From: sfeam Newsgroups: comp.graphics.apps.gnuplot Subject: Re: Fitting a function between two measurements Followup-To: comp.graphics.apps.gnuplot Date: Mon, 01 Oct 2012 20:36 -0700 Organization: gnuplot development team Lines: 37 Message-ID: References: Reply-To: sfeam@users.sourceforge.net Mime-Version: 1.0 Content-Type: text/plain; charset="ISO-8859-1" Content-Transfer-Encoding: 7Bit Injection-Date: Tue, 2 Oct 2012 03:36:01 +0000 (UTC) Injection-Info: mx04.eternal-september.org; posting-host="de68407e1e0278021e2f738a9cf7b10c"; logging-data="5696"; mail-complaints-to="abuse@eternal-september.org"; posting-account="U2FsdGVkX1/oTiuNuluQ385vL6o57TIl" User-Agent: KNode/4.4.9 Cancel-Lock: sha1:KmW49GmaD6ahSmN46PhJ0ieWr+M= Path: csiph.com!usenet.pasdenom.info!news.stben.net!border3.nntp.ams.giganews.com!border1.nntp.ams.giganews.com!nntp.giganews.com!eweka.nl!hq-usenetpeers.eweka.nl!newsfeed101.telia.com!starscream.dk.telia.net!eternal-september.org!feeder.eternal-september.org!mx04.eternal-september.org!.POSTED!not-for-mail Xref: csiph.com comp.graphics.apps.gnuplot:1395 Voodjin wrote: > Hi I have a little question and I'm not even sure if it's directly > accesable through a gnuplot function. I have two different > measurements (or better: their points) - let's call them true and > false - in one single plot. What I now want is to plot a function > which seperates the whole plot into two fields. One true-area and one > false-area. What I've done up to now is to put some points by hand > into the graph and then make a fit throught them (in the form f(x) = > (a/(x-b))-c). However, this is obviously not very scientific as my eye > isn't a very precise mathematical apparatus. So what I would like to > have now is a way to tell Gnuplot: Find a good function which > seperates both value-fields the best. Does something like this even > exist? It sounds like you are describing construction of a support vector machine (SVM). The Wikipedia page gives a good illustration. http://en.wikipedia.org/wiki/Support_vector_machine Gnuplot does not implement SVM fitting, but you might try R. Ethan > > I already thought it might be better to make a different plot with > only those values which have many overlapping values from both fields > and use them for a fit. Would this be a better attempt than the first > one with putting some fitting points by hand? > > Of course a "gnuplot, you do it"-function would be the best, but like > already mentioned I don't have any idea if something like this exist. > > I also appreciate responses which can guide me to some mathematical > background that can tell me how good my approximation seems to be. > > Thanks in advance!