|
discussion |
| Primary Documents |
| Conceptual Documents |
| Originating Ideas (Oct. 1998) |
| Current Implementation Ideas |
|
| Links |
| Other Tool Ideas |
|
Orgnet Inflow |
| Licensing Info |
 This work is licensed under a Creative Commons License. |
© Copyright 2003 Michael Bowen.
Last update: 4/12/2003; 6:57:54 PM.
| Sunday, January 05, 2003 |
|
Researcher extraordinaire grimjim has found an interesting paper with work that I believe may prove valuable as we get closer to looking at both the sleeve and the master taxonomy in XR. Recent advances in the study of voting classification algorithms have brought empirical and theoretical results clearly showing the discrimination power of ensemble classifiers. It has been previously argued that the search of this classification power in the design of the algorithms has marginalized the need to obtain interpretable classifiers. Therefore, the question of whether one might have to dispense with interpretability in order to keep classification strength is being raised in a growing number of machine learning or data mining papers. The purpose of this paper is to study both theoretically and empirically the problem. First, we provide numerous results giving insight into the hardness of the simplicity-accuracy tradeoff for voting classifiers. Then we provide an efficient ``top-down and prune'' induction heuristic, WIDC, mainly derived from recent results on the weak learning and boosting frameworks. It is to our knowledge the first attempt to build a voting classifier as a base formula using the weak learning framework (the one which was previously highly successful for decision tree induction), and not the strong learning framework (as usual for such classifiers with boosting-like approaches). While it uses a well-known induction scheme previously successful in other classes of concept representations, thus making it easy to implement and compare, WIDC also relies on recent or new results we give about particular cases of boosting known as partition boosting and ranking loss boosting. Experimental results on thirty-one domains, most of which readily available, tend to display the ability of WIDC to produce small, accurate, and interpretable decision committees.
10:25:45 PM  comment []
|
|
zizka infuriates me for the moment. so much that i'm blogging again at an ungodly hour. well, it's happy hour somewhere on the planet. not here, not now. what is infuriating is the false dichotomy, this fictional bifurcation, this imaginary duality, the fabricated halving of american political opinion. it's so much more complicated. up until this time, we simply haven't been able to account for it all. i believe that is all about to change, slowly but surely. my prediction is that we will, with information systems, do for political ideas what linnaeus did for the classification of life, and all the bible stories about how certain policies came about will become as antiquated as, well bible stories. which is not to say that people will not adhere to these creation myths with morbid tenacity. there is a great deal to be lost, especially in the world of big media, as opinion making gets disintermediated from the zogbys of the world. long ago, there was a goofy little show on race that was hosted by bryant gumbel. it was touted as the first time that (whatever network) was going to take live phone calls from all over america on the issues. even though it was a cliche already, the promise of technology was such that the network believed 'the switchboards are jammed' couldn't happen again. it happened. i called. i'd bet half a million people called. today, i'd bet my nickel that a joint like ebay could handle a simulcast auction of osama bin laden's nuts in a baggie and handle the traffic without a hiccup, and besides that google could breakout the demographics in realtime. ok i exaggerate the demand for osama's gonads. but the fact remains that there is massive, massive infrastructure for completely outsourcing all opinion polling. the surprise is that we already own it. call it 'votester'. somebody will build it and every red blooded american with dialup and an opinion will come. and it will be ugly as sin to find out what kind of snarly creatures we are when given the chance to word our own opinion surveys. been on usenet lately? there's your hint. but that's not the point. the most crucial obstacle to votester is harnessing and corralling the interests of wonks, middlebrows, lowlifes, propogandists and all matter of polloi onto singular subjects. this is where linnaeus comes in. what your congressman knows that you don't know is how bills are marked up. and what his staff knows that he doesn't know are the implications of markup vis a vis similar legislation and/or existing law, and what the lobbyists know that the staffers don't know is how much that costs in real life. or so goes my theory. what nobody seems to know, or care, is how some other legislative body did this same thing 20 years ago and the pros and cons of that. so imagine a master taxonomy which enumerates all of the interests and implications all of the debates and deals and all of the interested parties that had something to say. what you begin to approach is the realtime marketmaking of political opinion. how many people are weighing in on abortion rights today? 13.5 million? that's a 15 week high! which direction is the sentiment going? well the partial birth abortion foes are losing ground in the following zipcodes... you get the picture. all opinion all the time 24/7 from every stakeholder in the voterverse. i'm sure this all sounds like utopia, and for many i'm sure it sounds like dystopia. that doesn't change the fact that somebody is going to try and build a true, open political opinion market and that people are going to track it. i've said enough at this time of the morning, but at least i'm happier having said it than i was after reading zizka. no offense zizka, but i've seen the future. 1:21:42 AM comment []
|