Path: news.f.de.plusline.net!news-fra1.dfn.de!newsfeed.hanau.net!newsfeed2.scan-plus.net!news.germany.com!out01b.usenetserver.com!news.usenetserver.com!in02.usenetserver.com!news.usenetserver.com!cycny01.gnilink.net!spamkiller2.gnilink.net!gnilink.net!trndny07.POSTED!3abab865!not-for-mail From: "David Cressey" Newsgroups: comp.databases.theory References: <1178130430.502230.295300@u30g2000hsc.googlegroups.com> <1178174344.866049.272810@h2g2000hsg.googlegroups.com> <6ui_h.5979$YW4.662@trndny06> <4639fde8$0$4017$9a566e8b@news.aliant.net> <1178211370.791765.118760@l77g2000hsb.googlegroups.com> <1178221012.371056.145700@u30g2000hsc.googlegroups.com> <1178260494.811737.293620@y5g2000hsa.googlegroups.com> <8kY_h.7025$2v1.1573@newssvr14.news.prodigy.net> <1178376836.208957.255990@u30g2000hsc.googlegroups.com> Subject: Re: A new proof of the superiority of set oriented approaches: numerical/time serie linear interpolation Lines: 56 X-Priority: 3 X-MSMail-Priority: Normal X-Newsreader: Microsoft Outlook Express 6.00.2800.1409 X-MimeOLE: Produced By Microsoft MimeOLE V6.00.2800.1409 Message-ID: Date: Sat, 05 May 2007 17:30:04 GMT NNTP-Posting-Host: 72.70.235.201 X-Complaints-To: abuse@verizon.net X-Trace: trndny07 1178386204 72.70.235.201 (Sat, 05 May 2007 13:30:04 EDT) NNTP-Posting-Date: Sat, 05 May 2007 13:30:04 EDT Xref: news.f.de.plusline.net comp.databases.theory:43797 "Cimode" wrote in message > Thank you for your friendly warning but do not worry about me. To > me, it's a matter of expectations and feedback quality. I have > already came to the conclusion that a few good points made by some > people here may worth the hassle of accepting noise, off topic or > uninformed (unaware?) feedback. > > In my quest of refining and characterizing a computing model that > would correctly represent relations and relation operations, such > quality input is quite refreshing and allows me to double check that I > am not following a dead end conceptual lead. > > fine. Back to the main point. First, inferences and data are not the same thing. An interpolation, or a point derived from regression, or whatever, is an inference, not data. In your original exposition of the thread, you presented it as if the inferences could be returned to the user in place of data, where no actual data is available. My claim is that you do this at your peril. It has to be possible for the user to find out whether the supposed data provided by the DBMS is data that was provided to it, or whether it's an inference based on available data and a model for how the data works. Bob Badour already gave a function (something like (x-1) / (x-1) ) whose value is predictable everywhere except where x = 1. Inference by interpolation is risky with this function. Allow me to add another example: suppose you have some (necessarily incomplete) data about the location of the planets as observed at known places and times (like Tycho Brahe's data). Suppose you have a model of how the planets move (like Ptolemy's or Copernicus' or Kepler's, or Newton's or Einstein's) and suppose to interpolate (or extrapolate) to provide answers at other points in time. I believe this is precisely what the software that comes with some home telescopes does. How bad the answers are will depend, among other things, on how bad the model is. I'm interested in where you're headed, but it think it's a slippery slope. Aside: I wish my French were half as good as your English. I remember a maxim from a high school French book, that I'll try to reproduce from memory, without too many spelling errors: "Les conseils de la vieillesse sont comme le soleil d'hiver. Ils eclairent sans echauffer." For some reason, I think of this maxim often when I read this newsgroup.