July 24: A Presentation Approaches!
This morning I got up and had dorm breakfast of stewed lotus root and konnyaku and a ground beef omlette (with rice and miso soup). I got out of the dorm on time and caught my usual bus. This time, however, on the ride to work I saw some rather spectacular English on an it-would-be-generous-to-call-her-middle-aged lady. It was blue, with a circle of darker blue stars, with the top of the circle replaced by letters saying "LIVE FOR SPEED", and then, also in dark blue, cursive print swooping across the lower right of circle asking the rather appropriate question "Why the hell?". Not "Engrish" per se, as everything was pretty much correctly formed, but it was just so... random and, like I said, was being worn by a lady who was probably 70+. "Why the hell?" indeed. O_O I see a lot of "English" on clothing here, so I'm getting kinda blind to it, but this was just too spectacular, and I almost guffawed out loud on the bus.
At work I talked to Kabutoya-san about my presentation, and he had a bunch of suggestions that were relatively minor changes, but did make quite a bit of sense. My presentation is easily separable into 3 parts, the first two of which are very, very similar, so Kabutoya-san gave me feedback on the first part so I would have something to do while he looked over the second two parts. When he came back with feedback on the second and third parts, we came to the lovely discovery (two days before my presentation) that he had never actually communicated to me that I was supposed to benchmark LDA as well as my multinomial mixture. D: Yikes! Oh well, the LDA code was his anyways, so all I had to do was write a Ruby test suite, which took me less than half an hour, as I made a decent test for my multinomial mixture that has easily modifiable code and I just worked from that. I just hope that the results of my LDA (which runs slower than my mixture model due to increased model complexity) test will be ready to put in my presentation by the time I have to give it haha. >_<
At lunch I ate hiyashi-chuuka, which was lovely. I'm still a bit wierded out by the sweetness of the sauce, which is something we definitely do not subscribe to at home. Oddly, people seem to put wasabi in the hiyashi-wakame-soba, but it looks like they put MUSTARD in hiyashi-chuuka! This is most heretical, and I am going to have to stage a surgical strike operation against the mustard supply to the office. ;D
I played handiball at lunch, and my shoulder felt like it was juuuust about ready to get better, so I really didn't strain it at all. Hopefully I can get back to normal play, as my teammates don't seem to understand that ONLY serving seems to be a problem, and have stopped passing to me as much. Either that or they just mysteriously hate me now... D:
After lunch was more presentation stuff. I'm kind of glad that this presentation is wrapping up, because I'm sounding a bit like a broken record in the blog these days. >_< I actually have mostly wrapped up the changes Kabutoya-san asked for, and will submit my slides to him again tomorrow morning. In the meantime he's asked me to try to understand the Variational Inference method detailed in the actual LDA paper (we used Collapsed Gibbs Sampling as an alternative to this). Apparently Kabutoya-san is not familiar with this Variational Inference business, so if I can grasp it, I think that will be a big win for everyone. The trouble is that with this kind of stuff, it's always a question of "how deep?". I mean, the update equations are right there on the page, so actually using it would probably be quite easy as they've laid it out. I also think I understand the process of how to derive said equations as is detailed in the paper, but to actually do it is faaar beyond my ability at the moment. It appears to require calculating conditional expectations, using Jensen's inequality (an old nemesis haha) and using the Kullback-Leibler divergence, all of which are beyond my reach at present. I will certainly try though, because actually doing something honestly useful is sounding very attractive after a month and a half of work haha.
Kabutoya-san actually left before me today, which I guess explains a bit about my humourously easy hours, as the only people who are there when I arrive and leave consistently are the team lead, the deputy team lead and Kawanaka-san, who sits next to me, and is surely given extra hours as punishment for using spray-on deodorant! D:< I actually don't mind too much, as it's mild and does not resemble Axe in the slightest, but after 3 years of high school PE, I have a very, very strong prejudice against spray-on deodorants. Kawanaka-san's product does win brownie points for having the rather humourous name of "BODY SPLASH BY SEXY BOY". >_< Sounds illegal in most countries to me...
As a result of finishing my slides off and Kabutoya-san not being there, I left work "early", at about 6:10. I find that I'm getting better and being operational while hungry, but reading papers while hungry when I'm not even technically supposed to be there still is not something I yet choose to indulge in. My walk home was pleasant, although I am quite sad that I do not have "Groovin' Magic" available to put on my music player, though I suppose I *could* fix that.
At the dorm I ate dinner of stewed squid and daikon with a ground beef korokke (which was actually mostly onion haha). Oh, and how could I forget the miso soup and rice!? After dinner I just kinda flaked, although I managed to forget to write my blog, and just as I was about to get ready for bed at the highly respectable time of 10:30, I remembered, and now here I am up at 12! I did actually brush and floss my teeth and catch up on a couple of Gunnerkrigg Court posts as well though, so I've only actually spent about an hour on the blog, which is pretty normal.
Anyways, I am enjoying my life on 8 hours of sleep a night (as opposed to less, of course. Derp.), so I am going to leave you with that, and bid you good night.
Night!
No comments:
Post a Comment