tapas | you men the corresponding curves {x| g1(x) = 0} etc.. |

eigenburak | (i'm trying to add constraints to a linear combination of vectors, and defining my constraints too conservatively, i'm afraid, will leave me with no solution) yes, that is what i mean |

tapas | i think it depensd on the dimensionality of your space. in 20 dimensions, having two constraints might leave a 18 dimensional solution space iirc |

eigenburak | yeah, that's true unfortunately my constraint is that i want the x for which i'm solving to have x1>x2>x3>...etc |

tapas | hmm, this inequality stuff is interesting. *gets out pen and paper* |

eigenburak | yeah, the way it's used is actually very cool i'm just annoyed because i can't take advantage with my constraints, argh or, rather, my constraints shouldn't be my constraints |

tapas | :) |

eigenburak | for the lagrangian L=f(x)+lambda*g(x), if you're solving for a minimum, you just say lambda has to be positive and make it so that g(x)<=0 no wait yes |

tapas | hmm |

sexcopter | i have an integral from notes that reads http://www.mathbin.net/9659 and am really not convinced how this result is got. is there a particular technique used? Any hints or pages i can read? |

daYZman | i got a simple questino about SVMs. why is w in w.x + b = 0 is perpendicular to the separating hyperplanes. |

sexcopter | hah, just having typed it, i think the penny dropped |

tapas | daYZman: rephrase question |

eigenburak | yeah, sexcopter try defining max() has a piecewise in terms of x and a and see that penny in the bank daYZman: it's perpendicular because you define it to be perpendicular |

sexcopter | eigenburak: i was trying to picture two distinct scenarios, one where a>x and other where a<x to get two different integrals, but of course x is a variable, so as you say there are two pieces to the one integral |

eigenburak | daYZman: as in, you want it to be the vector that you use for comparison...is it class [-1, 1] daYZman: so because the hyperplanes that give max margin seperation have to be parallel, the w that intersects them to give an appropriate metric is perpendicular to both sexcopter: bingo. i have gone through the details but it seems that you're on the right track. can i borrow that penny? i think i need it for this lagrangian thing haven't* |

daYZman | umm |

eigenburak | daYZman: imagine subtracting b from all of the data points and then using w to classify each point. in other words, imagine a situation where b=0. it might be easier to see |

tapas | eigenburak: what i said was bull. staring at my drawing showed my my error :) |

eigenburak | tapas: which thing? the 20 vs 18 dimension thing? |

RadSurfer | is anyone here familiar with the Linux version of Maxima ? |

tapas | eigenburak: about the justification for transforming every constraint into the form g < 0 and making it a minimizatin prob eigenburak: it's more difficult than i thought :) RadSurfer: a little bit [very little] |

RadSurfer | Some of the rpm's are expecting a specific user, and won't install for me. I wanted to learn more about it |

daYZman | eigenburak, yep. but then, why use a plane that is perpendicular to the actual separating plane to classify? |

tapas | sounds more like a distribution prob than a maxim thing really |

RadSurfer | So try to reach someone through the Sourceforge page for that project? |