Wikipedia:Reference desk/Archives/Computing/2011 September 12

Welcome to the Wikipedia Computing Reference Desk Archives
The page you are currently viewing is an archive page. While you can leave answers for any questions shown below, please ask new questions on one of the current reference desk pages.

September 12

Support Vector Machine time complexity

For a support vector machine, does the time complexity of classification increase linearly even when each item being classified only has a constant-size subset of the features (e.g. short messages)? Also, has any theoretical work been done on how to choose old training data to delete in order to keep the training data set limited in size (and thus work around the O(n^3) training time complexity) in cases where the data population is partly evolving and partly constant? NeonMerlin 05:57, 12 September 2011 (UTC)

Skylight Illumination references

I was wondering if there exists any free descriptions of the sky light illumination for different conditions (sun angle, altitude). The article on cube mapping mentions that considerable research has gone into modeling skylight illumination.

Given that the sun is about 0.5 degrees diameter, this basically needs to be a color and illumination value for the sun (per tiny unit view cone) and a simple phi, theta representation (or spherical harmonics) of the color and illumination of the sky (excluding the sun, per unit cone).

Without going into too much detail, I guess what I'm looking for is best expressed linearly, gamma=1.0 and in terms of r, g and b as would on average look realistic on a good calibrated computer monitor.

While very good discussions are easily found on Wikipedia I always struggle to find usable data. 193.171.188.3 (talk) 13:55, 12 September 2011 (UTC) Eon

...With or without clouds? But seriously, this is the least of your worries.

The most insidious problem is that the variation in luminous intensity between sky and sun is huge - we're talking, dozens of orders of magnitude - so there's no reasonable way to represent this linearly. You're already familiar with the results of this nonlinear, enormous magnitude problem: things like camera bloom, sun bleed, CCD "smear", saturation, and just generally screwing up exposure settings and auto-exposure algorithms. If the sun is in the scene, the illumination is very difficult to model linearly.

Photographers know this, and use optical filters when trying to photograph the sun; computer animators use a variety of ad-hoc tricks to render realistic-looking sun/sky (usually, by faking the electro-optical non-idealities like overexposure and sensor saturation). Nimur (talk) 17:02, 12 September 2011 (UTC)

If I understand you, you want to create a computer model of a room lit by a skylight. Is this correct ? If so, I'd expect that to be the same as a room lit by sunlight, when the Sun is visible through the skylight, and by "sky", at other times. Perhaps you were thinking it would be more yellow, when just the Sun shines through and the rest of the bluer light from the sky is blocked, but I doubt that the blue light has much effect relative to direct sunlight, even in a greenhouse. StuRat (talk) 17:12, 12 September 2011 (UTC)

I effectively want a computer model (HDR environment cube) to render realistic shadows that look like they are caused by sunlight. I don't want to have my "camera" look at the sun, I just need the sun to illuminate what the camera is looking at. Oh and yes, I mean a cloudless sky. There's nothing wrong with linear and I acknowledge that the sun will be maybe like a million times brighter per unit area than the sky, but that is no problem for floating point values.

Also, I disagree that everything other than the sun is insignificant. If that were the case then shadows would be almost 100% black. From insolation maps I found on the internet I gather that the total contribution of the sky dome (excluding the sun) is about 15-20% of the total light on average. — Preceding unsigned comment added by Eonzo (talk • contribs) 17:46, 12 September 2011 (UTC)

Pure black shadows assume that no light is reflected off objects in the room, and I didn't say that. Try a yellow spot light for the sun (not a point source, as the shadow edges would be too sharp) and white to bluish-white ambient light for the reflected light (and/or skylight). And, yes, a 20% difference in light level is barely noticeable, since perception of light is highly nonlinear. StuRat (talk) 18:31, 12 September 2011 (UTC)

Sorry, didn't mean to misquote you then. But I never meant a room. I'm not sure if "skylight" implied a room, but it is outside that I am interested in. Still I think a shadow outside on a totally flat meadow of grass (no objects nearby) will still show significant light. Also, while I want to do all the computations in linear I would eventually gamma compress the render result so even if shadows were very dark then they would still probably show non-black in the final image.

Oh, OK, I was thinking by skylight you meant a window in the ceiling of a building. So, there's no question of the sky being blocked then. In that case, add a bit more ambient bluish-white light, but retain the yellow spot light. This is for a single frame. If you want a realistic movie, then you might want to vary the light level, and even model shadows from passing clouds, to make it more realistic. StuRat (talk) 23:33, 12 September 2011 (UTC)

How about reading Global Illumination (Number 38 of NVidia's GPU Gems #2). If you want to render realistically, you should not aim for rendering scientifically-accurately; I'd worry less about a true model of sky illumination that is scientifically matched to empirical measurements; and I'd worry more about scene parametrization and rendering algorithms that make your final output look like it's illuminated by the sky.

#16 Accurate Atmospheric Scattering is also worth reading. In the words of the author, "Let's find out how poorly these equations will perform..." As I said, eschew accuracy, go for results. Nimur (talk) 18:16, 12 September 2011 (UTC)

Thanks, those are excellent resources.... but I really am only interested in how it looks "infinitely far" away. I don't want to actually model the scattering, I am just interested in the 2D cube map. I can probably measure it myself for my location at a given time of day if I have a color camera with known curve with adjustable exposure. The problem is probably going to be insufficient resolution around the actual sun (as is the case with most outside HDR cube maps available).

Why not just assume a uniform diffuse field? The difference between a shadow cast by a uniform diffuse field and a shadow cast by the actual sky luminosity function is small. I think you may be conflating diffuse shadow mapping with cube-mapping; I can't think of any good reason to use cube-mapping of the light source for rendering shadows. (In conventional algorithms, cube maps are used to store intermediate results - projected shadows of an object on the scene - are you planning to cast shadows on the sky?!) I think you just want a soft shadow, as if illuminated diffusely by the sky or other soft light. Here's an algorithm and implementation of soft shadows for NVidia GPUs, PCSS; and Blender's diffuse shader manual. If, after you read up on diffuse shadows, you are still convinced that you need a projective map of the sky luminosity, read our rayleigh scattering article and the diffuse sky radiation; Rayleigh sky model (a model of polarization, not luminous intensity); you can use the Rayleigh equations as an idealization of the sky luminosity for a given sun- and observation- angle. In reality, atmospheric effects, clouds, and so on, will have a huge effect; and there are other scattering physics details to consider (dispersion; Mie scattering; Raman scattering/frequency-splitting; and so on). I really think you just need to use a diffuse light-source to get the effect you want. Can you elaborate on your rendering tool? Are you using off-the-shelf software, or are you trying to write your own renderer/raytracer? If you're using off the shelf stuff, you should consult the user-guide for diffuse lighting techniques; if you're writing your own, I can point you toward some heavier math papers for statistical sampling and diffuse shadow generation. Nimur (talk) 03:51, 13 September 2011 (UTC)

I have written my own Monte Carlo ray-tracer with global illumination (albeit with limited functionality, but it works beautifully for the stuff I want). I already get soft shadows and diffuse lighting by illuminating my scene with a "sky dome" light source - that is an infinitely big sphere that illuminates to the inside. I've recently converted to a "sky cube" light source (which I find to be a bit faster). So now I am concerned with accuracy for the most convincing daylight shadows. You are right, assuming a single color blue for the sky and a slightly yellowish tint for the 0.5 degree diameter sun will probably be OK. Thank you SO much for your persistence to help me. Only thing is all I seem to get are discussions and formulas, but no hard NUMERICAL data. Like, I still don't know the actual sun to sky dome ratio (I gave my unreliable estimate of 15-20%). I love, love, love, Wikipedia, but if anything could be done to improve it it would be more numerical data. — Preceding unsigned comment added by Eonzo (talk • contribs) 08:08, 13 September 2011 (UTC)

DVD

How long do they last? 100years? I've heard about DVD rot.--213.107.74.132 (talk) 17:35, 12 September 2011 (UTC)

This page has a lot of information about this. The life expectancy of some writable DVDs is, unfortunately, as short as 1.9 years, it says. The estimates of 100 years or 1000 years or whatever are of course wild estimates, since the tech has not been around for that long yet. Comet Tuttle (talk) 17:55, 12 September 2011 (UTC)

Engineering design of DVDs is sufficiently well advanced that you can be sure they will last a lot less than 100 years. A lifetime as long as that could not possibly be in the interests of the recording industry and might well constitute a breach of fiduciary duty to the shareholders. Thincat (talk) 19:03, 12 September 2011 (UTC)

I hope that last statement was sarcasm. If not, I have to challenge you to find a single case worldwide where manufacturing a long-lived item was ruled to be a breach of fiduciary duty. Comet Tuttle (talk) 17:38, 13 September 2011 (UTC)

Extremely short lifetimes like 2 years apply to recordable ("burned") DVDs, which use a color-changing dye that can fade and blur over time. Cheaper recordable DVDs use less stable dyes as a rule, I think. Factory-made ("pressed") DVDs have physical pits in them. Ordinary DVD drives detect the pits by shining laser light through them and reflecting it off a shiny metallic layer behind the pits. That process can fail for various reasons, some easily avoidable (surface scratches) and some less so (rusting of the reflective layer, delamination, warping). But the pits are still there in any case, and I see no reason why they couldn't be read by some means even after thousands of years, if the plastic lasts that long. -- BenRG (talk) 23:21, 12 September 2011 (UTC)

I'm talking about DVD-/+R's, and RW's.--213.107.74.132 (talk) 08:52, 15 September 2011 (UTC)

Best way to extend WiFi signal range in a home

I have DSL Internet service at my home and my current wireless router does not provide a strong enough signal throughout the whole house. One option would be to move the current router to one half of the house and purchase a wireless repeater to serve the other half. Another option might be to purchase a second router, locate it in the underserved half of the house and set it to use a different wireless channel so as not to interfere with the first router. Which would be preferable, or is there a better solution? --Thomprod (talk) 17:41, 12 September 2011 (UTC)

Personally I fixed my WiFi signal problem by moving the router and also replacing the network card in my PC that was experiencing the bad signal quality; another option you might want to investigate is building a reflector and putting it behind the antenna in order to increase signal strength in one direction. Comet Tuttle (talk) 17:57, 12 September 2011 (UTC)

Along with moving the router is moving what blocks the router. Since I have wifi on my phone, I walked around dead spots and looked at the bars moving up and down. I moved the router and tried again. I pinpointed interference to some things that I did suspect (the refrigerator) and some I didn't suspect (an oil painting). I can't move the fridge, but I did move the painting to another wall to keep it from blocking the signal. -- kainaw™ 19:34, 12 September 2011 (UTC)

I have already tried moving the router. Unfortunately, this doesn't help since the house is all on one floor and quite spread out. There is no one location that adequately covers the whole house. Which one of my options would be better? --Thomprod (talk) 13:19, 13 September 2011 (UTC)

The wireless repeaters I have researched all effectively cut your local bandwidth in half. This may not be an issue for you — it wouldn't be for me, because my DSL connection runs at about 1/20th the Wireless-G maximum. An advantage of the repeater is that all the devices in the house can get configured the same way; but an advantage of having a second router is that the amount of wireless traffic that each router has to handle is cut in half. If you're going to be moving around the house a lot with even one of your devices, I'd probably elect the repeater, so the device isn't switching networks all the time as it gains and loses signal. Note that some routers can also be configured as repeaters. (My Netgear routers claim to be able to do this, though I have never done so, and I don't really recommend Netgear products anymore anyway.) Comet Tuttle (talk) 17:42, 13 September 2011 (UTC)

image sent from iphone

How can I download or save an image sent from an iphone to my Yahoo mail account? I can make the image appear on screen but I can't copy, save, or download it. Is there any way to download it other than screengrab? --Dr Dima (talk) 20:40, 12 September 2011 (UTC)

I don't have an answer, but what's wrong a screen grab ? Assuming the image fits on the screen at full res, you shouldn't lose any quality, and it might even be faster than a download, although a bit of cropping will be needed. StuRat (talk) 23:49, 12 September 2011 (UTC)

Three things are wrong with screengrab (vs. download): 1. you lose the whatever information was stored in the original file (e.g. date taken). 2. if you are saving into a lossy format (say, JPEG) you lose quality over the original. Otherwise, you end up with a file that is much larger than the original. 3. in most cases, some cropping occurs. All three problems can be, in principle, overcome; but why make it a project when there should be a simple and straightforward way to do this?! --Dr Dima (talk) 00:01, 13 September 2011 (UTC)

Can you explain in a little more detail? If the mail was sent with an email attachment (a mixed mode MIME multipart message per RFC 2046 §5.1.3, to be accurate - there are many types of multimedia messages an iPhone can send),, then you can just save the image file. It doesn't matter what source (iPhone or otherwise) it came from. Is there some reason Yahoo does not allow you to save attachments, or are you just struggling with the user interface? Here are instructions for dealing with attachments if you need them. Nimur (talk) 01:32, 13 September 2011 (UTC)

Those instructions are for Yahoo Mail Classic. If you're using the upgraded Yahoo Mail, the interface may look like this: Viewing and Downloading Images. --Bavi H (talk) 02:13, 13 September 2011 (UTC)

I know how to deal with attachments :) . Alas, this is not an attachment. This is some sort of embedded image; I can pull out the HTML line that generates it, it looks normal: <img id="..." src="http...">. However, when I try to follow the actual URL link the image does not show. Clicking on the image does not produce the usual "save image as" either; I can only copy or paste the useless hyperlink, but not the actual image. The actual image only shows when I paste the hyperlink in the yahoo mail compose window, but not anywhere else. --Dr Dima (talk) 05:43, 13 September 2011 (UTC)

OK, I've got it to work! The Yahoo mail site is so designed that it's possible to save the embedded images from the Internet Explorer but not from the Firefox. Attachments work fine in both browsers, of course, but some embedded images do not. Thanks anyway for your replies! --Dr Dima (talk) 06:03, 13 September 2011 (UTC)

Strange, very old encoding

Hi all:

I have a file format that contains strings that look like this:

AAEEEE\307ASJXXER\304 ...

I know that the \307 and \304 are supposed to be some way to encode a "reverse contrast" of one of the letters A-Z, but I have no idea what encoding it is or whether it's really called reverse contrast (the file was made by a program called "AE2" which I haven't been able to install). Any ideas? I need to convert them to a readable format so I can actually look at them...

Thanks for any help, 128.174.126.125 (talk) 20:52, 12 September 2011 (UTC)

If you strip the high bit from \307 and \304 they become G and D respectively (assuming this is ASCII). Some old software uses a set high bit to mark the end of a string. Of course, AAEEEEG ASJXXERD is still unreadable, so I'm not sure if that helps. -- BenRG (talk) 21:43, 12 September 2011 (UTC)

I doubt that this is an ASCII file. AE2 is probably Adobe After Effects, which deals with a whole bunch of different file formats, so without some further information I think this is going to be hard to resolve. Does the file have a filename extension? Can you tell us what the first bytes of the file say? (Often there is "magic" there that is useful for identifying a file format.) Looie496 (talk) 22:00, 12 September 2011 (UTC)

AE2 is a multiple alignment editor (these are protein sequences so AAEEEFG actually makes sense, although of course ASJXXERD could also make sense). ACtually the easiest thing for me to do may just be to BLAST The sequences and use the online sequence to fill in the blanks. But in any case, the extension is .ae and the first bits of the file are comments (#) - it's a text format of some sort. Thanks for the help. 128.174.126.125 (talk) 22:24, 12 September 2011 (UTC)

Aha, that clears things up quite a bit. Looking around, this page gives a lot of info about the program and how it works, and says that a program called convp can covert the file to other formats -- however, since this is Unix software dating from 1992, that might not help you very much. Looie496 (talk) 22:35, 12 September 2011 (UTC)

Looking around a bit more, it seems that there is a program called ARB that can read those files and convert them to other formats -- it appears to be maintained, but it requires a Linux environment to install and run. Looie496 (talk) 22:45, 12 September 2011 (UTC)

It seems that Ben was indeed correct that \307 and \304 correspond to G and D... I wasn't looking at it as if it was octal numbers, so wasn't expecting it to "skip" some of them (e.g. from 307 to 310) but it makes sense now. Thank you very much for the help! Mattb112885 (talk) 15:42, 13 September 2011 (UTC)

Freezing and hanging of desktop shortcuts

Recently, I've been experience several oddities of my Windows 7 laptop. Whenever I open something big (and by that I mean graphic intensive), the entire system freezes - I can't move the mouse, nor use the keyboard, and the only solution is to hold down the power button to turn it off. I suspect that this might be a problem with the graphics card, but I don't know what I could do about it.

Another problem that I experienced this morning was that, when I turned the laptop on and logged in, it seemed that Windows Explorer got hung on something. The start bar seemed fine, except that the network icon was always "loading", so to speak. Furthermore, none of the desktop shortcuts would appear, and hovering over the desktop background turned the mouse pointer into the circular loading cursor. Killing and restarting explorer.exe only produced the same problems. I finally fixed it by booting into safe mode and restarting the computer from there.

I know this is a (very) long shot, but what do you think might have caused these issues? I ran several antivirus scans both in Windows and at boottime, which all came up clean. I know that it's not likely that you can pinpoint the issue without knowing my system, but I would like some suggestions regardless so I know where to start looking. Thanks all. 66.36.130.43 (talk) 23:39, 12 September 2011 (UTC)

Those sound like two distinct problems, to me. As for the first one, an issue with the graphics card does seem likely. Do you happen to know how much graphics memory you have available ? If the problem is repeatable, you might want to try lower resolutions/color depths/refresh rates, to see if they hang, too, under identical circumstances. StuRat (talk) 23:44, 12 September 2011 (UTC)

I have two GPUs, actually. It's that Nvidia Optimus technology which switches between a Intel HD Graphics 3000 and GeForce GT 540M. According to the information in the Windows Control Panel, the Intel has 1969 MB of available graphics memory, while the GeForce has 3795 MB available. 66.36.130.43 (talk) 00:51, 13 September 2011 (UTC)

In that case, do you have the ability to force it to use one or the other ? (If not, I suppose you can always pull one out.) That might allow you to test with one, and then the other, to see where the problem occurs. If it doesn't occur in either, when used alone, perhaps it's the attempt to switch between GPUs mid-operation that causes the lock-up. StuRat (talk) 21:20, 14 September 2011 (UTC)

It sounds like you know all about Task Manager, as you mentioned killing explorer.exe. Anything anomalous looking on how much CPU the various things were taking? Anyway, the next thing I would check is Event Viewer, which will probably show you, at some level, what the problem is. In my case, my laptop also locked up for a minute at a time when it was doing something big and graphic-intensive, and Event Viewer indicated it was a hard disk controller error, which further Googling indicated was "almost certainly" (according to an authoritative-sounding microsoft.com page) a hardware problem. Comet Tuttle (talk) 17:53, 13 September 2011 (UTC)