What I Learned Today

I recently came across Steve Yegge’s emacs tips. He’s a man after my own heart:

“Using the mouse is almost always the worst possible violation of economy of motion, because you have to pick your hand up and fumble around with it. The mouse is a clumsy instrument, and Emacs gurus consider it a cache miss when they have to resort to using it.”

Cache miss! I love it.

His article is long, detailed, opinionated, and a goldmine of useful emacs nuggets. It will only be interesting to you if you are at least an intermediate emacs user and, say, you agree with the philosophy noted above. Here I include some of my own tips and commentary on his post.

Steve suggests mapping the CapsLock key to be Control, which is sensible in an emacs world, since you use Control orders of magnitude more often than Caps Lock, and ergonomically CapsLock has the more desirable position. But I’ve always been reluctant to do this, because it’s non-standard, and once you’ve adapted to the unusual layout, you render yourself frustrated and mistake-ridden when using anyone else’s keyboard.

Navigation: I already use Ctrl-s (search forward) and Ctrl-r (search backward) liberally to move around in documents (I agree that this is very handy!) and I do use temp buffers for notes — or rather, only one temp buffer. I only use *scratch* for this, because I like being forcibly reminded that there is no auto-save going on and anything I put there won’t be saved unless I explicitly do so (otherwise my compulsively periodic saving goes on as a background processes, in addition to emacs’s autosave, and I generally assume that anything in an emacs buffer has a good dose of permanence). However, I’ve rarely used the regexp search commands, because it was so tedious to type Meta-x-isearch-forward-regexp. Happily, in this article I learned that there are bindings to be had: Ctrl-Alt-s and Ctrl-Alt-r, respectively. (Actually, on my Mac they are Ctrl-Apple-s and Ctrl-Apple-r.) I agree that this is a spectacularly awkward binding to type, but I don’t use it that often (yet) so I’m willing to try it out before remapping it.

The other super-useful navigation bindings that I regularly use (not mentioned in his article) are Ctrl-[ and Ctrl-] to jump to the top and bottom of the file, respectively.

Buffers and windows: I already use most of the buffer and window management commands he listed (splitting, swapping, closing, and listing buffers) but did learn a new one: Ctrl-x + to balance window sizes. Cool!

GUI stuff: I don’t use the menu bar in emacs (sometimes it’s there, sometimes not; I generally ignore it either way). I do, however, like having a scroll bar. I don’t use it much (again, I dislike using the mouse), but it gives a nice visual indication of where I am in a file and how big it is. Yes, the status bar will tell me that I’m 61% through the file, but the visual is much more intuitive, especially because it shows me how much of the file, proportionally, is currently shown in the buffer window. Steve backs off from his initial recommendation to get rid of the scroll bar with an interesting analog/digital argument.

Steve cites “region selection” as a case where the mouse is actually helpful in emacs. I’ll agree, in the case of rectangular region selection (I don’t even know how to do that with the keyboard, much less efficiently), but I do normal region selection all the time with the keyboard. I set the mark (Ctrl-space), move to the end, and then do whatever I needed to do with the region: copy (Meta-w) or cut (Ctrl-w), usually. He seems to view the “move to the end” as the slow bit, but with Ctrl-s (search forward), as noted above, generally this is speedy as well.

Help: I agree that Meta-x-describe-bindings can be a fascinating (and revealing) exploration of the current mode and the powers available to you. I’ve used this in bibtex mode to good effect.

Query-replace: This is one of my favorite bits of emacs power. I love that I can so easily specify text to find and replace, and do it interactively, skipping from instance to instance and individually saying yes or no, or opting to just say yes to all such occurrences (with a !).

Other tidbits I gleaned from this article are

• Meta-b to skip back a whole word. I can’t believe I never knew this one. I imagine then that Meta-f goes forward a whole word. Hey, it does! I also use Ctrl-t regularly to swap two characters (usually when I typed them in the wrong order), and sure enough… Meta-t swaps two words. Whee!
• M-x list-matching-lines: this one blew my mind. It takes in a regexp and shows you every line in the buffer that matches it (like grep -n -e on the command line, but hey, you’re still in emacs, and if you go to one of those entries and hit return, your cursor jumps to that position in the buffer).

Any additional tips are welcome!

The Dock in Mac OS X is a quick-launcher for your favorite applications, but it seems to encompass more than that; it also provides the ability to minimize and maximize windows as well as to quickly swap between running apps with a keystroke (apple-tab), which may well be my most frequently executed keystroke, second only to ctrl-K in emacs. But sometimes, the Dock hangs. Suddenly, you’re paralyzed, only able to work within the current app (and possibly only its currently window). Ugh. Rebooting does solve this, but that seems like a big hammer. I’ve looked for a “restart the Dock” option under the Apple menu (nope) or for Dock entry on the list of apps you can “Force Quit” (and hopefully Force Restart).. nope.

The most recent time this happened, I was left with only Firefox working, so I was able to google while in a Dockless state. I came across this tip, which gives the so-obvious-in-retrospect Unix-flavor solution:

1. Find “Dock” in your current list of processes (ps -auxww | grep Dock).
2. Get its process id.
3. Send it a hangup signal to force a restart (kill -HUP <pid>).

Done! (A single-line AppleScript solution is also mentioned. Someday I may venture into the world of AppleScripting. As of now, it is a complete unknown to me.)

ImageMagick’s convert utility is the Swiss-Army knife of command-line image editing. I won’t get into all of the amazing tricks it can accomplish (you can even make animations and anaglyphs), but instead share a cool new thing I learned to do with it today. This command will generate a grayscale image that contains only the values from the “channel” you specify (R, G, B):

convert colorimage.jpg -channel R -separate grayimage-red.jpg
convert colorimage.jpg -channel G -separate grayimage-green.jpg
convert colorimage.jpg -channel B -separate grayimage-blue.jpg

So for example, you can take this good-looking guy:

Red	Green	Blue

You may be surprised to learn that this has direct bearing on my Geology Master’s thesis. I’m analyzing grayscale images of layered rocks to determine if they contain evidence of past life—and the way you convert from color to grayscale may matter (we’re going to test this).

Lifehacker suggests the following command to list your most frequently used commands:

history|awk '{print $2}'|awk 'BEGIN {FS="|"} {print $1}'|sort|uniq -c|sort -r

The idea is to identify your frequent commands and then perhaps alias them to something shorter. Unix being what it is, of course, you’re unlikely to save many keystrokes unless you’re aliasing not just the command but also its arguments. Anyway, on my Mac, I get the following top 10 list:

• ls: what’s here?
• cd: go somewhere else.
• open: Mac-specific; spawns a Finder window if the argument is a directory or spawns the appropriate viewer/editor for a graphics file, Word document, PDF file, etc. I love this command.
• ssh: connect to another machine (sorry, elysium!).
• mv: move something to a better place.
• latex: there can be only one typesetter.
• less: less is more. No, wait, less isn’t more. Never mind; just display the contents of a text file.
• df: how much disk space is left? (I was getting close to the limit of my 100 GB!)
• convert: magical utility that turns any graphics format into any other one.

On my linux machine at work, the same exercise turned up slightly different patterns:

• ls: or maybe it was the same pattern.
• cd: like I said…
• jobs: oh good, something different! I run a lot of jobs on pavonis and often need help remembering what’s still running in which shell.
• fg: foreground a job, leaving the others to languish in the darkness.
• screen: oh, the joys of a detachable screen! I want to learn to wield this magic even more.
• less: less is still more! Or maybe it isn’t.
• cut: I love cutting and pasting in the shell.
• bg: dismiss this job into the background wastelands, where it may toil in obscurity!
• nice: try not to crowd out other users on the same machine.
• top: find out who is crowding me out on my own machine.

I don’t see myself aliasing any of these in the near future, but it’s a fun self-data-mining exercise. How do you use your shell?

I recently attended the annual Conference on Artificial Intelligence, held this year in Chicago, IL. I took away several new ideas and thoughts. Here are some highlights:

• Alyosha Efros gave his invited talk on a variety of cool things you can do with a single image–from inferring 3D geometry to automatically filling in occluded parts to estimating the location where the image was taken, all of his work is exciting even if you’re not a computer vision researcher. (Bonus: how to digitally “shave”)
• Stuart Russell noted that the graduate Machine Learning course at Berkeley has the largest enrollment of any grad course in the entire university; enrollment has reached as high as 140 students. In his invited talk, he issued a call to arms motivating AI researchers to tackle bigger research problems. “Fruit flies can recognize digits!” he noted. He urged the development of agents with probabilistic first-order logic, that track their internal state, and have the ability to abstract both behavior and their lookahead abilities. Ambitious, of course!
• Two presentations discussed methods for training a learning algorithm with zero data. No, really! Both pointed out problems in which the class labels or identities themselves can be represented in feature space, so you don’t really need to further train on real data. This is pretty obvious–effectively, it’s for problems where you already have a class prototype or representative, which is one way to represent a learned model anyway–but I think the papers are still useful in that they cause us to stop and think about what we can leverage a priori, instead of just hunting for more and more training data.
• Explicit Semantic Analysis (ESA) is an interesting text analysis technique that is useful when analyzing short text snippets, like keywords or web search queries. There isn’t much content in a two-word phrase, but ESA can remedy that by searching for the phrase in wikipedia and then expanding the “semantic content” (or representation) of the phrase using the titles of all wikipedia articles that contain it. Pretty clever!

I also co-chaired the Colloquium on AI Education, which was very well attended, energizing, and fun. I may even talk myself into doing some teaching again this year. :)