Infinisearch - Sweet links!! Thanks for that one!!

Timkin - essentially I'm using EP to develop the complete structure of the decision tree. The feature used for split, the value of the feature, and the topology follows. I'm hoping to put together a short tutorial on what I've done - I'll let you know when I'm finished. Hopefully it'll be done this week. (hope hope hope) And actually, it's not game related, I work in Cheminformatics/Bioinformatics - I've probably mentioned that before - but I think it could be useful for game related issues, ala my Pac-Man idea above. My bigger concern is giving away the technique before I have a chance to publish it with my name on it. ugh. Crazy academic pangs.

Speaking of academia, how has the job interview gone?

-Kirk

BTW, do you have any links to decision tree research pages at Monash? Thanks!

[edited by - KirkD on October 1, 2002 8:35:42 AM]

quote:Original post by kirkd
Timkin - essentially I'm using EP to develop the complete structure of the decision tree. The feature used for split, the value of the feature, and the topology follows.

How does your evaluation stack up against the standard comparitive algorithms for evaluating trees (eg C4.5)? Have you been using the standard UCI data sets for performance evaluation? I'd love to hear how your algorithm performs compared with current methods.

quote:Original post by kirkd
My bigger concern is giving away the technique before I have a chance to publish it with my name on it. ugh. Crazy academic pangs.

That's not normally a problem if you've got a draft paper prepared. It's not really an issue though since you can't copyright your algorithm, only the specific implementation of it. You wouldn't really want to anyway. If you've got a good method, people will want to try it out for themselves and test it out again and again. Just write up a draft of a paper and submit it to a conference or journal... it's very unlikely that someone would try and claim the idea as their own once you're at that stage.

quote:Original post by kirkd
Speaking of academia, how has the job interview gone?

I had a meeting over at La Trobe this morning to discuss research ideas and funding options. Sounds very promising as they have a lot of industry support and research funds. I'm putting in a submission to do research on Dynamic Enemy Strategies, which incorporates some user modelling and plan recognition with adaptive behaviours. On other fronts, I've got a tentative offer to do a post-doc in seizure prediction (neuro-physiology) which placates my maths background and interest in the brain... I just can't decide if I want to indulge my need for money (the whole non-linear prediction bit pays very well in industry) or my love of computer games (which doesn't pay so well)!!!

quote:Original post by kirkd
BTW, do you have any links to decision tree research pages at Monash?

You'd think a comp sci department would have good web pages... and well, they look pretty, but...

You can probably find some publication information on the web pages of David Dowe, Chris Wallace, Ingrid Zuckerman and Geoff Webb. Ingrid, in particular, spent her sabbatical last year over at Microsoft, helping them implement some decision tree stuff developed at Monash by Chris Wallace. Geoff does a lot of work on machine learning and classification, as does David and Chris. Chris is the father of MML, which is used extensively in classification problems.

I look forward to hearing how your research turns out!

Cheers,

Timkin

[edited by - Timkin on October 2, 2002 2:16:18 AM]

quote:Original post by Timkin
How does your evaluation stack up against the standard comparitive algorithms for evaluating trees (eg C4.5)? Have you been using the standard UCI data sets for performance evaluation? I'd love to hear how your algorithm performs compared with current methods.

I haven't compared it to C4.5 specifically, but in comparison to the standard Recursive Partitioning algorithm in S-Plus and one of the industry standard applications we use (Cerius2) it gets about 5-15% better predictions on an external test set of observations. I've been doing a 10-fold cross-validation process and averaging the results across multiple runs, so I'm confident of the overall performance.

quote:Original post by Timkin
That's not normally a problem if you've got a draft paper prepared. It's not really an issue though since you can't copyright your algorithm, only the specific implementation of it. You wouldn't really want to anyway. If you've got a good method, people will want to try it out for themselves and test it out again and again. Just write up a draft of a paper and submit it to a conference or journal... it's very unlikely that someone would try and claim the idea as their own once you're at that stage.

I'm not too worried about copyright or patent issues. I'm more worried about having time to get it all put together and getting it published in a good journal before someone else with more time (and likely more hands) scoops me.

quote:Original post by Timkin
I had a meeting over at La Trobe this morning to discuss research ideas and funding options. Sounds very promising as they have a lot of industry support and research funds. I'm putting in a submission to do research on Dynamic Enemy Strategies, which incorporates some user modelling and plan recognition with adaptive behaviours. On other fronts, I've got a tentative offer to do a post-doc in seizure prediction (neuro-physiology) which placates my maths background and interest in the brain... I just can't decide if I want to indulge my need for money (the whole non-linear prediction bit pays very well in industry) or my love of computer games (which doesn't pay so well)!!!

What??!!?? I thought game developers made boo-koo bucks! Of course, in academia you can do what you want, when you want, at the pace you want.....yeah...I fell for that one myself.


Thanks for the links. I'll check 'em out. And, I'll let you know once I get things written up and in a semi-stable form...

-Kirk

[edited by - KirkD on October 3, 2002 8:35:09 AM]

quote:Original post by kirkd
What??!!?? I thought game developers made boo-koo bucks!

Definitely not... it''s a very underpaid industry for the skills required, particularly when compared to other areas of IT.

quote:Original post by kirkd
Of course, in academia you can do what you want, when you want, at the pace you want...

So long as you''re publishing, you can do what you like (up to a point)... it''s all PUBLISH, PUBLISH, PUBLISH... and if you don''t, you don''t get funding and then you are soon out of work!

Good luck with getting everything finished.

Cheers,

Timkin

Ferretman,

Hmmmm...Game AI Programming Wisdom II, eh? Hmmmm...I hadn''t thought of that as I consider myself an absolute amateur. What is the deadline for such a submission/proposal? If I can get the Pac-Man control system together in a reasonable period of time and get it working properly, that might be a good idea. Thanks for the tip!

-Kirk

Fup,

Thanks for the link. I may just put together such a proposal. The question is whether I can get my system working for a game oriented article in the time frame. Sounds reasonable assuming the process goes well and I have some solid results (data) to show for the effort.

Who''d''a''thought I would start out making decision trees for Cheminformatics and Bioinformatics and then write it up in AI Wisdom. 8^) "If I know one thing for sure, it''s that I don''t know where I''ll be tomorrow."

-Kirk