"A Man's a Man for all that!" - Rabbie Burns

"Religion? No thanks. I prefer not to outsource my brainwashing."
Trying to get your average Joe creationist to understand the phrase scientific theory is as hard as getting a fish to enjoy mountaineering. Its an unimagined world for them - it requires a complete reversal of their normal modes of thinking and being. The fact that humans could explain the complexities of this world without a creating God is a world view they cannot grasp. It's like asking a tuna if it appreciates the view from the top of Mount Everest.

Aug 19, 2010

Geeky Wickerman

Before I start posting about new stuff I should really give an update about what I've been up to during the period when this blog went quiet.

Most of my time as usual has been spent either directly doing my day job or travelling up and down Scotland to get to our main offices. There is a major re-organisation coming up for us and I have been part of a small team which has done some scenario planning for how we are going to operate in the future. I had to do the write up and pepare the scenario planning report which was a major task. In the current financial climate nothing is certain for anyone so we'll need to see what the next six months or so brings.

As part of some background preparations for that scenario planning task I started doing some background investigation of different modeling tools that might be useful for widening our understanding of the kind of data that we deal with. This has ended up being a really interesting journey ( well for a geek like me - ,maybe not for others).

I have had an interest in the whole subject of machine learning and data mining for some time and in particular I have dabbled with the use of neural nets and SVM's ( support vector machines ) before but over the last five or six months I have come across some excellent pieces of software ( most of it open source) that can be used for a range of modeling tasks.

In particular I have found excellent software for using bayesian networks, and a whole range of open source machine learning and data mining libraries and GUI's.

I know this isn't the sort of stuff that floats everyone's boat but on the off-chance that someone else might have some interest in it I may post a little more about this in future posts.  I will probably not go into great detail in  this blog though as I'm thinking I may start a separate blog where I can post more detailed stuff on these subjects.

A couple of weeks back I was at the Wickerman festival in Dumfrieshire. I guess I must look like a criminal because my car was stopped on the way in and we got the full police search with sniffer dogs and everything. I don't think I saw half the bands that were playing - it was my usual festival experience - wandering from stage to stage only to find that I catch the end song from the set - then wandering to the next stage and the same thing happens. Maybe that's just my luck at festivals - or maybe I should buy a programme!  The Wickerman burning was spectacular though. I thought it was going to be a wickerman that was say about 50 ft high.  Far from it.  The Wickerman was HUGE and made an amazing sight as they set light to it at the end of the final night.

5 comments:

Looney said...

The technical areas sure sound like fun. If you do post, I will be an enthusiastic reader!

Rummuser said...

You are so right! It doesn't float my boat!

Bunc said...

Rum,
I had a feeling there would be a split on this one - that's why I reckoned on only posting a bit about it here and leaving the details for another blog.

Looney,
I'll get into some of the technical stuff in more detail elsewhere. As a taster I have recently been using FAMM ( a C neural net library.) I've been training an ANN with a model of a normal distribution so that it outputs the CDF. As you'll know probably better than I do I gather there are no exact methods for calculating this for a normal distribution but there are a number of numerical method approximations. I have been interested in seeing if I was able to train an ANN with an example set of z scores and associated CDF values so that the ANN could then be used to produce interpolated CDF values for z scores it hadn't been trained with. I.ve also been using NETLOGO ( a mutli agent simulation environment in Java) which is both a lot of fun and has some interesting modelling possibilities ( Repast is a very popular mutli agent sim system that i've been trying out also). You might also wanat to check out GENIE - generates and operates with Bayesian belief nets. The main datamining tool I've been using is RapidMiner.

Bunc said...

Thats me not checking my spelling again - it's FANN not FAMM.

Looney said...

That FANN system looks like a good little learning platform. I had been wanting to learn something about these, so I will down load and see what is in there. Thanks.

Related Posts by Categories



Widget by Hoctro | Jack Book
About Us | Site Map | Privacy Policy | Contact Us | Blog Design | Ayrshire Blog Creative commons License