| --- Log opened Mon Jan 07 00:00:44 2008 |
| 00:21 | <@McMartin> | TF: Well, the thing about perceptrons is that once you chain NANDs together it's not a perceptron anymore. |
| 00:22 | <@ToxicFrog> | Yaeh. |
| 00:22 | <@McMartin> | Perceptrons are a dead end. The last word on them is "If it's a linearly separable function, you can do it. Otherwise it's completely impossible. End of story." |
| 00:22 | <@McMartin> | It's hard to get many papers out of three sentences. |
| 00:22 | <@ToxicFrog> | My complaint is that rather than going "well, I guess we need more than one perceptron to solve most things", they went "oh, these things are completely useless" |
| 00:22 | <@McMartin> | Well, then you get into The Land Of Spiders. |
| 00:22 | <@ToxicFrog> | Single perceptrons are, but they're the building blocks of neural networks! |
| 00:23 | <@McMartin> | My classwork with neural nets largely involved training a function and being unable to interpret the results. |
| 00:23 | <@McMartin> | Which we weren't graded on because not even the professor could reliably train them. We were graded on whether or not we had implemented back-propagation properly. |
| 00:23 | <@ToxicFrog> | ... |
| 00:24 | <@McMartin> | And that was, like, three perceptrons in a row or something. |
| 00:24 | <@McMartin> | And anything much more complex than XOR was pretty hopeless unless yout rained it on the entire input space. |
| 00:25 | <@McMartin> | Well, including XOR because if you train it on anything less you're not training XOR~ |
| 00:26 | <@ToxicFrog> | We had some fun with them, although we didn't do anything hugely complicated. |
| 00:26 | <@McMartin> | Yeah. We were supposed to be doing shape recognition, which once you get past the perceptron level was lolno. |
| 00:26 | <@ToxicFrog> | I'm just baffled. It's like looking at a single transistor and concluding that, since it's not logically complete in and of itself, it's of no use at all. |
| 00:27 | <@McMartin> | IMO the flaw was that they looked at a transistor and decided the problem they really wanted to solve was causing an FPGA to learn to be an x86 processor. |
| 00:27 | <@McMartin> | The non-CS part of that course we had much more fun with neural nets. |
| 00:28 | <@McMartin> | In particular, we played God and designed retinas capable of detecting various shapes. |
| 00:28 | <@ToxicFrog> | Sweet. |
| 00:28 | <@McMartin> | But then, that tends to fall into the perceptron range or just past it. |
| 00:28 | <@McMartin> | But yes, the programming part was one step away from the joke AI problem "teach a randomly wired neural net to play Tic-Tac-Toe" |
| 00:32 | | You're now known as TheWatcher[T-2] |
| 00:34 | | You're now known as TheWatcher[zZzZ] |
| 00:40 | | Orthia [~reiver@121.72.1.ns-3455] has joined #Code |
| 00:40 | | Orthia [~reiver@121.72.1.ns-3455] has left #Code [] |
| 00:47 | <@McMartin> | In other news, ha. Latest bug fixed in UQM: gas giants with negative mass. |
| 00:47 | < Finerty> | There were gas giants with negative mass? |
| 00:48 | <@McMartin> | Yeah. 16-bit values and cubes don't mix. |
| 00:48 | < Finerty> | ah, so. |
| 00:49 | < Finerty> | anything over 31 will make you negative. |
| 00:51 | <@McMartin> | Examples include: Eta Vulpeculae II; Alpha Maksutov II; Beta Gorno III; Alpha Lacaille VII; Beta Doradus II |
| 00:53 | <@McMartin> | http://bugs.uqm.stack.nl/show_bug.cgi?id=1025 |
| 01:51 | | Serah [~Z@87.72.35.ns-26506] has quit [Killed (NickServ (GHOST command used by ZLOK))] |
| 01:51 | | Serah [~Z@87.72.35.ns-26506] has joined #Code |
| 02:09 | | Finerty is now known as Vornicus |
| 02:55 | | gnolam [lenin@85.8.5.ns-20483] has quit [Quit: Z?] |
| 04:38 | < MyCatVerbs> | Why're you using 16 bit values? oO |
| 04:38 | < Vornicus> | Actually reading the bug it's 32 bit values |
| 04:40 | < Vornicus> | anything over 1290 (12.9 earth radiuses, in game values) overflows. |
| 04:41 | < Vornicus> | >>> (1<<31)**(1.0/3) |
| 04:41 | < Vornicus> | 1290.1591550923497 |
| 04:42 | < MyCatVerbs> | What do you use for a calculator? |
| 04:42 | < Vornicus> | Python |
| 04:43 | | * MyCatVerbs loves REPLs. ^_^ |
| 04:43 | < Vornicus> | REPL? |
| 04:43 | < MyCatVerbs> | Read-Eval-Print-Loop. |
| 04:43 | < Vornicus> | ah |
| 04:44 | < MyCatVerbs> | Python has one, Ruby has one, Lisp, Haskell, GhostScript, Scheme, all have one. Pretty much most enough all "nice" programming languages do, more or less. :) |
| 04:46 | <@McMartin> | Perl should have one, but doesn't. |
| 04:46 | <@McMartin> | BASIC has one! |
| 04:47 | < Vornicus> | Perl doesn't have one? |
| 04:47 | <@ToxicFrog> | ...perl doesn't have one? |
| 04:48 | <@ToxicFrog> | (Lua has a REL, although some people have written REPLs for it. Bash is a REL, as are most(?all) shell languages.) |
| 04:48 | <@McMartin> | OK, fine, BASIC has a REL, not a REPL. |
| 04:48 | <@McMartin> | Perl doesn't have one unless you write one yourself. |
| 04:48 | < Vornicus> | Suck. |
| 04:49 | < MyCatVerbs> | RELs are usually just about good enough, anyway. Just prefix things with "print" or whatever the local equivalent is. |
| 04:49 | <@ToxicFrog> | Well, writing one is, what, three lines? |
| 04:49 | < MyCatVerbs> | ToxicFrog: start: get line into x; print `eval x`; goto start; ? |
| 04:50 | < MyCatVerbs> | (translated, obviously, from horribly vague pseudocode to the local dialect) |
| 04:50 | <@McMartin> | IIRC it's a bit more than that because you have to prevent the loop from getting mucked with by the stuff typed in. |
| 04:50 | <@ToxicFrog> | Something like that. |
| 04:50 | <@McMartin> | It's in the Camel Book somewhere though. |
| 04:51 | < MyCatVerbs> | McMartin: eh, who cares? Just let the silly luser break things, it's their own fault. ;) |
| 04:51 | <@McMartin> | There's also the matter of $_ to worry about. |
| 04:52 | <@McMartin> | You don't want the REPL to reassign user-visible variables behind the user's back. |
| 04:53 | < MyCatVerbs> | I can see how that could get more involved. |
| 04:53 | <@McMartin> | Though yeah, it's still only a few stanzas of code. |
| 04:57 | < MyCatVerbs> | Fuck, the terminator is coming. |
| 04:58 | | * MyCatVerbs hides. Being able to see daylight progress across the face of the Earth with xplanet is cool, but also scary in that it brings the clock to life in a rather direct fashion. :) |
| 07:10 | <@McMartin> | Oh hey. The MS IDE will crash during debug on a use-before-def. |
| 07:48 | | Vornicus is now known as Vornicus-Latens |
| 08:11 | | You're now known as TheWatcher |
| 09:03 | | C_tiger [~c_wyz@Nightstar-5325.nycmny.east.verizon.net] has quit [Quit: And away she goes!] |
| 10:47 | | AnnoDomini [AnnoDomini@Nightstar-29093.neoplus.adsl.tpnet.pl] has quit [Ping Timeout] |
| 10:54 | | AnnoDomini [AnnoDomini@Nightstar-29119.neoplus.adsl.tpnet.pl] has joined #Code |
| 10:54 | | mode/#code [+o AnnoDomini] by ChanServ |
| 13:57 | | gnolam [lenin@85.8.5.ns-20483] has joined #Code |
| 13:58 | | mode/#code [+o gnolam] by ChanServ |
| 15:12 | | Otto_Flick [lenin@Nightstar-10613.8.5.253.static.se.wasadata.net] has joined #Code |
| 15:13 | | gnolam is now known as NSGuest-4183 |
| 15:13 | | Otto_Flick is now known as gnolam |
| 15:14 | | NSGuest-4183 [lenin@85.8.5.ns-20483] has quit [Ping Timeout] |
| 16:28 | | C_tiger [~c_wyz@Nightstar-5325.nycmny.east.verizon.net] has joined #code |
| 17:04 | | * AnnoDomini yays. Has just passed Reprogrammable Digital Systems ahead of schedule, as he has completed the Snake-like thing in VHDL. |
| 17:37 | | Forj [~Forj@203.211.126.ns-3357] has joined #code |
| 17:37 | | mode/#code [+o Forj] by ChanServ |
| 17:38 | | Forj [~Forj@203.211.126.ns-3357] has quit [Quit: Gone] |
| 18:03 | | Vornicus-Latens [~vorn@ServicesOp.Nightstar.Net] has quit [Ping Timeout] |
| 18:03 | < gnolam> | VHDL is a product of the Snake. :P |
| 18:04 | < gnolam> | I'm still not sure if it stands for "Verbose Hand-Destroying Language" or "Very Highly Demonic Language". |
| 18:07 | <@AnnoDomini> | My Snake's got five segments, each 10 by 10 pixels, moving at 10Hz. It can pause, doesn't reverse direction in place (usually), and loops around the screen. |
| 18:12 | | Vornotron [~vorn@Admin.Nightstar.Net] has joined #code |
| 18:49 | | Otto_Flick [lenin@Nightstar-10613.8.5.253.static.se.wasadata.net] has joined #Code |
| 18:51 | | gnolam [lenin@Nightstar-10613.8.5.253.static.se.wasadata.net] has quit [Ping Timeout] |
| 19:07 | | Vornotron is now known as Vornicus |
| 20:34 | | Otto_Flick is now known as gnolam |
| 21:34 | | * McMartin mutters at either gcc or MSVC, and for once he can't tell who's at fault. |
| 21:34 | <@McMartin> | Either MS is using an incompatible extension that by rights should be part of the standard since it's obviously semantically consistent, or gcc is being stupid about forward declarations. |
| --- Log closed Tue Jan 08 00:00:50 2008 |