Disqus - Latest Comments for brendano

Re: Amelia McNamara | Blog | Participatory sensing

brendano — Fri, 08 Apr 2016 18:01:12 -0000

btw i totally just used the dawid/skene model (the original version before the ipeirotis paper, where you dont know the gold standard labels) to teach EM ...

Re: Amelia McNamara | Blog | Participatory sensing

brendano — Fri, 08 Apr 2016 17:59:19 -0000

great stuff! i dont know if dan sheldon's worked on data quality specifically but has done a lot of modeling with the ebird (or similar?) data ... https://people.cs.umass.edu...

Re: Jobs I've Held - The Gong Show

brendano — Fri, 09 May 2014 10:34:28 -0000

+1 for college mascot :P

Re: What’s the matter with logistic regression?

brendano — Mon, 11 Nov 2013 14:38:54 -0000

re the separation plots, these things are quite fun. Here are some related plots I made a while ago: http://web.archive.org/web/...

Re: Topic Models aren't hard

brendano — Wed, 24 Apr 2013 11:40:29 -0000

"The next 5 or so years led to a flurry of incremental model extensions and alternative inference methods, though none have achieved the popularity of their namesake." ==> I'm not sure that's true. I see tons of people use the Steyvers and Griffiths collapsed Gibbs Sampler in applications papers -- it's conceptually and computationally simpler than the Blei et al. variational EM method.

Re: http://medriscoll.com/post/13034410709

brendano — Sun, 20 Nov 2011 00:52:09 -0000

Yes, it is in the data. Text is rather non-interesting from a machine learning perspective. It's already had dimension reduction to the important semantic elements; this is why bag-of-words linear models do so well for many text tasks, and to get deeper semantics out of it you need linguistics in your models. Sensory data, on the other hand, has all sorts of statistical redundancies.

Re: Stephenson’s Subjective History

brendano — Thu, 22 Sep 2011 18:26:56 -0000

This essay got me so excited about Linux and open-source back at the start of college. I still think it's by far the most entertaining account of that era and its wars over the desktop OS. I mean, every other page (I have it in book form!) has gems like "This is a little bit disingenuous... like the Batmobile dealer claiming that he is not really in competition with the others because his car can go three times as fast as theirs and is also capable of flying." :)

Re: cheat sheet for scala syntax

brendano — Tue, 05 Jul 2011 17:31:58 -0000

it's open source, i encourage you to help

Re: cheat sheet for scala syntax

brendano — Tue, 05 Jul 2011 17:25:26 -0000

you should write it up as a pull request on github :)

Re: cheat sheet for scala syntax

brendano — Tue, 05 Jul 2011 17:20:03 -0000

thanks. fixed.

Re: Machine Learning: A Love Story

brendano — Fri, 06 May 2011 14:26:28 -0000

i feel like machine learning is relevant these days in part *because* it embraced statistics.

Re: Machine Learning: A Love Story

brendano — Fri, 06 May 2011 14:07:50 -0000

When was machine learning a marginalized field?

Re: http://blog.smola.org/post/1130285201

brendano — Thu, 28 Apr 2011 10:06:57 -0000

Hi, your last point:

"""Many practitioners are scared to pick up a paper with many equations but they might be willing to spend 10 minutes reading a blog post."""

is absolutely true. When I was working in industry I definitely felt like this. And if your background is from outside the field in question, it's much more useful to read blog posts than research papers; the latter take longer to evaluate their relevance, and typically you have to go through a lot of them to find what you need.

Re: Sharing code, API's, and a Readability API - A Computer Scientist in a Business School

brendano — Mon, 13 Dec 2010 02:23:50 -0000

APIs are only as stable as the platform they run on. They have to be maintained just like other software. I think they seem more stable because we're used to thinking of servers as not changing.

But we use lots of library dependencies and other not-replicable practices because they make development easy. I think the solution is to save the dependencies in a smart manner. Various languages have different ways of doing this (in Java, includes all the .jar's; in Python, use "virtualenv", etc., so the only dependency is the language runtime), or it could be done at the system level, by saving a virtual machine image (so the only dependency is an EC2-like service that can run Xen x86 or whatever VM images).

Re: Hacker News, automagically organized [unofficial] - MetaOptimize

brendano — Mon, 18 Oct 2010 16:37:38 -0000

Would be nice to remove "http" as a token...

Re: The Evolution of Report Summaries in WikiLeaks Data Over Time

brendano — Sun, 17 Oct 2010 17:35:27 -0000

Do you think LDA as a pre-processing step is useful for this sort of exploratory analysis, compared to analyzing the individual terms' changes to frequencies over time?

Re: cheat sheet for scala syntax

brendano — Wed, 01 Sep 2010 22:39:32 -0000

> You have the following at the start of the Object Orientation section
>
> class C(x: R) same as class C(private var x: R)
>
> I think that should be "val x" not "var x".
>
>
> Here's the simple code to test it:
>
> class Foo(x: String) { x = "hi"} generates a compile time error: reassignment to val
>
> class Foo(private val x: String) { x = "hi" } generates the same error
>
> class Foo(private var x: String) { x = "hi" } compiles without error
>

Re: KEA Keyphrase Extraction as an XML-RPC service (code release)

brendano — Fri, 20 Aug 2010 01:45:48 -0000

Nice summary of all those tools out there...

Re: Open Source iPhone Apps

brendano — Wed, 11 Aug 2010 19:59:11 -0000

Er... Don't take me as an authority, I wan't using Macs during the time in question. I think I read that somewhere, and from what I've seen using a mac for the last several years, it seems true...

Re: Open Source iPhone Apps

brendano — Sat, 07 Aug 2010 15:23:30 -0000

The Macintosh developers' culture, going back a long time, was always shareware-centric and not opensource- or freeware-centric. I personally think this culture leads to more highly polished apps with better user experiences, albeit non-free. Apple replicated a form of this for the iPhone ecosystem; micropayments are a better form of what shareware was.

Re: cheat sheet for scala syntax

brendano — Fri, 25 Jun 2010 13:40:04 -0000

> Nicholas Sterling <nicholas.sterling@gmail.com> (unregistered) wrote:
>
> This is the greatest page -- I hope the Scala dev team knows about it!

I might've sent it to one of their lists before... I forget.

>
> Once you've read a book on Scala, this is just the kind of to-the-point
> reminder you need of what is available and how to incant it.
>
> Thanks so much for your work on this; I hope you have the time to keep it
> going.

I actually haven't used Scala too much for a while. If I get to back to it
I'll probably work more on this -- I found it the only way to keep track of
Scala's complexity. Which maybe is why I stopped using it...

Re: Boston-Bound: Leaving USV and Next Steps

brendano — Tue, 30 Mar 2010 15:07:54 -0000

Good luck!

Re: This is an experiment in scientific methodology

brendano — Mon, 01 Mar 2010 22:29:03 -0000

> However, I would hardly say that MT has a culture of open-source.
> I think it's like most of NLP that some packages get released, but most never do.

fair enough. however, they have had lots of success using open-source systems as a common shared platform, as a base for further experiments. that at least makes things more replicable compared to always building from scratch.

Re: This is an experiment in scientific methodology

brendano — Mon, 01 Mar 2010 19:05:42 -0000

sounds like fun!

fyi, in machine translation there's a good research tradition of big opensource systems that everyone can share for precisely these reasons of replicability and the like. e.g. giza++, moses, and now joshua

Re: Serious Business looks for life beyond Friends For Sale!

brendano — Mon, 30 Nov 2009 19:17:41 -0000

Mistake: Ryan Ferrier was not the founder of Powerset. As he says himself: http://twitter.com/pearlzof...