Linear Algebra Libraries - n.2

In a previous post I discuss some some options about linear algebra libraries. Especifically I'm looking for a tool to eigenvalue (and maybe eigenvectors) calculation. In the previous post I have mentioned Eigen which is a (free) C++ template library so is easily to "install" (just copy the files somewhere) and use (just include the headers). With Eigen you can calculate eigenvalues (and eigenvectors), and is very easily (intuitive) to use. So Eigen is very appropriate for prototyping some program that requires linear algebra tools at some point. The people that develop Eigen says that Eigen is very fast (in some cases even fastest that other non template libraries). But I think (I have not tested it yet) that Eigen may be fast in basic linear algebra operations. In the case of eigenvalue problems it gives to me the sensation that is slow. So I'm looking for an alternative to it, and found about CULA, which basically is an implementation of LAPACK for GPGPU (graphic cards accelerators)
.
I'm just very new about this. I have several questions.

1) In my laptop I think that I do not have a GPGPU card. So: ¿I will need to work directly with the machines of the cluster where I work (machines that have a GPGPU card on it)? ¿Are there some kind of GPGPU simulator? 2) ¿What need to be installed and how do you install it in order to start doing some trials with CULA? 3) ¿Does CUDA need to be installed also?
In order to answer this questions, in the getting started section of the CULA home page one can find some relevant info. It seems that CUDA need to be installed, and also a (nvidia) GPGPU card is needed.

I think that, as CULA is a GPGPU implementation of LAPACK, maybe I can use LAPACK on my program in the laptop, and then, with relative ease adapt the code to run CULA. So, maybe, starting a try with LAPACK will be useful in the first place. Also playing with LAPACK will give to me an idea of the performance of Eigen.

Here I found that it seems not so much directly to use LAPACK from a C++ code.

Here another example of using LAPACK from C++ code.

Here another example of using LAPACK from C++. I like the style of this blog.

So, it seems that deal with LAPACK is a "crap". But I remember that Armadillo may be a convenient way of using LAPACK from C++ (but it will not serve to the purpose of learning LAPACK in order to after learn CULA). Note that Armadillo not necesarilty requires LAPACK and some BLAS library. But Armadillo can be linked to LAPACK (and some BLAS library) in order to achieve better performance and bigger functionality.

The latest version of LAPACK is 3.3.1, but my ubuntu comes with version 3.2.1 previously installed, so I will try first without installing the latest version of LAPACK. On the other hand, for the BLAS support, my ubuntu hasn't got installed ATLAS. F...k, here says that the installation process of ATLAS is a f...ng crap. It says that it may take several hours.

TODO ... finish this.

Salut !

Learning - BGL and property map

So, when reading about BGL I found difficult to grasp the idea behind property maps. Luckily, in this post there is a nice series of questions and answers that helps to clarify the matter.

Here, another post that shows some simple examples about property map's on graphs.

In this post, it is discussed about a compressed graph representation for static graphs. Here the corresponding link to the BGL home page.

Salut !

R statistics - Dendrograms in R

The idea here, is to explore how to plot dendrograms using R.

For the moment I just will put a list of links:

http://addictedtor.free.fr/graphiques/RGraphGallery.php?graph=79
http://rwiki.sciviews.org/doku.php?id=rdoc:stats:dendrogram

Keywords :

dendrogram
heatmap
hclust

Salut !

Buying a netbook 2011

So, I'm planning to buy a netbook.

I had an offer in a local store that ask for $2285 for a BGH MT10. In mercadolibre there is an offer about $2199 (se here). It seems a nice machine.

Also, here one can see that it has a relatively nice processor (for being a netbook).

Salut !

Resources - Complex Systems

Here a list of resources about Complex Systems and Complexity.

: Institutions, Organizations

:: Complexity-NET
York Centre for Complex Systems Analysis
Bristol Centre for Complexity Sciences
Institute of complex system symulation (ICSS)
Warwick Complexity Science Doctoral Training Center
Santa Fe Institute
Center for the study of complex systems
Brooking Center on Social Dynamics and Policy
Northwestern University on Complex System
Nederlantse Organizatie voor Wetenschappelijk Onderzoek
IFISC (españa)
Reseau National des Systemes Complexes (Francia)
Agent Based Computational Economics and Financial Modelling (Essex University)
Complexity and Networks Group (Imperial College, London)

: Blogs

:: Nirkmukta (Promoting Science and Free Thought in India) - COMPLEXITY EXPLAINED

: Books

:: Cosmic Evolution: The Rise of Complexity in Nature - Eric J. Chaisson

Salut !

Neural Networks - Resources

So, as maybe neural networks will be one of the topics in which I will work in, I use this post to collect resources about neural networks.

sources of answers to FAQs about NN

monthly posting to the Usenet newsgroup comp.ai.neural-nets

databases

Face Recognition Home Page

Salut !

Numerical integration methods of ODEs - n.1

So, in this post, I will collect information about methods for numerical integration of ODEs (ordinary differential equations).

For the moment, I'm not looking for something very precise, but it is preferred some method that can deals with stability and stiff problems.
A list of links:

First order ordinary differential equations
QUAM- A novel algorithm for the numerical integration of stiff ordinary differential equations (In this "paper" there is a definition of stiffness)

What is stability and stiffness?

: Stiffness

A problem is stiff if its jacobian matrix has at least one eigenvalue with a large magnitude and a wide spread of eigenvalue magnitudes. (source here)

: Stability

TODO ...

Salut !