Hey, remember these? I do. Vaguely. But vaguely isn't good enough, so here's another one. Since I've discovered that the secret to understanding machine learning problems well enough to implement them is mostly a matter of strategically procrastinating, it seems only fair that I should run one of these. So before cross-entropy actually starts making sense to me, what would you like to know?

deleted

So I just popped 175K on my '98 Taurus. I'm quite pleased with this vehicle. The AC is dead, and I've only had to replace the transmission once at 125K, when the planetary gear chewed itself to shit going to Washington back in 2008.

Any other gearheads reading this journal? What're you driving, and what're you pushing mileage-wise?

Took me a while, but after more careful digging, I've figured out what is happening. Last.fm is indeed lying. They are not listening to your media players. They are listening to your sound card output. Disabling sounds in Skype will disable every Skype event log except for two - updates to mood/music playing, and any time you make a skype call. Those to events trigger Last.fm logging. In the log, there's no real identifying information. What can be gleaned, though, is how long you made that call.

"If you choose to download the Last.fm software, it will monitor all the music you listen to on your computer (even if you are not online). It will then send this information to Last.fm via Audioscrobbler. This information will then be integrated into Last.fm to enhance the depth of your record collection and the accuracy of your profile.

The Last.fm software cannot and will not collect and transfer any other information from your computer other than as described above."

That is a flat-out LIE. And you're not even told about the integration with Skype when you install Last.fm's Audioscrobbler. You have to find out yourself by looking through log files, or go back to an obscure 2006 Skype Blog post.

This was most fun!

That'll teach me to cut my nails too short!

Someone on Slashdot recently claimed I hadn't read Keep the Aspidistra Flying because I thought the ending was depressing. After I finished my PhD in 2007, I've managed to avoid the same fate and have successfully avoided having a real job for almost five years. I've done freelance programming and written four books, and had a lot of time to post on Slashdot (as you can tell from the fact that, so far, I've posted more than anyone else this quarter) and do open source stuff (Ohloh ranks me in the top 2,000 geeks with no life^W^W^W^Wopen source developers).

That's about to change though. I had two interesting job offers recently (I seem to get job offers from banks very often, but I have a very low tolerance for tedium, so I'd probably have been fired around day 3 if I'd taken any of them). One was from Google in Paris (yay!) but working on boring things (boo!). The other was from Cambridge University, which is about as well paid as you expect in academia (aww!) but basically involves working on the same stuff I do for fun (yay!) with some very intelligent people (yay!). Oh, and it's in a city where a quick search found four tango classes (yay!) and property prices not much lower than London (oops!) and which is both small and flat enough that I can cycle everywhere (yay!) and so does everyone else (look out!).

So, in a few weeks I'm moving to Cambridge. I'll miss looking out at the sea, but being able to dance tango more than once a week should be some compensation. There also seems to be a lively salsa scene, although having to learn yet another set of names for the same Rueda steps is going to be a little tiresome...

When I visited, I went for drinks with some of the makerspace guys the night before my interview (I have no idea how much I drank, but it didn't seem to affect my interview performance too badly...) and met someone who worked on the C++11 atomics spec (which I was in the middle of implementing at the time) and someone who had ported 2BSD to a 32-bit PIC with 128KB of RAM, so it definitely seems like a city with no shortage of geeks...

After a little tiff with some angry man over on Reddit, I had to go conduct a few experiments of my own regarding the effectiveness of green light on photosynthesis.

To get to the meat of it all, between ~450 - 1800 umol, green light tends to drive photosynthesis more efficiently than red and blue light, especially in plants with much thicker leaves, like actual trees, succulents, cacti, etc. A tiny bit higher than 1800 umol, and your ROI drops rapidly as you've pretty much hit saturation of the chloroplasts, for all ranges of light. The law of Diminishing Returns comes to bite us in the butt, and no wavelength is ultimately more efficient than any other, and in fact, you begin reaching damaging levels of photon saturation which can burn a plant.

Now, there are multiple food crops that have an optimal light requirement below 500 umol. Multiple types of lettuce grow very well with ~200 umol, for an example. For that purpose, LED lights using red and blue light do well more than simply suffice. For rooting cuttings, red and blue light works very well. Fruiting plants typically require much more light, Jalapenos like it around 1500 umol. I've grown nice-sized jalapenos, bigger than what I'll find in most supermarkets, under a red and blue only panel, but I suspect I might be able to do a bit better with some focused green light. I'll do that when I get one made with equivalent specs.

Of course, getting to that intensity for larger plants will likely require me going solid white with a carefully-tuned color temperature. I have seen units now that on equal power draw with their HID counterparts are beating them by roughly 20%. However, these still require either EXTREMELY LARGE or very exotic cooling systems for anything over 300w and in a reasonably small package, and are not totally feasible for either green house or consumer gardening, this will likely be stuck in the parking light/street light arena for a while.

I have seen the benefits of green light, but there isn't MUCH additional benefit versus splitting up into different LED modules so you can get the light deeper into the canopy with the red/blue units. Perhaps for those doing bonsai, or growing cacti, and possibly underwater corals/plants, there is a good use, I can see it. Well, at least I'm sure of why an HPS works rather well despite having a very 'poor' spectral output. It's more than just the 660-680nm + IR that it dumps in massive quantities, it is also the green light having its own effective role.

Kinda goes totally against the conventional knowledge that red and blue are the most efficient, eh?

It's probably a bad idea to run one of these while I'm slogging through the project bulge of my last semester of undergrad, but let's take a stab at it anyway. Have at it! Clever trolls welcome!

I've gotten a couple of requests now for another one of these, so here one is. Please, bring to me your curiosities and questions about the strange and mysterious biological sciences. I can probably answer (or research answers to) most questions. (Also, if you keep missing these, I've decided I'm going to vaguely hint at them in mysterious ways on Twitter (@rhet0rica) from now on. I guess even Twitter has to have some utility.)

So Jorge Cervantes did a Reddit AMA. And he went on to say LED is not as good as CFL or HID.

As you see in the response to that post, I didn't let that slide, at all.

Don't dare talk shit about my profession when you're totally uninformed about the subject.

http://www.youtube.com/watch?v=9ZTikdxj8AI&feature=youtu.be

Finally, I have a copy of the video from the filming session done at the end of last year. Sorry that took so long to acquire!

It was great getting to watch (albeit remotely) the filming session.
Got any questions?

Bring me your curiosity! As before, if this expires and you want to ask a question, just slip me a comment (or an e-mail) and I'll put up another one of these. No question is too trivial; no thinly-veiled troll too transparent! (I'm going to regret saying that, I'm sure.)

Trying to figure out a way to go modular. I think the company I'm working with has a piece that I could modify into a modular plug and play system for my horticultural panels, so I can simply warranty repair with a swap out. It is going to take a fair amount of modification, however.

http://i.imgur.com/kGmNh.jpg

http://i.imgur.com/tY6C5.jpg

Can't show much more than that, unfortunately.

Looks like my tech is gaining some attention - the BBC's program "CountryFile" will be at the UK facilities for filming Monday.

This should be rather interesting. Shame I'm here in the USA - I guess I'll have to settle for a mere mention on the program instead of being physically present to explain things.