Re: ST Update
Timmy Wilson
I've been spending time learning/playing with Edwin Chen's RBM implementation:
http://blog.echen.me/2011/07/18/introduction-to-restricted-boltzmann-machines/
I'm using it to find niche twitter communities:
https://github.com/smarttypes/SmartTypes/blob/master/smarttypes/neural_nets/cluster_twitter_users.py
I think the key is layers (ie recursive RBMs) -- i haven't got there
yet -- that's my next step
I'm also messing around w/ a sparsity constraint --- although i have
it commented out right now
line 68 -- https://github.com/smarttypes/SmartTypes/blob/master/smarttypes/neural_nets/intro_to_rbm.py
My goal is to get a serial algo working well (finding niche
communities) -- and then work on speeding it up -- either through the
GPU -- or some form of parallelism -- or both
There's a great write-up on a nice ubuntu/python math environment --
http://blog.adamdklein.com/?p=416 -- this is what i'm using for the
server -- with the addition of hadoop and/or theano
(http://deeplearning.net/software/theano/) -- or maybe just
http://docs.python.org/library/multiprocessing.html
I've been working w/ a guy locally (Rob Tucker) who's helping me get
the hardware in order -- Rob used to manage the datacenter @
http://www.fidelityvoice.com/ -- Rob 'knows a guy'
We're toying around w/ the idea of selling on-demand machine learning
+ GPU + hadoop services
Sorry to be so verbose -- net/net things are moving along
Oh -- i'm planning on taking these 3 classes next semester, which
should help as well:
- http://www.launchpad-class.org/
- http://www.nlp-class.org/
- http://www.pgm-class.org/
On Sat, Nov 19, 2011 at 7:36 AM, Joe O'Donnell <gal...@yahoo.com> wrote:
> Hi Timmy,
> That was definitely exciting to see the RepSoft algo. I am still studying Neural Nets to get a better handle on how to use it. Microsoft announced they are giving up on their competitor to map reduce and going with Hadoop:
> http://www.zdnet.com/blog/microsoft/microsoft-drops-dryad-puts-its-big-data-bets-on-hadoop/11226
>
> That combined with their social analysis product announcement would seem to mean that we are in a crucial highly competitive early phase for social analysis products right now. It's nice to have that confirmation of the existence of a market.
>
> Take it easy,
> Joe
>
>