CoNLL2000 example

[bratao]

Hello @generall ,

Thank you for this awesome project, paper and code base.

I´m always looking for algorithms that can surpass the CRFSuite for a sequential labeling problem that I have. And your paper shows some promising results, with a very novel idea. Congratulations!!
I would love to see if applies to my problem.

In the paper you use some specific "distance functions" ( such as MVDM) and uses as context "Two elements before and after current one."

Could you please show an example in sknnsuite similar to this scenario ? With the actual code I can´t see an straightforward way to do it.

Thank you again!

[generall]

Hello @bratao, thank you for your interest in my research.

I will try to add an example with CoNLL2000 in nearest time.
It is possible to implement CoNLL2000 labelling with existing code if you will apply external preprocessing of sentences and define your own distance function (MVDM requires statistics of all existing data, so you need to "learn" distance function separately too).
If you are familiar with ruby language you can look at ruby implementation of my experiment https://github.com/generall/SkNN-ruby/blob/master/CoNLL2000.rb

But please be aware of using default implementation of data searchers, It performs only full comparison search for nearest elements, so it may take a lot of time to perform labelling.