Disqus - Friends of momcaedmons

Re: How to: Use Python to Solve Optimization Problems

asmeurer — Thu, 02 Jul 2009 15:39:56 -0000

Hi folks. I know this post is a little old, but I wanted to note that there is no need to use regular expressions to convert input into a SymPy expression. SymPy has a built-in function sympify() that converts strings into SymPy expressions:
>>> from sympy import *
>>> sympify("x**2")
x**2

Re: Vote Now for Your Favorite Video

asmeurer — Mon, 06 Jul 2009 23:13:20 -0000

The first one is the best (the other ones are horrible. Come on guys, these are supposed to target everyone, not just people who already agree with you). The only thing is, you need to make it so that it shows the text full screen like the other ones (except the second one with the crazy spinning effect. Just do it normal). You don't even notice that it is on the bottom. Just cut the video in a few places to show the text but leave the audio.

Re: Refine module

asmeurer — Mon, 17 Aug 2009 18:25:38 -0000

It looks good, though now it needs expansion. For example, in the example you give:
>>> refine(exp(I*x*pi), Assume(x, Q.integer))
exp(pi*I*x)

This should return (-1)**x.

Will the new assumptions system let us implement assumptions like "f(x) is continuous" or "diff(f(x, y), x) == diff(g(x, y), y)"?

Re: Refine module

asmeurer — Wed, 19 Aug 2009 02:16:32 -0000

That's great. Clearly, this is better than the old system. Of course, the real work would be modifying refine to do things with those assumptions. For example, the following should return 0:

>>> refine(f(x, y).diff(x, y) - f(x, y).diff(y, x), Assume(f(x, y), Q.continuous))

And of course, once we have an interval class proper, you would assumedly be able to assign that assumption (or any assumption really) over an iterval. There could be real power in that, especially if it can gather some of those things automatically, like knowing that sin(x)/x has a discontinuity at 0.

You could probably make a whole new GSoC project out of just adding in assumptions next year :)

Re: KOB.com - Proposed school schedule change draws ire of APS parents

asmeurer — Fri, 26 Mar 2010 16:12:31 -0000

I'd say that the parents who showed up to this meeting aren't the ones who aren't caring about their children's education.

Re: ShowPass

asmeurer — Sat, 16 Oct 2010 14:26:45 -0000

Wow. That's a really stupid idea.

Re: PyCon 2011: Exhibition of Atrocity - PyCon US Videos - 2009, 2010, 2011 - blip.tv

asmeurer — Mon, 09 May 2011 01:41:13 -0000

Another option for the duck punching example if you are on Python 2.5 or greater is to use a with statement context manager. The __exit__ method will be called both if the function finishes normally and if an exception is raised.

Re: PyCon 2011: Sunday Morning Lightning Talks - PyCon US Videos - 2009, 2010, 2011 - blip.tv

asmeurer — Fri, 20 May 2011 04:36:31 -0000

Any idea how to get the ogg version in the new interface?

Re: Region Plots with Interval Arithmetic

asmeurer — Tue, 15 May 2012 23:18:07 -0000

Well this is already impressive. I was able to plot figure 4a from Tupper's paper, for example. Is it possible to plot equations, or are only inequalities supported for now?

The speed of this unfortunately is way too slow. I tried plotting 1 < sin(x)**2 + cos(x)**2 as you suggested, only using the 1024x1024 default from your script, and it took several minutes to complete. I guess using numpy or even stdlib math instead of mpmath will make this faster. We'll have to see if that makes it fast enough, or if we need to dig deeper.

Have you tried having it update the plot in real time (e.g., after each iteration of k)? I'm not proficient enough in matplotlib to figure out how to do it, but it should be possible. This won't make things faster--in fact, it will make them a little slower--but the increased response will make it feel faster.

Regardless, you should start putting this in a real branch in the SymPy repo, which will make it easier to track and test your changes.

Re: Region Plots with Interval Arithmetic

asmeurer — Sun, 20 May 2012 23:42:02 -0000

Big +1 on putting priority on merging Stefan's branch.

Re: Adaptive sampling for 2D plots

asmeurer — Sun, 27 May 2012 03:30:14 -0000

This looks nice. I tried a lot of plots, and they all look much nicer in your branch than in Stefan's. One thing I did notice though is that it is a lot slower. I guess this is another place where the ability to update the plot would be nice. You could show the initial plot and update it once you add the additional points.

Re: Growl Fork: Get Growl Notifications On Mac OS X 10.7 Lion For Free

asmeurer — Thu, 31 May 2012 18:23:11 -0000

Growl is open source, so there's probably no difference. Someone just compiled Growl and shipped it. You can do the same yourself if you have the know-how (it's a little tricky, though, because you have to create your own developer certificate, but the Growl developers helped me do it when I did it).

Re: GSoC Week 3

asmeurer — Tue, 12 Jun 2012 16:46:10 -0000

This is indeed great. Do you know of any ways to avoid the error problem, other than simply making the interval size less than one pixel?

So if implicit equations aren't supported, I take it you plotted y**2 = x**3 - x as two separate equations.

What I'd like to see, perhaps in a future post, is how this compares to just regular plotting for explicit equations, both what the advantages and what the disadvantages are. Also, maybe you could code up a simple naive implicit plotter to compare against.

Re: GSoC Week 6

asmeurer — Wed, 04 Jul 2012 01:57:15 -0000

How did you determine that the Mac OS X grapher uses this algorithm?

Re: GSoC Week 6

asmeurer — Wed, 04 Jul 2012 02:00:23 -0000

It looks like they use a different algorithm (maybe I'm wrong, though). I did see one thing there that could be useful: to get plots with complex values, plot log(expr), and it will go complex where expr is negative. That is a useful way to test the complex value problem.

Re: GSoC Week 6

asmeurer — Wed, 04 Jul 2012 04:01:03 -0000

I'm not sure if any code is shared between these two products. Grapher is made by Apple, and Graphing Calculator was a light version of a proprietary (non-Apple) software that they shipped with Mac OS 9.

Re: Your photos, simplified (Part III)

asmeurer — Sat, 14 Jul 2012 23:24:25 -0000

Is there a way to tell how much of the 3GB I've received?

Re: OS X Mountain Lion (10.8) Is Not Good News For Developers

asmeurer — Sun, 19 Aug 2012 22:02:48 -0000

Google Chrome Canary uses notification center, and I definitely didn't download it through the App Store.

Re: Tip: How to sort folders by size with one command line in Linux

asmeurer — Sun, 28 Oct 2012 01:41:12 -0000

On non-Linux, you'll need to use GNU sort to get the -h option. This is part of the GNU coreutils package, and might be called gsort on your system.

Re: Unification in SymPy Enabling logical programming in Python

asmeurer — Thu, 01 Nov 2012 14:14:12 -0000

Will it be possible to extend unify by telling it about external universal identities that it can use to try to find a match? For example, your first example can also be matched in two different ways because the transpose of the transpose is the original matrix. Of course, here we will get infinitely many matches, because we could use the transpose, or the transpose of the transpose, or the transpose of the transpose of the transpose, ...

Re: Unification in SymPy Enabling logical programming in Python

asmeurer — Fri, 02 Nov 2012 02:45:45 -0000

Re: Branching Strategies handling divergent rules

asmeurer — Sat, 10 Nov 2012 20:51:31 -0000

I think the Python version of your toy problem is missing the otherwise condition.

Re: Preliminary BLAS Results putting it all together

asmeurer — Sat, 10 Nov 2012 21:02:14 -0000

"How can we ensure that the best solution is in the first few?"

I take it this is an open question, since you didn't seem to answer it. I guess you should also make a way to write down rules for what methods might be better. For example, you note that POSV is better than GESV. This will lead to a partial ordering on set of possibilities, from which you can create a smaller set which is optimal with respect to that ordering. Adding more ordering rules will make this set smaller (alternately, providing more input assumptions will make more rules apply). Note that you'll have to order the rules themselves, so that you can apply them lexicographically (or else they may not combine to a true partial order).

That will work if the yielded set is finite. If it is infinite, or combinatorially very large, you'll need to modify the yielding algorithm, itself so that it uses those orderings. I don't know if a general solution is possible in this case (and hence, if the code can truly be separated like you like).

You could also play around with more fancy stuff, like fuzzy logic or some AI/machine learning algorithms to pick the best possibility. This will be a good idea only if you really do end up getting a bunch of possibilities, even after making obvious eliminations using the ordering idea, or if all but the most trivial cases lead to infinite or combinatorially large possibilities.

Re: Preliminary BLAS Results putting it all together

asmeurer — Sat, 10 Nov 2012 21:05:27 -0000

By the way, Disqus lets me subscribe to each of your posts by email. Do you know if it is also possible to subscribe to new posts, so that I will get an email each time you post something new?

Re: Preliminary BLAS Results putting it all together

asmeurer — Sun, 11 Nov 2012 03:28:05 -0000

Yeah, Mountain Lion dropped RSS support in Safari, and I've been too lazy to setup Google Reader (plus I'm not too high on the interface). So I've decided to just move to email updates for blogs I care about. Actually, a nice thing would be email updates from the planet.