WIP: MULAN for ranking+multi-class classification

[danyaljj]

http://mulan.sourceforge.net/download.html
https://github.com/tsoumakas/mulan

adding the basic definition of the function
issues:

• the based classifier is not serializable

[kordjamshidi]

@danyaljj why is Mulan important to add? Basically Saul should allow us to make sophisticated structured output models so adding specific packages like Mulan seems inconsistent with the goal of Saul, to me.

[danyaljj]

I needed a ranking algorithm ....

[kordjamshidi]

So why as a part of Saul, you can use Saul as a dependency and write another project.

[kordjamshidi]

BTW, it would be great to have a CCM formalization of ranking and see how we can do it in Saul, in a way that is consistent with our general formalization of structured output prediction using a specific loss, some constraints and classifiers.

[danr-ccg]

Hi,

Please note that Kai-Wei implemented ranking with Illinois SL, and it would probably be easier to port/implement in Saul. The formulation is quite simple, and I’ll be happy to discuss if you want to incorporate in Saul.

          Dan

From: Parisa Kordjamshidi [mailto:[email protected]]
Sent: Friday, October 7, 2016 10:50 PM
To: IllinoisCogComp/saul [email protected]
Subject: Re: [IllinoisCogComp/saul] WIP: MULAN for ranking+multi-class classification (#422)

BTW, it would be great to have a CCM formalization of ranking and see how we can do it in Saul, in a way that is consistent with our general formalization of structured output prediction using a specific loss, some constraints and classifiers.

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHubhttps://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_IllinoisCogComp_saul_pull_422-23issuecomment-2D252400639&d=DQMFaQ&c=8hUWFZcy2Z-Za5rBPlktOQ&r=EoEIPOcXriNKNTRQxPa5uv7hPypxUE6tE_RlGOluh-c&m=j8pT0cfvCnTL5R8JE5mfuIpuEOyDANbpprvz9_b98c8&s=HBWDwRxguSS3pMUs6csWf-Swq8MFVBxYy804YNqr5yM&e=, or mute the threadhttps://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_notifications_unsubscribe-2Dauth_APLC6itvj2gIxPloBm-2Dxck4RlQGIsvwrks5qxxLjgaJpZM4KRjhU&d=DQMFaQ&c=8hUWFZcy2Z-Za5rBPlktOQ&r=EoEIPOcXriNKNTRQxPa5uv7hPypxUE6tE_RlGOluh-c&m=j8pT0cfvCnTL5R8JE5mfuIpuEOyDANbpprvz9_b98c8&s=HQhvtwy8bAiozHgv09bD4tUDxmBG1TIPBBpyfEh39vA&e=.

[kordjamshidi]

Yes, I agree with @danr-ccg and I think that is the way to go for having ranking models in Saul. I will look at Kai-Wei's implementation.