Libraries

[Chengi Liu]

Hi,

  A very dumb question. Why doesnt graphlab have "popular" basic ML algorithms.. like Logistic Regression, Linear Regression?

[Yucheng Low]

Hi,

These algorithms don't require GraphLab. There are easier simpler ways to do those on multicore / distributed systems. (relatively recent work in the distributed case).

That said, it might be a good idea to implement them, even if its just directly on top of the RPC system.

Yucheng

On Apr 15, 2013, at 4:36 PM, Chengi Liu <chengi...@gmail.com> wrote:

Hi,

  A very dumb question. Why doesnt graphlab have "popular" basic ML algorithms.. like Logistic Regression, Linear Regression?

--
You received this message because you are subscribed to the Google Groups "GraphLab API" group.
To unsubscribe from this group and stop receiving emails from it, send an email to graphlabapi...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.
 
 

[Chengi Liu]

Yeah,

    I mean.. I have a constructive criticism for graphlab team. I am a noob in front of you guys.. 

You guys have to admit that both the current algorithms and the programming style are quite sophisticated.

Baring CF toolkit (which has been popular because of netflix prize), as an average "Joe", first I have to search the algorithm, and then you have a distributed implementation where there is no documentation (except references to again a very dense publication) and the option is to read the code which is not that bad except it is because you guys are great coders and your C++ code is also very dense.

So, why I asked this question is if you have something like Logistic/Linear Regression out there, that is one less "dense" publication to read.

Also,it would greatly help me (us) if in the documentation you can include how a particular algorithm can be implemented in GAS abstraction.

For example: https://cwiki.apache.org/confluence/display/MAHOUT/Logistic+Regression

or maybe have a simple plain vanilla implementation without any crazy optimization for some basic algorithms. 

But let me just say, how much appreciative I am for you guys to develop this awesome opensource project.

Thanks

[Yucheng Low]

We do have rather extensive documentation here: http://docs.graphlab.org/ .

The code in toolkits/ tend to quite a bit more optimized (thus less readable). But for some stuff there are simpler versions (i.e. simple_undirected_triangle_count), or in demoapps/simple_pagerank which are quite heavily commented to be readable.

Instead of Logistic Regression / Linear Regression, our "simple" case is PageRank, for which we do have a walkthrough here: 

http://docs.graphlab.org/using_graphlab.html

Yucheng