trading, neural nets & data viz

Luiz Paulo · Post by **Luiz Paulo** » Tue Mar 15, 2005 8:47 pm

apine,

check the following white paper out. It is from a company called Prediction Dynamics, which built a system to use NN ensembles in financial modelling.

they have some neat ideas regarding rule extraction from NN models - their approach basically uses backprop NNs.

i believe they went under though - the website www.predictiondynamics.com went down a couple of months ago...

[url=/beta/User%20Files/80/Prediction%20Dynamics,%202003%20-%20Nonparametric%20multifactor%20time%20series%20modelling.pdf]Attached File: Prediction Dynamics, 2003 - Nonparametric multifactor time series modelling.pdf[/url]

chiral3 · Post by **chiral3** » Tue Mar 15, 2005 8:56 pm

could you re-post that paper and make sure it is a pdf or that the extension is on. thx

Luiz Paulo · Post by **Luiz Paulo** » Tue Mar 15, 2005 10:15 pm

there seems to be some kind of problem, which i can't fix. will keep on trying, other wise i'll try pm'ing to someone (anyone?) who could post.

[url=/beta/User%20Files/80/pred.pdf][/url]

InspectorKojak · Post by **InspectorKojak** » Tue Mar 15, 2005 10:22 pm

Right click on the link, then select "Save As..." and save it do disk as type "all files".

Luiz Paulo · Post by **Luiz Paulo** » Tue Mar 15, 2005 10:25 pm

it worked for me

thx Inspector

apine · Post by **apine** » Tue Mar 15, 2005 11:30 pm

Luiz Paulo, thanks for the paper. it will take some time for me to digest.

some practical notes on PC:

there are a few data visualization programs available online, like kdnuggets. but they miss the point of interactively exploring the data like Inselberg's software does. last time that I asked, he was charging $5K for it. maybe its worth it, but it is hard for me as a sole prop to justify.

apine · Post by **apine** » Thu Mar 17, 2005 3:26 am

i read the paper. it looks interesting. i certainly have not seen anything else that breaks a neural net into a decision tree. i saw on the net an article on the firm from mid-2004. but the web site is "forbidden."

is anyone aware of academic work that Prediction Dynamics used to build their software. are these guys out of a university?

apine · Post by **apine** » Thu Mar 17, 2005 3:29 am

i guess they did go under:

"Prediction Dynamics® is a registered trademark of Stewart W. Wilson. There is no connection with a defunct Ireland corporation that had been using the name."

i saw this on Stewie's site.

apine · Post by **apine** » Thu Mar 17, 2005 3:41 am

http://www.cs.tcd.ie/Padraig.Cunningham/online-pubs.html

this is the site for one of the Prediction Dynamics guys. look at the

Explaining Predictions from a Neural Network Ensemble One at a Time

article. i printed it out to read. let me know if anyone has looked at this before.

Luiz Paulo · Post by **Luiz Paulo** » Thu Mar 17, 2005 5:11 pm

the founder of pred dynamics was john carney, a ph.d under cunningham. i tried to track down his thesis but seemed unable to find it - it was the basis for the system they were trying to sell (and the white paper). if someone obtained the thesis it would be super cool

i recall reading that Deutsche was using the system...

i got another paper from carney lying around here somewhere.