September 2010
15 posts
“When the first one was released back in 1977 it was phenomenal — a pulpy space...”
– I think I’m hoping the world does end in 2012 : Pharyngula
A bunch of tutorials for RapidMiner
If you are into data mining (everybody, IMHO should be) and you are interested in RapidMinder (you should!), then there is a great resource site with tutorials from Thomas Ott. The tutorials are not only interesting for RapidMiner users in the sense that they show how to use the various features of what is probably the best open source tool for data mining, but also give some insight on how you...
“There are a couple of techniques critical to building high quality graphs to...”
– Making High Quality Charts
“What’s also nice, is that you can build a least-squares regression model...”
– The Gompertz Equation
This is what I was looking for a long time; I did not find a good introduction to using Gompertz curves, and this even comes with an Excel for illustrations.
The statements the author gives in the quote above have to be taken with a grain of salt - sigmoid curves like...
“Jesus and his disciples were walking around one day, when Jesus said, “The...”
– Geeky Religious Joke : Dispatches from the Culture Wars
“A lot of my friends like to take pictures,” writes Erin McCann, a blogger at...”
– TSA Campaign Targets Photographers in Hoodies | Threat Level | Wired.com
1 tag
How to write error messages (not)
I am doing a test drive of RapidMiner, the optically most beautiful open source software for data mining. There are currently two things that drive me mad (I always get mad at new software, there is a low frustration tolerance somewhere):
(1) The tutorials have a New-Age-music soundtrack that is annoyingly loud. There is also exactly the same New Age track for all tutorials, which does not help...
PuzzlOr: Disaster Relief
Here is some mean little problem: optimize the best positions for two air-dropped crates on an island where the villagers need some help. From Analytics magazine’s PuzzlOr, the 10x10 grid “village” looks like this:
I wont spoil the fun by giving a solution, but here is a CP model that should do the job (unless there is a typo in the huts location). Note how the constraint...
Logistic Regression with Eureqa
Logistic Regression (see Wikipedia’s good explanation) is a popular method for building binary predictors/classifiers where the intended result is a probability of the sample belonging to one certain target class.
With Eureqa, you can build logistic regression classifiers by using the logistic function in the modeling task and using only constants, multiplication, substraction and addition...
1 tag
Binary Support Vector Machines as Linear Programs
There was a very interesting talk by professor Carlo Vercellis at OR 2010 on how SVMs can be expressed as linear programms: If you use the L1-norm for the vectors, SVM-learning on linear classifiers becomes linear itself, whereas the standard form with the L2 norm is quadratic. A little bit of googling got me an cookbook explanation of how this is done (see below,...
Thoughts on OR 2010
This was my first OR conference. The positives: Unlike the CADE conferences I used to hang out some 10 years ago, there was a good share of talks where a non-theorist could actually understand what was going on. There is less variety in methods, so a little bit of modelling knowledge will let you much further into advanced topics than a little bit of logic brings you into the field of theorem...
“I think there are some people in the Bush Administration that should be in the...”
– Addington Joins the Heritage Foundation : Dispatches from the Culture Wars
“Wow! That is pretty amazing. It is also completely ridiculous and false.”
– Michael Trick’s Operations Research Blog : Five Sundays, Mondays, and Tuesdays and Operations Research Training 