution: Logan, JEP:LM&C, 1992
  -Ex-Gaussian, Log-normal & Gamma distributions: Ratcliff, 1978
  
************************** new versions *********************************

    Pastis version 3.02 (11/02/02)
      Added the Wald distribution.

    Pastis version 3.01 (30/11/01)
      The limits for the parameters of the learning curves
      are increased from 10,000 to 100,000.
  
    Pastis version 3.0 (09/06/01)
      A new built of the program, using the more stable DJGCC 
      c/fortran compiler. In order to compile the source code,
      you need a ANSI c comliant compiler with a compatible
      fortran compiler. The gcc and g77 (from the GNU community)
      works fine. 

    Pastis version 2.45 (04/06/01)
      Added the two-parameter Gaussian distribution (normal)
      for comparison purposes, and because it was asked.

    Pastis version 2.35 (04/12/00)
      When the data presented to PASTIS are averaged date, you can
      use the resulting averaged curve formula. Two are provided:
      AEXPONEN  to fit average data using an exponential curve, and
      APLAWCURV to fit a power curve to the average data.
      cx cy are columns containing first and last trials of the 
      averaged pool of raw data. Contact the author for a preprint.
 
    Pastis version 2.25 (16/2/99)   
      The full (3param) displaced logNormal distribution is added.
      The Double-Exponential (or Type I) distribution is
      Also integrated.  Finally, the Ex-Gaussian distribution
      is fixed.  Previous version tended to produce worst fit
      due to rounding errors.
  
    Pastis version 2.15 (31/5/96)
      The log-normal distribution is now implemented.  It is the
      non-displaced distribution, with two parameters, mu and
      sigma of the normal underlying distribution.  To compare
      -lnL results with distributions having three parameters,
      you have to use AIK statistic (see Lambert, JEP:General, 1995
      You start a log-normal analysis with -a LOG-NORMAL
  
    Pastis version 2.1 (27/5/96)
      A new feature has been introduce in Pastis.  Using -h option,
      the value of a parameter can be constraint.  For example,if
      you have some reason to think that the first parameter of a
      Weibull analysis must be less than 500, you can use the command
            pastis -r data.file -h p1<500 -a WEIBULL
      With the same command, you can limit the range of values of a
      parameter.  For exemple, this command says that parameters 1 is
      between 400 and 600:
            pastis -r data.file -h p1<600 p1>400 -a PLAWT
      This feature is superseded by any = constraint.
      WARNING: on any UNIX system, < and > are pipes.  Use the escape
         sequence \\> and \\< instead.
  
    Pastis version 2.0 (1/5/96)
      Pastis has been updated, bringing enhancement to two aspects:
      1- There is now an optimization step.  Pastis looks at the size
         of the input file and determine the best course of action to
         minimize the read-from-disk operations, that are lenghty.
  
      2- Praxis has been replaced by another minimization algorithm:
         Stepit was found to converge more efficently toward a good
         minimum.  In fact, it is so performant that we do not need
         multiple call to pastis. However, it is slower to converge
         (about twice as much function call than with PRAXIS).
         Nevertheless, overall performances of
         PASTIS should be at least 10 times better.