Is there some particular reason why the LateX stuff is untransformed?
Why a pro/con list is 75% as good as your fancy machine learning algorithm
11–20 of 29 posts
Re: Why a pro/con list is 75% as good as your fancy machine learning algorithm
#12So the author thought, "I'll make an article about pro/con lists", but he thought he'd make it sexist and racist while he's at it. That's hilarious. Couldn't you pick something less controversial, like say, comparing religions, political parties or sports teams?
There is nothing sexist or racist about this article, unless you believe that stating a black person is black is somehow offensive towards black people.
Re: Why a pro/con list is 75% as good as your fancy machine learning algorithm
#13Aren't we, like, not supposed to do this kind of thing anymore?
Do you mean the checklists or the stereotyping / judging on superficial attributes?
Re: Why a pro/con list is 75% as good as your fancy machine learning algorithm
#14Is there some particular reason why the LateX stuff is untransformed?
What browser do you use? Are the files from cdn.mathjax.org not loading?
Re: Why a pro/con list is 75% as good as your fancy machine learning algorithm
#15Is there an obvious reason for drawing vectors from the Dirichlet distribution?
Re: Why a pro/con list is 75% as good as your fancy machine learning algorithm
#16So the author thought, "I'll make an article about pro/con lists", but he thought he'd make it sexist and racist while he's at it. That's hilarious. Couldn't you pick something less controversial, like say, comparing religions, political parties or sports teams?
There is nothing sexist or racist about this article, unless you believe that stating a black person is black is somehow offensive towards black people.
Well, the black person also got 0 points for "smart"...
Re: Why a pro/con list is 75% as good as your fancy machine learning algorithm
#17So the author thought, "I'll make an article about pro/con lists", but he thought he'd make it sexist and racist while he's at it. That's hilarious. Couldn't you pick something less controversial, like say, comparing religions, political parties or sports teams?
There is nothing sexist or racist about this article, unless you believe that stating a black person is black is somehow offensive towards black people.
Actually, he marks black as a "pro" which is an almost hilariously mathematical form of racism.
Re: Why a pro/con list is 75% as good as your fancy machine learning algorithm
#18Is there an obvious reason for drawing vectors from the Dirichlet distribution?
The Dirichlet distribution I chose is merely the uniform distribution over the unit simplex. I.e., it means that all possible h-vectors get equal weight.
It's an attempt to choose as uninformative a prior on h as possible.
Re: Why a pro/con list is 75% as good as your fancy machine learning algorithm
#19Re: Why a pro/con list is 75% as good as your fancy machine learning algorithm
#20Aren't we, like, not supposed to do this kind of thing anymore?
Author here. I know that treating the L^2 ball as somehow equivalent to binary vectors is invalid - I even say so in the post. If it helps, I explicitly point out the place where the real work for the binary vector would have to go. The basic idea is I think is to use the central limit theorem on dot(u,d) and dot(p,d) - but unfortunately the CLT does not show that these two dot products are independent.
Just remember: Your name is on this and you're intentionally pouring gasoline onto an issue that didn't need it.