Content deleted Content added
m Dating comment by Bodhi.root - "" |
Bodhi.root (talk | contribs) No edit summary |
||
Line 89:
[[User:Mskinn99|Mskinn99]] ([[User talk:Mskinn99|talk]]) 23:24, 13 January 2008 (UTC)
== Page Cleanup ==
I extended the page with a brief overview a semi formal description of the algorithm, referenced some performance improvements and included an example. I personally feel that the matrix based description is best to follow. Additionally I felt that it is most helpful to understand the workings of this algorithm through a numerical example. Therefore I included a (rather extensive) example based on an example presented in the widely used AI text book of Russel and Norvig. I also referenced a repository of source code where java code implementing the algorithm may be found.
I am generally new to editing at Wikipedia but tried to follow the guidelines I found to be relevant. If the content is not found suitable, contains errors or is otherwise unsuitable, I certainly welcome feedback, critique and any edits, including deletions, deemed necessary :)
Line 94 ⟶ 95:
I just finished cleaning up the page a bit to try and make it easier to follow. Hopefully I didn't make things worse. I am new to the subject but described the problem as it is addressed in the first reference (Rabiner, 1989). I also modified the notation so that the forward probabilities use subscripts of the form: <math>\mathbf{f_{0:t}}</math> (zero-based instead of one-based). This provides a consistent form that describes the problem and allows us to calculate the state vector at time t=0. I also used the "hat" to distinguish between the actual probabilities of an event sequence and the normalized values that describe the state vector. I hope the notation is OK, and I worry about my terminology (is "state vector" appropriate? despite my reading I can't find what this is actually called). Also, my opinion is that the matrix example is very helpful. I found it useful for checking my results when I developed an HMM program and think it should be left in. [[User:Bodhi.root|Bodhi.root]] ([[User talk:Bodhi.root|talk]]) <span style="font-size: smaller;" class="autosigned">—Preceding [[Wikipedia:Signatures|undated]] comment added 19:36, 15 September 2009 (UTC).</span><!--Template:Undated--> <!--Autosigned by SineBot-->
*Sorry, had to modify things once more. The explanation I gave before was not technically correct. The new version explains the underlying probabilities and how to interpret the forward, backward, and smoothed probabilities. It is a little harder to follow, but it is now technically correct. I also modified the example section to use the new notation. The matrix examples should probably be reworked so that they use the formulas given earlier (i.e. scaling by the <math>c_t</math>'s), but I'm done for now and don't want to recalculate all that. [[User:Bodhi.root|Bodhi.root]] ([[User talk:Bodhi.root|talk]])
==Explaining this algorithm with Matrices==
| |||