Warming Up

Leif Johnson — 13 Feb 2011, 21:02

Waking can be an invigorating experience. Austin has—or at least I have—spent the weekend waking from a forced hibernation, induced by not-one-but-two recent spells of severely cold weather.

It was so cold here last week that I woke one morning to find that the water from the faucet—destined first for warming and then for tea—gushed, sputtered, and then petered out. The shower, bathroom sink, and outside faucets behaved similarly. The pipes had frozen overnight, and I spent the next three hours panicking about broken plumbing, donning long underwear, rooting the hair dryer out from the bathroom cupboard, and crawling under the house to thaw out the surprisingly unassuming pipe that brings water from the main into the house. The next day, the meyer lemon that we bought to decorate the front yard curled up its leaves in fright and froze, even though I'd covered it with an empty inverted rain barrel. (I hold out hope that it will somehow sprout some new branches this week.) In the midst of all this, my friend C arrived for a visit, and we woke the next day to find the entire neighborhood covered in snow and ice, witnessing a veritable comedy of vehicular errors as we drove downtown for breakfast.

But this weekend the spell broke, and the city stretched its collective self out in the sunlight. The Saturday farmer's market afforded parking only on the second floor of the garage at Third and San Antonio ; for that matter, parking lots and roads all across town were crowded. Austin expresses itself through car use. It's been so nice outside today that L spent part of the afternoon reading in the backyard, lounging on the cardboard that marks the soon-to-be garden beds. All through the weekend, birds, pedestrians, and bikers peeked their heads outside for the first time in days, greeted by a warming sun and a lovely tropical breeze bearing clouds from the Gulf. This evening people sat outside on porches as the bus went by.

I love—love—how Austin doesn't even venture outside when the weather is bad. Last year I went to a class that was almost 90 percent empty when it was raining outside one day : I figured most people just didn't want to face the gray skies—I know I didn't. When snow showed up on the roads on Friday, the entire university closed, and the bus system changed to run on a Sunday schedule. Even the crazy people riding bikes—who, astoundingly, and often without hats or scarves, are the last ones to head inside when it gets cold—didn't show up on campus at the end of last week. It's better here just to hunker down for a couple of days, because it will soon be sunny and warm again. This time it took a couple of weeks, actually, but the sun seems to have returned to warm us all.


The past week has been a fairly productive one for me. On the coding side of things, I have capped development on my matching pursuit code after vectorizing it for a pretty nice speedup. I can encode and decode sounds with it, and the training time is wonderfully short for short codebook vectors. The Smith-and-Lewicki version of the code needs more testing to uncover a hard-to-find indexing bug near the end of the signal being encoded, but that will come along this week. I also hacked a good bit on the driving simulator for B, getting the code to a good state where the dynamics match up well with the reinforcement learning and state quantization side of things. (It turns out to be surprisingly difficult to model a dynamic world using discrete bins.) We still need to explore the reasons behind the non-smooth Q tables for some modules, but I think it has something to do with the state quantization.


At home, L and I did a good bit more painting over the past couple of weeks. Thanks mostly to her work on the trim, we are nearly done with most of the house ! Today was so beautiful outside that I pulled up some dead bermuda grass on the once-and-future patio, and then piled up cinder blocks behind the garage in anticipation of the rain barrels back there. A house is truly a stream of things to add to the todo list, but it's also a more satisfying stream than many of the things that crop up in coding work.

I also did a lot of computer shuffling this weekend. I got a new RAID-5 array set up on a machine at the lab, and copied my data to it in preparation for setting up a second RAID-5 array on the second machine. The linux install process hit some snags this afternoon, so it will take some more system administration time to get that fixed up and working, but I hope it will go by quickly after getting some fresh eyes on it tomorrow.


This week promises to hold a lot of progress in my research work. I will make some more progress on the speech learning project, by training up a few matching pursuit dictionaries of different lengths and sizes—probably {20,50,100,200}ms time windows, times codebooks of {20,50,100} vectors. We are going to compare encoding efficiency against raw FFT to see if we're capturing anything interesting with MP. I also hope to train up some MP dictionaries on the FFT power spectra from recorded sound, to try for speech generation using those MP codebooks as a control space. If this works, there might be some interesting collaboration with J and R in the lab, to see if we can do something similar for generating motor signals.