Re: Euphoria Contest
- Posted by petelomax at blueyonder.co.uk Nov 13, 2002
- 390 views
Although the AI thread has taken over somewhat, I had a few thoughts about a "contest" which would actually produce a few useful things. Basically, start with a framework of an application, designed modularly enough that bits and pieces could be coded and judged separately; but together comprise a reasonable real-world application. What is needed is an application that can quite easily be broken down into such modules; several people submit their "components" and in the end the best of breed in each category wins, subject I guess to each contestant only submitting eligible entries for one such module Examples (obviously other suggestions welcome): 1) TV listing. There are numerous sites on the web where TV listings are freely shown. Some are fast and relatively poor quality, some are slow, difficult to navigate, some are either overloaded or down for hours/days at a time, some are chock with adverts and/or excess repeated details/icons/images/etc. a) scan available sites within a set time period b) store the details in a suitable database locally c) filter depending on user preferences/available channels/ country of viewing d) interesting user interface d.1) multi-channel rows/columns/timeslots d.2) compare known details of VCR settings d.3) printout suggested viewing times. e) purge/reorganise utilities (automatic, naturally) f) of course a print preview would be nice 2) Betting Expert. As we know, bookmakers base their odds not actually on the real chances of winning, but on the total amount of bets placed by the public. There is of course no better way to be sure of winning than not to bet in the first place, but this is an interesting concept. a) get all available info from net a.1) historical a.2) save todays race results b) suitable database, eg age sex time of month for mares jockey going handicap length of race owner racecourse number of horses in race finishing place <lots more...> c) ranking algorithms for each race d) suggested bets d.1) winnings/losses paper/real monies e) comparison of guessed ranking/actual results e.1) identify conditions program predicts well/bad f) registration spanner in the works eg: 10% of suggested bets are randomly picked (aka drivel) until the app is registered That is only two suggestions. I suspect something simpler is needed. Obviously, the framework would have to be rather loose/inefficient, eg defined fields for a database referenced by name rather than a simpler enumerated type, with an unrecognised field returning zero or null; module communication via routine-ids with the same limit (ie unimplemented routines return zero/null). An unpredictable (luck) element appears: the "best of breed" may not clearly be one module over another, but if j&d is better than either k&d or k&b, j wins despite j&b being worse than either k&b or k&d. (subject to the modules being from different authors?, or not??) Before you start, someone needs to produce a "naff" version which just about works. My last word: I enter contests for the fun & in hopes of the kudos; an actual prize over my name in lights is irrelevant. Pete