Stephen: computer related

Blog-style comments

These are mainly quick little tips about Vim, latex and Matlab, since I use all three of these on a daily basis.

June 2 2011: a professional proof-reader just proofed my thesis, and I made many recurring mistakes that I thought I would share since I was not aware of all of them.
- Quotes. When you quote a sentence, like "I have a dream," then the punctuation ("," or ".") goes before the final quotation mark. For example: "Let them eat cake." But what I did not know is that when you quote a word to emphasize it, like We compare our method to an "oracle" method. The "oracle", which ... , then the punctuation comes after the final quotation mark, so it is "oracle", and not "oracle,".
- Most of the corrections were due to hyphens. There are three types of hypens ("-", "--", and "---" in latex). The "--" is easy because it's just used for numbers, like 45--50 (and do not put spaces in between the dashes and numbers). The "---" is also easy because it is used to break up a sentence. When using "---", there should be no spaces between the words. For example, I like candy---because it is sweet---and therefore I have cavities.
  
  The trickiest is the "-", because sometimes you have no space, sometimes you have "-", and sometimes you use a space; e.g. eyewitness, eye shadow, eye-opener. A good reference is this grammar page. Other times, it depends on how the compound word is used. An example (based on the grammar page above), "The well-known actress won the award because she was so well known." Confused? You should be. The grammar page says that some words just have to be looked up in a dictionary because there is no reliable rule to tell when to hyphenate.
  Some common mistakes I made:
  - "tradeoff" should be "trade off"
  - "lowpass" should be "low-pass"
  - "non-zero" should be "nonzero"
  - "large scale" should be "large-scale"
  - "nonnegative" should be "non-negative" (and "nonlinearity" should be "nonlinearity")
  - "The inequality constrained version..." should be "The inequality-constrained version..." (but I assume "This version is inequality constrained" is correct).
  - "single pole integrator" should be "single-pole integrator" (but I assume "The integrator has a single pole" is correct).
  - "nonsmooth function" should be "non-smooth"
  - "stepsize" should be "step size"
  - "least squares" should be "least-squares"
  - "subproblem" should be "sub-problem"
  - "large scale problems" should be "large-scale problems"
  - "total-variation" should be "total variation"
  - "second order cone" should be "second-order" cone
- "e.g." and "i.e." are always followed by commas. In latex, it should look like i.e.,~the phrase goes here. There is no need to italicize them, since they are now de facto English words. Similarly, there is always a comma before "respectively", as in Methods A and B are faster and slower, respectively, than method C.
- Figure captions. When do you end a figure caption with a period? All sentences except the last one should end in a period (duh) so that you can distinguish sentences. But the last sentence sometimes should not end in a sentences. The general rule is that if the last sentences is not a complete sentence, then it should not end in a period. For example, if the last sentence is Left: comparisons, using all the methods then do not end it with a period. But if the caption is Left: comparisons, using all the methods. We noticed that our method was the best. then you do use a period at the end.
- Section titles: do not capitalize words in the section/chapter titles unless they are the first word or a proper name. Similarly, it is the "Fourier transform", not the "Fourier Transform".
- The Caltech proof reader likes commas for all items of a list, as in I like apples, bananas, and oranges (as opposed to I like apples, bananas and oranges). I was taught that both ways are acceptable and merely a stylistic choice, but I'm mentioning what the Caltech proof reader likes so that other Caltech thesis writers can save themselves some corrections.
- Do not capitalize names even when you want to emphasize the abbreviation. Thus Magnetic Resonance Imaging (MRI) is incorrect, and it should be Magnetic resonance imaging (MRI).
- For the bibliography, if you're unsure what the standard abbreviations for journals are, you can check out (or even use the macros provided) IEEEabrv.bib which is available at ctan.org's IEEEtran directory. If you include this bib file, then, for example, the string IEEE_J_IT will expand to "{IEEE} Trans. Inf. Theory" in the bib file.
October 30 2010: check out Detexify, a LaTeX symbol classifier. You draw a symbol, and it tries to match it with an existing symbol and show you the LaTeX code. Cool! It worked fine to find \succsim (which I can never remember).

October 2010: comments about Matlab's Just-In-Time compiler. For an introduction, see Mathworks note about JIT and profiler, another Mathworks JIT document (PDF), and this undocumentedmatlab.com post about JIT options.
I want to point out how important understand the JIT is, because JIT accelerated code is much, much faster (in certain situations). This is less of an issue with things that are vectorized, or already call optimized libraries (e.g. the fft). Below is some simple code, with 3 other variants. The table below that shows the various running times that I've recorded over the years on different computers.

JIT

JIT, line A

JIT, line B

JIT, no JIT


N = 1000000;
a = randn(N,1);
b = randn(N,1);
tic
s = 0;
for k = 1:N
    s = s + a(k)*b(k);
end
toc


N = 1000000;
a = randn(N,1);
b = randn(N,1);
tic
s = 0;
for k = 1:N
    s = s + a(k)*b(k);
end
toc
k = 19;     % line "A"


N = 1000000;
a = randn(N,1);
b = randn(N,1);
tic
s = 0;
for k = 1:N
    s = s + a(k)*b(k);
end
toc

k = [4,5];   % line "B"

feature accel off
N = 1000000;
a = randn(N,1);
b = randn(N,1);
tic
s = 0;
for k = 1:N
    s = s + a(k)*b(k);
end
toc


feature accel on

Here are the running times on various computers, OS, and Matlab versions. I can run the code in one of two ways: either with the cell mode (e.g. cntrl + enter), or execute the whole file (e.g. F5).

variant	R2006a, Windows XP, Pentium 4 3 GHz	R2008a, Windows XP, Pentium 4 3 GHz	R2006a, Redhat RHEL4, Pentium 4 3 GHz	R2007a, Debian, Pentium 4 3 GHz	R2010a, Redhat RHEL5, Pentium i7 2.7 GHz
normal, cell mode	2.04 s	0.04 s	1.73 s	1.80 s	0.87 s
normal, F5	.019 s	.014 s	.011 s	.011 s	.007 s
line "A", F5	.042 s	.0435 s	.038 s	.018 s	.038 s
line "B", F5	2.447 s	1.87 s	2.90 s	1.90 s	1.03 s
no JIT, cell mode	NA	NA	1.76 s	1.79 s	0.81 s
no JIT, F5	NA	NA	2.20 s	2.40 s	1.30 s

Notice how lines A and B (and especially line B) ruin the performance. Also, when JIT is off, cell mode is actually faster than running the script normally -- very interesting, and unintuitive.

Oct 2010: I use Vim to write my latex documents, but haven't used the Vim spellchecker until now. However, the spell checker checks for misspelled words in the wrong place: it checks comments, not the source. If I turn on spell checking, the default is this:
After my fix (see below), it looks like this:
To do this automatically, on Unix/Linux, edit the file ~/.vim/after/syntax (not all of those directories exist by default, so you may need to make them), and make a file tex.vim containing the following:
```
set spell
syn spell toplevel
setlocal spell spelllang=en_us
syn cluster texCommentGroup contains=texTodo,@NoSpell
hi def link texBlockComment      texComment
```
and that's it. (to get syntax highlight in html, see this blog comment).
Oct 2010: Matlab issues with RHEL. Our department computers run Red Hat Linux, and there have been some weird issues with recent Matlab versions (R2009a and later) that Mathworks has been unable to solve. The basic problem is that when the window focus changes, sometimes the keyboard doesn't work. You can generally eventually get it to work by clicking on a menu or changing focus several times, but it's very annoying. The simplest way to reproduce the error is by typing figure from the Matlab command, and then after the figure is open, change focus back to the editor (but don't close the figure). If you can still type, type close. After this, if your system has the bug, you shouldn't be able to type.

The bug has to do with javascript, but the Mathworks people and my system administrator had no luck fixing it. However, I found a workaround that is a bit weird but so far has been foolproof. I noticed that if you ssh to your own computer, and then launch Matlab, there's no bug. So, change your icons / shortcuts so that instead of just launching Matlab, they call it like this:
ssh -f 127.0.0.1 /opt/matlab/bin/matlab -desktop
The 127.0.0.1 is just localhost (that is, your own computer). Of course, change the path to Matlab if you've installed somewhere other than /opt/matlab/bin. I use an ssh keyserver so this doesn't require me to enter a password, but if you don't use that kind of system, then you may need to enter a password or reconfigure it.
Sep 2010: quick Vim hack. Suppose you're editing a text file and you want to do some syntax highlighting. For example, it's a "to do" list, and you want to highlight every occurrence of "DONE". Then do something like
:syn keyword Statement DONE
or
:syn keyword Todo DONE
and "DONE" will be highlighted as if it is a Statement or Todo (which have default highlighting). There are many other options instead of Statement...
Aug 2010: I use ssh key agents. On Linux, it's easy to enter your passphrase once when the desktop manager loads, and from then on you have your key in memory. But on Windows, although I've used pageant for a long time, I didn't know how to have it load automatically. Finally, a few months ago, I figured it out. First, you want pageant to load the key automatically (as opposed to having to specify which key to use). Then you put this command in your "Startup" menu. For my account on Windows 7, it looks like this:
"C:\Program Files (x86)\PuTTY\pageant.exe" "C:\Users\Stephen\My Documents\id_rsa.ppk"
Now, when I login to Windows, it asks me once for my passphrase and then remembers it until I logout. Several programs, like Xwin32 and Filezilla, work very well with pageant, so I don't need to enter any passwords.
July 2010: I use dropbox a lot, especially for collaboration with coauthors. Because dropbox will alert other users whenever you change a file, this means that everyone else will be bothered if you just edit a file with Vim, since Vim makes a temporary swap file in the directory.
You can fix this by telling Vim to put these swap files in a single location (e.g. I made a directory called ~/.backup on my Linux laptop). Then, in your Vimrc file, add a link like
set directory=~/.backup//
The double trailing slashes are important, because they tell Vim to mangle the name with a unique path identifier. This means that if you have two files of the same name but in different directories, you'll have unique swap files for each. More info at winterdom.com You can do the same thing for MATLAB temp files, but it doesn't work as well because MATLAB won't mangle the name, so you can have collisions if you have many files with the same name.
July 2010: I got a new laptop two months ago and have been running it mainly with Ubuntu 10.04 LTS. The version of gnome (2.30) has a really cool desktop background effect: it allows you to play a "slideshow." This means it cycles through a group of pictures. Incidentally, Windows 7 can also do this. For gnome, it's rather laborious to make your own slideshow, so I've written a little script to make it much easier.
You can download the makeBackgroundXML.sh script. To use it, go to anywhere you have background pictures; the default location is /usr/share/backgrounds. I then make a subfolder and put whatever background pictures I want in the slideshow into this folder. Run the script (you may need root privileges) in this subfolder, and it makes out.xml, which is what gnome needs to run the slideshow.
To add the slideshow, first open nautilus to the directory that contains the new xml file, then right-click on the desktop, select "Change Desktop Background", and drag the xml file to the window. Voila! If you want to change the timing, you can edit the script and make the obvious changes, and then re-run the script (no need to drag the xml file to the Desktop Background window again, for it will update automatically).
To see someone else's approach, checkout this "Create Your Own GNOME Background Slideshow" website.
Feb 2010: I put some effort into getting Vim's syntax highlighting to work as desired. Below are some files that are run in addition to the usual syntax highlighting files. To get them to work this way, place them in the special directory $VIMRUNTIME/after/syntax (you may need to make this directory), where $VIMRUNTIME is often something like ~/.vim; the original syntax highlighting files, which you probably shouldn't touch (and may not even have write access to) are in $VIM/syntax, and you can use these as examples. Vim is pretty smart, and if you edit a C language file, it will look for syntax files called c.vim. So, some useful files:
- matlab.vim When editing a MATLAB source file, this lets Vim recognize block comments (these comments start with %{ and end with %}, much like /* and */ in C ).
- c.vim This adds support for highlighting keywords from the pthreads and MPI libraries. You can use it as a template for whenever you are using a new API frequently. I find that syntax highlighting greatly reduces the number of typos in my code (and if you find a typo in this file, please let me know).
- cpp.vim Same thing, but for C++ instead of C (there's not much difference).

Stephen: computer related

Blog-style comments

Other links