Talk:Conditional random field

Main Definition Appears to have a typo
The last displayed equation trails off. Also I don't see an occurrence of the tilde ~ in the definition itself. aetilley 20:37, 12 July 2015 (UTC)

Relationship with Maximum Entropy Model
It will be nice to see how this model brings the best of HMM and MEMM models. There exists a section describing its relationship to HMM, but there is none for MEMMs.

Prasen 15:15, 24 January 2005 (IST)

Article a bit biased to sequential applications?
I think this article is a bit biased to sequential applications, for example CRF's are being used for images, games (i.e. go), in place of naive bayes nets (Logistic Regression CRF), etc. —Preceding unsigned comment added by 128.250.6.244 (talk) 10:32, 21 February 2008 (UTC)

I tried to address this issue by rearranging the article. There is much missing on CRFs and I will include more examples and more general stuff soon. Good source: http://arxiv.org/abs/1011.4088 T3kcit (talk) 13:25, 30 December 2010 (UTC)

Rewrite
I've started a rewrite of this article so that it'll be easier to understand (hopefully). So far I've done the first part of the Description section, and I'll continue where I left off soonish. Any comments/objections? Arnsholt (talk) 15:57, 6 June 2011 (UTC)
 * I think it would be important to state what X and Y stand for in the definition. — Preceding unsigned comment added by 129.132.245.49 (talk) 14:53, 2 August 2011 (UTC)
 * Definitely! T3kcit (talk) 10:08, 24 August 2011 (UTC)

I think there sould be a discussion of inference methods and special cases, also of parameter learning. I feel you removed to much in your rewrite. T3kcit (talk) 10:08, 24 August 2011 (UTC)

I didn't notice but now the article claims that all CRFs are tree-structured. This is plain wrong and even inconsistent with the inference part.... It's not so easy to undo this so I just delete it for now :(

Where did you get that CRFs have a tree structure? There where several problems with the argument as it was on the page. The Markov property means that each variable is independent of the rest given it's neighbours (at least this is the case for undirected graphs, for directed, the concept of the so-called Markov blanket is a little bit more difficult). This is btw exactly what the definition you wrote said. In computer vision, CRFs are rarely trees. If all CRFs where trees, exact inference would always be possible.

I am not so happy about you removing my explanations and replacing them with wrong arguments and conclusions. Maybe my explanations weren't the best but now there is just a definition. So what now? T3kcit (talk) 05:33, 25 August 2011 (UTC)

Why did you remove the section "Relation to Markov random fields"? I think this is very important to the understanding of this subject. T3kcit (talk) 05:59, 25 August 2011 (UTC)
 * Agreed! I was just looking up this article in search of just such a section. I hope somebody will bring it back… Thomas Tvileren (talk) 12:49, 4 June 2014 (UTC)

Difficult to understand
This article is still pretty incomprehensible to people outside of the area of expertise. I want to know more about these in order to understand natural language parsing, but there are no simple introductions on the net. Can we look at simplifying this a bit?Deadlyvices (talk) 12:35, 25 June 2014 (UTC)

Yet another software
Hi. You can add one more CRF implemenration package. This is a linear-chain CRF Java package that I have developed recently. It is here: https://github.com/asher-stern/CRF 109.66.15.40 (talk) 18:48, 27 November 2014 (UTC)

Difference between linear-chain CRF and CRF unclear / incompletely explained
see subject line — Preceding unsigned comment added by 2A02:8071:41B0:2E00:A803:D936:14D9:6054 (talk) 16:48, 13 April 2017 (UTC)

External links modified
Hello fellow Wikipedians,

I have just modified 3 external links on Conditional random field. Please take a moment to review my edit. If you have any questions, or need the bot to ignore the links, or the page altogether, please visit this simple FaQ for additional information. I made the following changes:
 * Added archive https://web.archive.org/web/20130718001211/http://acl.eldoc.ub.rug.nl/mirror/P/P10/P10-1052.pdf to http://acl.eldoc.ub.rug.nl/mirror/P/P10/P10-1052.pdf
 * Added archive https://web.archive.org/web/20100421020327/http://crfpp.sourceforge.net/ to http://crfpp.sourceforge.net/
 * Added archive https://web.archive.org/web/20100707042144/http://cbioc.eas.asu.edu/banner/ to http://cbioc.eas.asu.edu/banner/

When you have finished reviewing my changes, you may follow the instructions on the template below to fix any issues with the URLs.

Cheers.— InternetArchiveBot  (Report bug) 00:53, 12 August 2017 (UTC)

The Connection with Logistic Regression Needs to be highlighted
Sutton & McCallum introduced CRFs. In their classic article ("An Introduction to Conditional Random Fields", http://homepages.inf.ed.ac.uk/csutton/publications/crftut-fnt.pdf), it was made clear that just as Hidden Markov Models extend Naïve Bayes, CRFs extend logistic regression. It is embarrassing that the phrase "logistic regression" is entirely missing from a supposedly authoritative Wikipedia article. Prakash Nadkarni (talk)