Comments for Justin Domke's Weblog http://justindomke.wordpress.com Thu, 10 Dec 2009 03:22:49 +0000 http://wordpress.com/ hourly 1 Comment on Automatic Differentiation: The most criminally underused tool in the potential machine learning toolbox? by John Sibert http://justindomke.wordpress.com/2009/02/17/automatic-differentiation-the-most-criminally-underused-tool-in-the-potential-machine-learning-toolbox/#comment-203 John Sibert Thu, 10 Dec 2009 03:22:49 +0000 http://justindomke.wordpress.com/?p=441#comment-203 Check out admb-project.org. ADMB is a nifty open source wrapper and 4th generation language build on C++ that makes it easy to build automatic differentiation into applications. I know the folks promoting ADMB are eager to enlarge the areas of application. Check out admb-project.org. ADMB is a nifty open source wrapper and 4th generation language build on C++ that makes it easy to build automatic differentiation into applications. I know the folks promoting ADMB are eager to enlarge the areas of application.

]]> Comment on Rushern Baker by justindomke http://justindomke.wordpress.com/rushern-baker/#comment-200 justindomke Mon, 30 Nov 2009 15:15:23 +0000 http://justindomke.wordpress.com/?page_id=518#comment-200 Rushern, If you could offer a non-illegal explanation of how your campaign obtained the email addresses of all graduate hills and gardens residents, that would go a long way in increasing your credibility for "honesty and ethical reform". When I've asked you and your staff about this in the past, I was treated with contempt, hung up on, or ignored. Of course, with your newly improved attitude, it looks like you don't care about actually answering my concerns, only avoiding embarrassment. Rushern,

If you could offer a non-illegal explanation of how your campaign obtained the email addresses of all graduate hills and gardens residents, that would go a long way in increasing your credibility for “honesty and ethical reform”. When I’ve asked you and your staff about this in the past, I was treated with contempt, hung up on, or ignored. Of course, with your newly improved attitude, it looks like you don’t care about actually answering my concerns, only avoiding embarrassment.

]]> Comment on Rushern Baker by Rushern Baker http://justindomke.wordpress.com/rushern-baker/#comment-199 Rushern Baker Mon, 30 Nov 2009 05:46:57 +0000 http://justindomke.wordpress.com/?page_id=518#comment-199 Good evening Justin I hope you are doing well, I just wanted to check and make sure you are not receiving my campaign emails. As we get closer to 2010, there will be a lot of exciting things happening with my campaign, i would love for you to be part of what we believe will be a successful campaign. As you know, honesty and ethical reform are the cornerstone of my campaign. Therefore, I want to personally invite you to become a member of Team Baker, should you decide to join Team Baker, please go to www. rushernbaker.com to sign up. Thank you and I look forward to working with you when I become County Executive. Good evening Justin
I hope you are doing well, I just wanted to check and make sure you are not receiving my campaign emails. As we get closer to 2010, there will be a lot of exciting things happening with my campaign, i would love for you to be part of what we believe will be a successful campaign. As you know, honesty and ethical reform are the cornerstone of my campaign. Therefore, I want to personally invite you to become a member of Team Baker, should you decide to join Team Baker, please go to www. rushernbaker.com to sign up. Thank you and I look forward to working with you when I become County Executive.

]]> Comment on Notation is evil by Hendrik http://justindomke.wordpress.com/2009/10/28/notation-is-evil/#comment-197 Hendrik Sat, 21 Nov 2009 22:41:11 +0000 http://justindomke.wordpress.com/?p=839#comment-197 Good points. Those would be handy. I think "incomplete" might be more appropriate than "evil" though. Just define your own notation for these and hope that it catches on. Worked for Euler. Good points. Those would be handy. I think “incomplete” might be more appropriate than “evil” though.
Just define your own notation for these and hope that it catches on. Worked for Euler.

]]> Comment on Mandelbrot in Scala by Brett England http://justindomke.wordpress.com/2008/11/29/mandelbrot-in-scala/#comment-195 Brett England Tue, 10 Nov 2009 16:41:21 +0000 http://justindomke.wordpress.com/?p=138#comment-195 The while loop can be removed if you use a tail recursive function. This is the functional approach that Scala aspires you to use: def iterate(z:Complex, c:Complex, level:Int, i:Int): (Complex,Int) = if(z.abs > 2 || i > level) (z,i) else iterate(z*z+c, c, level, i+1) ... val (z, i) = iterate(new Complex(0,0), new Complex(x,y), level, 0) out.write(if (z.abs < 2) 0 else i) ... The while loop can be removed if you use a tail recursive function. This is the functional approach that Scala aspires you to use:

def iterate(z:Complex, c:Complex, level:Int, i:Int): (Complex,Int) =
if(z.abs > 2 || i > level) (z,i) else iterate(z*z+c, c, level, i+1)

…
val (z, i) = iterate(new Complex(0,0), new Complex(x,y), level, 0)
out.write(if (z.abs < 2) 0 else i)
…

]]> Comment on Completing the square in N dimensions by david http://justindomke.wordpress.com/completing-the-square-in-n-dimensions/#comment-192 david Mon, 02 Nov 2009 23:26:10 +0000 http://justindomke.wordpress.com/?page_id=828#comment-192 I like this. Seems like all of numerics is reducing nonlinear problems to linear ones. Always wondered if there was a generalization of the quadratic equation to operators but never got this far.. I like this. Seems like all of numerics is reducing nonlinear problems to linear ones. Always wondered if there was a generalization of the quadratic equation to operators but never got this far..

]]> Comment on Mandelbrot in Scala by Ewald http://justindomke.wordpress.com/2008/11/29/mandelbrot-in-scala/#comment-191 Ewald Sun, 01 Nov 2009 10:54:41 +0000 http://justindomke.wordpress.com/?p=138#comment-191 http://wiki.cython.org/tutorials/numpy Look here for info on how to use numpy arrays effciently inside a cython routine. I mailed the code. cython -a yourfile.pyx makes an html file that highlights the slow (python calls) bits. Double click the lines to see the generated code. http://wiki.cython.org/tutorials/numpy

Look here for info on how to use numpy arrays effciently inside a cython routine. I mailed the code.

cython -a yourfile.pyx makes an html file that highlights the slow (python calls) bits. Double click the lines to see the generated code.

]]> Comment on Mandelbrot in Scala by Justin http://justindomke.wordpress.com/2008/11/29/mandelbrot-in-scala/#comment-189 Justin Sun, 01 Nov 2009 00:25:39 +0000 http://justindomke.wordpress.com/?p=138#comment-189 Very cool. Actually, since writing this post, I have basically abandoned Scala for python, so that is encouraging. (Python's libraries are 1000x better, which outweighs everything in the end) I am frankly amazed you got such good performance out of regular python! I've been using weave, rather than cython for performance, but I should probably take another look at that. I would appreciate looking at the code ((myfirstname).(mylastname).rit.edu, or just post it as a comment). I remember having a hard time figuring out how to get cython to run quickly. Very cool. Actually, since writing this post, I have basically abandoned Scala for python, so that is encouraging. (Python’s libraries are 1000x better, which outweighs everything in the end) I am frankly amazed you got such good performance out of regular python! I’ve been using weave, rather than cython for performance, but I should probably take another look at that. I would appreciate looking at the code ((myfirstname).(mylastname).rit.edu, or just post it as a comment). I remember having a hard time figuring out how to get cython to run quickly.

]]> Comment on Mandelbrot in Scala by Ewald http://justindomke.wordpress.com/2008/11/29/mandelbrot-in-scala/#comment-188 Ewald Sat, 31 Oct 2009 19:29:07 +0000 http://justindomke.wordpress.com/?p=138#comment-188 Hi, Not sure if this is relevant anymore. Just for fun I coded your second example in python and cython (the calculation bit). Using your numbers, n=2048 level=100 I get: Python Mandelbrot with n=2048 and level=100 in 58.295 seconds Cython Mandelbrot with n=2048 and level=100 in 1.218 seconds Same answer? : True Speedup: 47.8744235634 Laptop has Core 2 Duo T9300 with 3GB ram on Ubuntu 8.10 64bit. I can send you the code if you'd like. Using python/cython/scipy/numpy/matplotlib/f2py is probably a better way than more obscure (for number crunching and real world problems) languages. Cheers Hi,

Not sure if this is relevant anymore. Just for fun I coded your second example in python and cython (the calculation bit).

Using your numbers, n=2048 level=100 I get:

Python Mandelbrot with n=2048 and level=100 in 58.295 seconds
Cython Mandelbrot with n=2048 and level=100 in 1.218 seconds
Same answer? : True
Speedup: 47.8744235634

Laptop has Core 2 Duo T9300 with 3GB ram on Ubuntu 8.10 64bit. I can send you the code if you’d like. Using python/cython/scipy/numpy/matplotlib/f2py is probably a better way than more obscure (for number crunching and real world problems) languages.

Cheers

]]> Comment on Convex Functions by justindomke http://justindomke.wordpress.com/convex-functions/#comment-185 justindomke Mon, 19 Oct 2009 21:27:39 +0000 http://justindomke.wordpress.com/convex-functions/#comment-185 Well... OK. First, assume $latex f$ is convex, and let's prove $latex g$ is convex. We have that $latex g(\alpha t + \beta s)$ $latex = f(x + (\alpha t + \beta s)v)$ $latex = f(\alpha (x + t v) + \beta (x + s v))$ $latex \leq \alpha f(x + t v) + \beta (x + s v)$ $latex = \alpha g(t) + \beta g(s)$. Now, instead assume $latex g$ is convex for all choices of $latex x,v$. Then, $latex f(\alpha x + \beta y)$ $latex = f(x + \beta(y-x))$ $latex = g(\beta)$ (where $latex v=y-x$ in the definition of $latex g$). $latex \leq (1-\beta)g(0) + \beta g(1)$ $latex = \alpha f(x) + \beta f(y)$ In all honesty that wasn't as trivial as I remembered it being... Well… OK. First, assume f is convex, and let’s prove g is convex. We have that

g(\alpha t + \beta s)
= f(x + (\alpha t + \beta s)v)
= f(\alpha (x + t v) + \beta (x + s v))
\leq \alpha f(x + t v) + \beta (x + s v)
= \alpha g(t) + \beta g(s).

Now, instead assume g is convex for all choices of x,v. Then,
f(\alpha x + \beta y)
= f(x + \beta(y-x))
= g(\beta) (where v=y-x in the definition of g).
\leq (1-\beta)g(0) + \beta g(1)
= \alpha f(x) + \beta f(y)

In all honesty that wasn’t as trivial as I remembered it being…

]]>