I am presenting at the Teradata Universe Beijing this year.  I will be the first speaker for the telecommunications sessions (the other sessions being 'Banking' and 'National Accounts').

My talk will be a high level summary of the customer analytics we conduct, how we present to corporate executives, and the benefits our analysis has brought our customers and the company.

It's a single day conference, and I'll be taking a few extra days for sight seeing in Beijing with my wife.

Feel free to say hello if you see me at the Teradata Universe Beijing 2008, May 30th at The Kerry Centre Hotel, Beijing.

Cheers

Tim

 

 - More Info -
http://www.teradata.com.cn/en/main.html

Teradata Corporation warmly invites you to attend Teradata Universe Beijing 2008 which will be held on 30^th May at the Kerry Centre Hotel, Beijing.

With theme"Insights into Action", this year's conference focuses on how the world's leading companies effectively gain insights from large volume of data and transform such insights into strategies and actions that help them preempt the opportunities before its competition.

Join us to learn some of the world's data warehouse best practices from local and inter-
national speakers including Mike Koehler-President & CEO of Teradata Corporation,Step
hen Brobst- CTO of Teradata Corporation and speakers from Bank of America, Barclays Bank, ABN AMRO Bank, Chang Hwa Bank, Optus SingTel, SmarTone Mobile Communications, China Telecom Group Hunan Corporation, Intel, Nationwide Insurance, China Post and our key partners such as SAS, Informatica, LSI, and Guomai Technologies, Microsoft(China)Co.,Ltd., Beijing Donghuahechuang Digital Technology,TEAMSUN, Beijing Eastcom Info-Tec.Co.,Ltd and many more. They will be presenting at the conference and sharing with you their real-world experiences in leveraging data warehousing technology to turn insights into actions.

Teradata Corporation, the world's largest company solely focused on raising intelligence through data warehousing and enterprise analytics, has business in more than 60 countries. Our customers are leading companies in a variety of industries such as communica-
tions, financial services, securities, manufacturing, retail and hospitality, insurance, government and public services, media and entertainment, health care and pharmaceutical, and travel and transportation logistics. We are committed to help you outperform your competitors and taking the lead in the industry with our powerful solutions!

Teradata Universe 2008 is undoubtedly an indispensable event on your business agenda!


You can view the conference agenda online;
http://www.teradata.com.cn/en/Speaker_9.html


And oh god...
- You can even see a shockingly dodgy wedding picture of me http://www.teradata.com.cn/en/Speaker_9.html

If you would like to view the powerpoint slides that accompanied this presentation (and the practice run I did for the IAPA), then you can download it from here;

http://www.iapa.org.au/Environments/edoras/Resources/IAPA/Tim%20Manns%20Teradata%20200806.ppt

http://www.iapa.org.au

also the generic download section of the conference;

http://www.teradata.com.cn/en/download.html

Cheers

Tim

Hi Tim, congratulations for your opportunity.

Nice powerpoint, but I wanted to see your little secrets when it comes to field generation and model usage 

I hope someday you make a powerpoint specifically with these things!

Good luck in Beijing!

Hi Arkantos,

re:

Nice powerpoint, but I wanted to see your little secrets when it comes to field generation and model usage.
I hope someday you make a powerpoint specifically with these things!


->   If you are ever in Sydney then buy me several beers and just maybe 'll let a few secrets slip :) 

I usually use standard default options Neural Nets in Clementine.  For me its the data processing and derived fields that generate the results (not algorithm options).  I try to use the largest sample sizes I can (although I always Balance reduce and never boost data), and always remove problems of skewed values (eg problems with 'power law' occurs a lot for me).  Solve these data processing problems and your predictive and clustering models should benefit for it.

Cheers

Tim

Hi Tim,

Last year I was about to go to Sydney but finally I couldn't, this year I'll probably be going to Turkey. Anything interesting there? :) When I go to Sydney I'll let you know, I'll also get you drunk for your City of Heroes user and password =D

I noticed what you said as well... changing little bits of the model nodes is just fine tuning, the big picture you get it when you execute the default node. Do you always keep original balance even if the "1" class is very small? By skewed values you mean erroneous/impossible/contradictory data? Or just extreme values? Don't know what's power law :(

Cheers and good luck in Beijing.

Arkantos

Hi Arkantos,

I always balance the data by reducing "0" class to be near the same size as "1" class.  Some people boost "1" class, but I don't.

Power Law just means is very positively skewed data.  In my analysis 2% of customers (data) have values at least 1000's larger than anyone else.  This means that my data is very skewed, and the main problem is how to re-distribute the data into something that a neural network or decision tree can use.  In my experince, using raw and skewed data in predictive analysis doesn't produce good models.

Cheers

Tim