It's the little things...

[alexr_rwx]

New toothbrush! The bristles are just bristlier.

Also: the mighty Zach G yourusername came to ATLhack tonight. He brought up Fermat's Little Theorem and probabilistic tests for primality, which, to be honest, I hadn't thought about a lot, but reveal all sorts of math I'm going to have to go and learn. Mark and I tried to give him a basic overview of classification and supervised learning. Later, in our discussion of clever ways to lay out the datastructures for the Scrabblebot, we ended up finding out, from the mighty Joe and Erik and Tejus (shared journal: yourusername), that postgres lets you have multiple indexings on the same table, which somehow lets you optimize your lookups based on what you're looking up? This has to do with the fantastic world of query optimization, which I should really learn about -- if you're a database, given a particular question, how can you find a logically equivalent question that can be answered more efficiently? (rockstarling, you've spent some time thinking about this, yes?)

There is so much to learn in the world... (also: woo for invigorating ATLhacks!)

Comments

[lindseykuper]

Not much, actually! I have some other pet problems.

[oniugnip.livejournal.com]

Those are some pretty awesome pet problems to have. Useful heterogeneous data sources ahoy(?) !

[cola-fan.livejournal.com]

Database optimization can't be too far from functional program optimization, can it? Both are side-effect free, declarations of either process or set operations.

I think Tejus has moved on from yourusername to his blog at Vijedi (http://www.vijedi.net/wordpress/).

Glad to see you guys are still hacking.

[britta43.livejournal.com]

I knew there was a reason I was supposed to be at Octane...sigh. I am really forgetful. I blame it on Trinitarian theology.

[oniugnip.livejournal.com]

Yeah yeah! I mean, in the grand scheme, they're both just subsets of algorithm optimization in general, right? We want clear, concise, intuitive ways of expressing things, and we don't care how the machine ends up doing the operations, as long as it's logically equivalent.

Like... naively, multiplying some two-by-two matrices is done with some multiplications and some adds, but somehow or another, people find ways of doing it (http://en.wikipedia.org/wiki/Coppersmith%E2%80%93Winograd_algorithm) with lots of adds and fewer multiplications, instead? And the approaches they come up with are really unintuitive, but happen to calculate the same thing.

How do you find those unintuitive hacky-but-fast approaches in an automated fashion?