Edited by author 12-12-2007 01:25 AM
/m144 answer: I am surprised because the Viterbi algorithm only needs O(m*n) space where m is the number of tags and n is the length of the word. With m <= 4 and n <= 10 this should not be a problem!

If you can run ok with training based on 100 words but not 500, it seems you are not freeing up memory after each processing each word. Probably you need to find and fix this memory leak.

One way to clean up your code to reduce memory usage is to encapsulate the body of an outer loop into a function. This can help because all memory used by local variables of the function is released every time you return from the function.

Have you used the Matlab profiler to see where your code is slow? That may be where the memory leak is. Unfortunately the profiler does not directly tell you about memory usage.

You may find useful hints here: http://www.mathworks.com/support/tech-notes/1100/1106.html