Crunching galaxies at Kaggle
The guys at Galaxy Zoo are working on understanding galaxy formation, so they've started a kaggle challenge. It involves some clever feature extraction from astronomical photographs, and a fair bit of learning. The goal? Classify the galaxy's morphology as well as humans do.
"Sara, remind me to get some sleep when I get home". That's a geofence around your home allright, and if you are going to handle that server-side, for all the users? Well, R-Tree's will get you out of the pickle. But what if I wanted to figure out what friends come close to me? That puts a geofence around my position, meaning not only do you repeat spatial queries, you change the query every time! In off-the-shelf R-Tree's this grinds to a halt, since concurrent processes cannot update the tree simultaneously.
So, that's what I've been doing lately: designing data structures that overcome this problems. We are currently at close to linear scalability and hundreds of thousands of queries per second on a single machine!
Siafu, the GPL Context Simulator, is out!
Yes, it's taken a while, but I've finally pushed Siafu into GPL! Go ahead, check it out here.
What am I talking about? Ah well, from the website then:
Siafu is a versatile, large-scale Context Simulator written in Java. Download it, create or modify a scenario and fire it up. Your simulated world includes models for agents, places and the context therein. By tweaking these models, you influence the scenario, and can collect the context data for any of your agents. Use it for visualization, data set generation, testing or demonstration. Whatever you do with it, have fun and, if you like it, spread it!