CHANGES

--------------------------
Version 1.01
Date:  4 May, 1999
--------------------------

1. included math.h in hmm.h. Not including it was creating problems
   in forward.c.

2. BaumWelch(): Convergence criterion has been changed: now stops if
   log Prob(Obser|model_n) - log Prob(Obser|model_{n-1}) < DELTA
   where DELTA = 0.001  in baumwelch.c. 

3. esthmm: Allow user to specify the initial model by using a -I flag and
   and then giving the hmm model file name. 

   So, say you generate a sequence O using a model lambda_0. Now if 
   you estimate the model parameters lambda_1 from the observation using
   lambda_0 as  the intial parameter values: 
     log Prob(O|lambda_0) and log Prob(O| lambda_1) should be close.

   This is a sanity check for BaumWelch code.

4. genseq: now generates random sequences everytime you run it. It also
   has a way of specifying the random seed in the command line so that
   you can replicate pervious experiments.

5. Added a new executable: testfor. This allows you to compute
   log Prob(O| model) using Forward() and ForwardWithScale().

6. Added  postscript and PDF versions of a tutorial I gave on HMMs.

7. Tested on: 
 	Sun Ultra 2 running UNIX SUN Solaris 5.5 and GNU gcc 2.7.2
	DEC Alpha running  UNIX OSF version 4.0 and GNU gcc 2.8.1
        Dell i686  running Redhat Linux 2.0.36  and GNU gcc 2.7.2.3

--------------------------
Version 1.02
Date:  5 May, 1999
--------------------------

1. Test on:
       Gateway Pentium 400 MHz machine running NT. The compiler
          was gcc from Cygnus (version B20) You can get it from:
      ftp://go.cygnus.com/pub/sourceware.cygnus.com/cygwin/latest/full.exe 

2. Changed the README file, viterbi.c

3. Corrected few typos in the tutorial (hmmtut.ps and hmmtut.pdf).