"Smells like fish," Scheme said.
Another splotch of ink.
"You're writing this down, right Hu?"
"My observations. What I eat and drink. Smells. You're supposed to be taking notes."
Oh. Right. (Another splotch.) Wait, why?
"You never see the pattern as it's happening. Once, I kept track of everything I ate for six months and found out I had Thai food every nine days, like clockwork."
I started retroactively building a database. Although it only had one entry: an espresso at the Black Danube --
"And I want to correlate everything. When do I have the best ideas? When am I clever, and conversely, when can't I form a complete sentence? What have I been eating, drinking, absorbing? How much have I been sleeping?"
-- Annabel Scheme, by Robin Sloan.
HUGIN-19.LG.GRAILGRID.NET (a.k.a. Hu) is the narrator of Robin Sloan's Annabel Scheme. In the excerpt above, we see the kinds of things that Hu, an AI Watson to Scheme's Holmes, is capable of doing and recording.
I am making Hu. A rudimentary version of Hu, sure, but Hu nonetheless. Every ten minutes, I want my computer to take a snapshot of what's going on -- what time it is, what software I have open, what websites I'm looking at, how many unread emails I have, what music I've been listening to, which files are open, how long/big those files are (to track my work's progress), what the weather's like, what I've eaten (via services like Daytum), etc. From there, I can look at trends, graph things, measure things, and try to spot patterns.
What's more, Hu has a plugin architecture -- after all:
I can interface with anything! I'm infinitely extensible...
Want Hu to record how many items are in your trash can? Write a little bit of code that checks how many items are in the trash can and tell Hu you want that data recorded. Ditto for recording your weight. Or your computer's remaining battery life. Or how many events are on your calendar for today ("do I work better when I've got long stretches of time to work in, or do I get my best work done in little chunks of time?").
This is all wishlist stuff, though. Don't get your hopes up too high just yet.
Hu is a little bit delicate at the moment.
To use Hu, first download the source. Then, change your values for the various plugins (your last.fm username for the
hu_lastfm plugin, your location for the
hu_googleweather plugin, etc.). Finally, run
$ python hu.py. That will run Hu once. Hu's notes are stored in
Hu has a plugin architecture. When Hu runs, the first thing that happens is that he looks for every folder with a name that starts with
"hu_" and treats that folder as a plugin.
Here's how you write a plugin:
- You come up with a name for your plugin (with plugins, as with bands, coming up with a name is the hardest part). We'll call it
$PLUGIN. When you see
$PLUGINfrom here on out, substitute your own plugin's name. Nota bene: your plugin's name must start with "hu_" for the reason outlined just above.
- You write a python script called
$PLUGIN.py that contains a definition of a function called
getData(). You can include other functions, classes or any other kind of code, but you must define a function called
getData()returns exactly what is recorded in Hu's entry, so it should be XML, complete with newlines (no newline necessary at the end). No tabs/spaces are necessary, since that's carried out from within
$PLUGIN.py inside a folder called
$PLUGINfolder lives inside the
- Inside the
$PLUGINfolder, have a file called
__init__.py, which contains a single line of code:
That's all there is to it. If you're having any issues, look at how the default plugins (
hu_openapps, etc.) work.
Once Hu has a bunch of information about you, here are the kinds of things you can start to ask:
- What application do I spend the most time in? Find the most commonly occurring frontmost application.
- What websites do I visit the most often? Get the most frequently-appearing domain.
- How much of what I do at the Coffee Shop With the WiFi is work, and how much is refreshing Hacker News?
- How does the weather affect the kind of music I listen to? Compare the list of songs you listened to when the weather was "Cloudy" to the songs you listened to when the weather was "Fine."
- What time of day do I listen to music most at? See how many songs I listened to between 6pm and 12am and compare that number with how many songs I listened to between 9am and 3pm.
At the moment, you have to ask those questions at the command-line using your
grep fu. I plan on making searching and querying Hu a bit easier later on, but until then, you'll have to speak bleep-blorp like a robot.