Mercurial > forge

changeset 12493:32ce8fbfb857 octave-forge

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Converted calculation in hmmviterbi to log space.
author asnelt
date Sat, 24 May 2014 23:09:12 +0000
parents d7adff2641da
children e784cfb75305
files main/statistics/NEWS main/statistics/inst/hmmviterbi.m
diffstat 2 files changed, 8 insertions(+), 7 deletions(-) [+]
line wrap: on
line diff
--- a/main/statistics/NEWS	Sat May 24 19:47:26 2014 +0000
+++ b/main/statistics/NEWS	Sat May 24 23:09:12 2014 +0000
@@ -1,7 +1,7 @@
 Summary of important user-visible changes for statistics 1.2.4:
 -------------------------------------------------------------------
 
- ** Make princomp work with nargout < 2.
+ ** Made princomp work with nargout < 2.
 
  ** Renamed dendogram to dendrogram.
 
@@ -9,6 +9,8 @@
 
  ** Transposed output of hist3.
 
+ ** Converted calculation in hmmviterbi to log space.
+
  ** The following functions are new:
  
     cdf signtest ttest ttest2 vartest vartest2 ztest qrandn dcov
--- a/main/statistics/inst/hmmviterbi.m	Sat May 24 19:47:26 2014 +0000
+++ b/main/statistics/inst/hmmviterbi.m	Sat May 24 23:09:12 2014 +0000
@@ -189,17 +189,16 @@
     endif
   endif
 
-  # Each row in transprob and outprob should contain probabilities
-  # => scale so that the sum is 1
-  # A zero row remains zero
+  # Each row in transprob and outprob should contain log probabilities
+  # => scale so that the sum is 1 and convert to log space
   # - for transprob
   s = sum (transprob, 2);
   s(s == 0) = 1;
-  transprob = transprob ./ (s * ones (1, columns (transprob)));
+  transprob = log (transprob ./ (s * ones (1, columns (transprob))));
   # - for outprob
   s = sum (outprob, 2);
   s(s == 0) = 1;
-  outsprob = outprob ./ (s * ones (1, columns (outprob))); 
+  outprob = log (outprob ./ (s * ones (1, columns (outprob)))); 
 
   # Store the path starting from i in spath(i, :)
   spath = ones (nstate, len + 1);
@@ -211,7 +210,7 @@
   # Find the most likely paths for the given output sequence
   for i = 1:len
     # Calculate the new probabilities of the continuation with each state
-    nextpathprob = ((spathprob' .* outprob(:, sequence(i))) * ones (1, nstate)) .* transprob;
+    nextpathprob = ((spathprob' + outprob(:, sequence(i))) * ones (1, nstate)) + transprob;
     # Find the paths with the highest probabilities
     [spathprob, mindex] = max (nextpathprob);
     # Update spath and spathprob with the new paths