Advogato blog for vicious

23 Jan 2008

Wed, 23 Jan 2008 06:50:14 GMT

My palm is blinking red and I hear I should go to some carousel.

13 Nov 2007

Tue, 13 Nov 2007 15:35:44 GMT

I thought people would be able to drive better in the rain in Urbana. But it is just as bad as it was in California, where it was excusable for general lack of rain. I was actually passing cars on my bike because they were going too slowly. I guess people think that driving slow, erratically and breaking at very unexpected moments is much safer in rain. So 90% of drivers are driving in a horrible way and the other 10% have road rage and weave through traffic angrily because they are pissed of at the other 90%. I actually saw a traffic jam yesterday because locals seem to be unable to move through an all way stop sign, a problem made worse by the rain.

I guess I am more pissed off about the inability of people to move on the road when it is raining and I am on a bike and thus trying to get home quickly. When bike is faster than cars and you are not in a big city where that would be expected, than something is wrong.

8 Nov 2007

Thu, 8 Nov 2007 03:38:57 GMT

Had a talk at UW-Madison yesterday. I think it went pretty well. Walter Rudin came and seemed interested as well, I think that has to count for quite a bit in the ego boost category.

18 Oct 2007

Thu, 18 Oct 2007 15:24:15 GMT

Looking at the freedesktop.org vfolder menu spec, I am getting the feeling that the original intention got totally derailed. The reason I looked at it since I needed to fix up the genius .desktop file. Anyway, the original intention was not to enforce any particular menu structure, and to allow arbitrary queries based on exhaustive set of keywords (categories). The way most menus nowdays work is that if you define two main categories which do overlap for your application, you might appear in two badly designed menus. Some arbitrarily badly picked "main categories" subset of all categories was chosen, and most menus just treat those as basically menu titles. The OnlyUnallocated query is totally screwed up now according to spec, since it doesn't make a distinction from excluded and deleted. That means that if your queries are only slightly tougher than normal, some .desktops might not appear even if OnlyUnallocated was used for some submenu.

The intention of the original implementation was that all installed apps appear unless the user intentionally turns off the icons in some menu editor. Why would I ever install something unless I wanted to run it. Any other behavior is a critical bug in my opinion.

Picking "main categories" was exactly what the menu spec was supposed to avoid. That is, you should pick the CORRECT categories, not miscategorize your application just in order for it to appear in some currently badly designed menu. The set of category keywords was supposed to be independent of the menu structure!

Example of total miscategorization: Science/Mathematics apps. There's quite a bit of them, but they will never appear in a proper category, so they are all miscategorized just so that they appear. Properly, something like Octave or Maxima should include Application;Science;Math; and should appear in "Other" or in "Science" or "Science & Math" submenu. As it is, Maxima for example, appears in "Education" in fedora.

In any case the situation is no better (and it may in fact be worse) than it was before the vfolder menu spec. Stupid arbitrary submenus are pretty much enforced and you MUST categorize your application within these in order for your app to appear. So even if a good implementation came along, it wouldn't correctly categorize your menus since the .desktop files have bogus (or not enough) information in them. At least in the way things were done before all apps were at least displayed.

Anyway, I just felt like ranting.

16 Oct 2007

Tue, 16 Oct 2007 21:09:56 GMT

For about the past week, I am finally a proper Illinoisan, having at last gotten my local drivers license. It is quite a different experience from California, where they send you everything by mail usually much later than they promise. Here we got our licenses and the plates and all on the spot. I have to admit the test and application had some logically nonsensical questions. For example there is a question that says something to the effect of "Do you drink alcohol such that it inhibits your driving ability?" Now you have to answer no to this else they won't give you the license. Some idiot didn't think the question through logically. Of course I drink to the point where it would inhibit my driving ability. The question is: do I drive after that? But what they asked didn't have this component, thus anybody but an abstinent must lie to get an Illinois drivers license. Or if you are sure that drinking does not affect your driving ability, then you can answer no to the question, and still get hammered and drive drunk without having lied on your application. Hmmm...

After several weeks of not being able to produce anything useful, I've made a whole bunch of updates to genius and made a 1.0.0 release. It's not that it's so much more featureful than 0.7.7, but hell, it's been 10 years now, there ought to be a 1.0.0 release one of these days. I know I made a pledge to not release until it can do eigenvalues/eigenvectors in general, but I'm not sure when I will implement that. There is RQI code in there now, so you can get an eigenvalue and an eqigenvector for a general matrix, but I suppose the correct solution is to hook it up to some lapack code or some such. libgsl has some matrix stuff for double precision which I suppose would be fine, but it doesn't yet do arbitrary complex matrices either and I was too lazy to hook it up. Plus I'd need to depend on fairly new libgsl ...

I've made them install genius and maxima on all the linux boxes in the UIUC math department even though most people here use Mathematica since Wolfram is here in town. Strange as it seems, there are actual genius users out there, so perhaps it is time to make their ranks larger (no pun intended).

I'm also still working on a paper on which I was working when I last wrote a blog entry. I've yet again made the theorems slightly stronger in hopes of improving exposition (doesn't that sound weird). So it's still not arxiv ready I don't think for another week or so. So far people that have seen it like it.

25 Apr 2007

Wed, 25 Apr 2007 19:37:41 GMT

Defended my thesis yesterday, so you can call me "Dr." now See the abstract and pdf.

7 Dec 2006

Thu, 7 Dec 2006 17:12:46 GMT

Here's another helping of papers: a joint paper with John D'Angelo and Han Peters on proper ball mappings (don't send any puns, I've already thought of them all!). And another one recently on extending Levi-flat hypersurfaces.

As for coding, I haven't done much since the summer when I did a few updates to genius. BTW, genius is now in Ubuntu repositories. In related news, I sort of fell in love with Maxima recently. It lacks some things I can do with genius easier, but whenever I try to seriously use genius I try adding new features and optimizations which really throws me off schedule. And genius doesn't do any real symbolic manipulations which is what I use maxima for.

Maia has been growing up (and throwing up?) nice, see the gallery then scroll to 2006 and click on one of the Maia links.

I've now sent over 90 job applications for the 2007 year for postdocs, so we'll see what turns up. Apparently 90 is on the low end for number of academic job applications one now must send to please the gods.

21 Jul 2006

Fri, 21 Jul 2006 23:03:19 GMT

So apparently being the prime minister of israel carries the same requirement as being the kaiser of prussia. One must have at least one good war. All hail kaiser Olmert!

14 Jul 2006

Fri, 14 Jul 2006 07:03:46 GMT

So I needed to use genius for some research and I realized that the code I wrote is very demanding, and further that genius was leaking memory like a sieve.

It also crashed on me, which was annoying. So I spent last two days fixing up all the leaks and improving speed in some places. My test problem for parameter equal to 3 ran in 26 seconds this morning after the leaks were plugged. Now it runs in 8 seconds. Fast you say. Well, for parameter 4 it took 4 hours. For parameter 5 it ate all my memory and I had to drop to console to kill it, since my mouse pointer was too jumpy and everything but genius was likely swapped out.

Actually it's not a problem that should require lots of memory, that needs some more coding. Still it's unlikely parameter 5 is anywhere near feasible. Maybe if I rewrite it in C, but then I'd have to write some routines for row reducing which I find kind of boring. It's also probably not worth the trouble. The code is to check that a certain conjecture is true for specific values, and I'd much rather do an actual proof. Proof by having the computer do thousands of iterations of NullSpace just to test one parameter value seems entierly unsatisfying.

It's nice again to be doing some coding. On the other hand I'm coding on a piece of software of which I'm pretty much the only user (there are apparently people using genius out there, but mostly as a calculator, not for anything really serious).

I noticed that raph was writing Ghilbert. That is a cool piece of software.

29 Jun 2006

Thu, 29 Jun 2006 17:00:21 GMT

So two years worth of nonsense summed up in 21 pages, yay! The paper was posted on arxiv on the evil day (6/6/6), thus a small mistake crept in to one of the remarks and thus I am testing to see how arxiv handles replaces of papers. June 29th doesn't seem too evil, so hopefully no other mistakes got in there.

See the paper if you want to see what I'm doing instead of coding.

Still not a parent yet, but it's getting dangerously close. So I gotta go do some research (read: sit in the local cafe with an absent expression in my face doodling on paper) when I still have time.