2010-05-17.log

--- Day changed Mon May 17 2010
-!- marainein [~marainein@220-253-57-32.VIC.netspace.net.au] has joined #hplusroadmap00:46
-!- genehacker [genehacker@wireless-128-62-83-5.public.utexas.edu] has quit [Ping timeout: 246 seconds]01:31
-!- genehacker [genehacker@wireless-128-62-83-5.public.utexas.edu] has joined #hplusroadmap01:45
-!- clemux_ [clemux@lysine.clemux.info] has joined #hplusroadmap02:29
-!- clemux_ [clemux@lysine.clemux.info] has quit [Client Quit]02:29
-!- clemux_ [clemux@lysine.clemux.info] has joined #hplusroadmap02:39
-!- clemux_ [clemux@lysine.clemux.info] has quit [Client Quit]02:39
-!- genehacker [genehacker@wireless-128-62-83-5.public.utexas.edu] has quit [Ping timeout: 276 seconds]05:17
-!- quantumkat [~kat@ip72-196-96-148.ga.at.cox.net] has quit [Quit: Leaving]05:37
-!- JayDugger [~duggerj@pool-173-57-16-175.dllstx.fios.verizon.net] has joined #hplusroadmap05:54
-!- genehacker [genehacker@wireless-128-62-36-100.public.utexas.edu] has joined #hplusroadmap05:54
JayDuggerHa!05:55
JayDuggerRetweeted by not one, but two authors I like!05:55
JayDuggerThe "pleasant, simple, true; pick two" from here over the weekend.05:57
JayDuggerI did get one-upped with "smart, honest, effective." That's okay.05:58
-!- Incarnation [~lorem@66.51.248.231] has quit []06:04
-!- Incarnation [~lorem@66.51.248.231] has joined #hplusroadmap06:47
-!- marainein [~marainein@220-253-57-32.VIC.netspace.net.au] has quit [Quit: Ex-Chat]07:05
-!- wolfspraul [~wolfsprau@166-70-153-80.ip.xmission.com] has quit [Quit: leaving]07:48
-!- parolang [~user@8e4a01246100775874c4f448e9887093.oregonrd-wifi-1261.amplex.net] has joined #hplusroadmap08:11
-!- wolfspraul [~wolfsprau@elphelut.fttp.xmission.com] has joined #hplusroadmap08:13
-!- klafka [~klafka@cpe-66-66-5-254.rochester.res.rr.com] has quit [Quit: This computer has gone to sleep]08:22
-!- clemux [clemux@lysine.clemux.info] has quit [Quit: leaving]09:06
-!- clemux [clemux@lysine.clemux.info] has joined #hplusroadmap09:15
-!- Incarnation [~lorem@66.51.248.231] has quit []09:22
-!- clemux_ [clemux@lysine.clemux.info] has joined #hplusroadmap09:25
-!- klafka [~klafka@129.21.70.151] has joined #hplusroadmap09:27
-!- genehacker [genehacker@wireless-128-62-36-100.public.utexas.edu] has quit [Ping timeout: 252 seconds]09:30
-!- clemux_ [clemux@lysine.clemux.info] has quit [Quit: leaving]09:36
kanzurehttp://blueobelisk.shapado.com/questions/is-there-a-chemical-ontology-which-i-can-use-to-annotate-my-chemical-reactions09:36
-!- clemux [clemux@lysine.clemux.info] has quit [Quit: leaving]09:41
-!- clemux [clemux@lysine.clemux.info] has joined #hplusroadmap09:42
kanzureopen source nanotech http://news.ycombinator.com/item?id=133073109:44
kanzurebiocurious has a wiki http://www.biocurious.org/index.php?title=BiocuriousWiki:Current_events09:45
kanzurehttp://nextbigfuture.com/2010/05/billions-of-self-assembled-light.html09:45
kanzurehttp://www.cs.duke.edu/~alvy/papers/nsc-ret_final.pdf09:45
kanzurehttp://crnano.typepad.com/crnblog/2010/05/selfreplicating-nanorobots-now-possible.html09:46
-!- lepton [~john@149.sub-75-231-79.myvzw.com] has joined #hplusroadmap10:00
-!- kristianpaul [~kristianp@190.7.138.180] has quit [Quit: leaving]10:00
-!- kristianpaul [~kristianp@190.7.138.180] has joined #hplusroadmap10:01
-!- lepton [~john@149.sub-75-231-79.myvzw.com] has quit [Ping timeout: 240 seconds]10:12
-!- lepton [~john@248.sub-75-231-178.myvzw.com] has joined #hplusroadmap10:18
-!- lepton [~john@248.sub-75-231-178.myvzw.com] has quit [Ping timeout: 248 seconds]10:37
-!- jm|space [~jm@p57B9E4EC.dip.t-dialin.net] has joined #hplusroadmap10:39
-!- jm|earth [~jm@p57B9C0CF.dip.t-dialin.net] has quit [Ping timeout: 240 seconds]10:42
-!- lepton_ [~john@231.sub-75-220-194.myvzw.com] has joined #hplusroadmap10:48
-!- genehacker [genehacker@wireless-128-62-86-169.public.utexas.edu] has joined #hplusroadmap10:54
-!- lepton_ [~john@231.sub-75-220-194.myvzw.com] has quit [Ping timeout: 240 seconds]11:04
-!- lepton_ [~john@84.sub-75-231-133.myvzw.com] has joined #hplusroadmap11:05
-!- genehacker [genehacker@wireless-128-62-86-169.public.utexas.edu] has quit [Ping timeout: 260 seconds]11:22
kanzureblender: "11:31
kanzureblender: "11:31
kanzureIIRC, it's source code was purchased from the license holders and made open-source thanks to this same sort of thing.11:31
-!- quantumkat [~kat@ip72-196-96-148.ga.at.cox.net] has joined #hplusroadmap11:31
kanzurehttp://code.google.com/p/google-mobwrite/ patch theory code11:49
kanzurelepton_: what's the fastest turn around time that you've seen in a machine shop or engineering firm for idea-to-prototype?11:53
kanzurepreferably for internal ideas, but funded ideas too.. but i think they might be in separate categories11:53
lepton_How broadly are you defining idea-to-prototype11:54
lepton_Because I've "oh no we need a bracket" to the bracket existing in less than 24 hours11:55
kanzurehm11:55
lepton_But that's a pretty narrow scope (simple parts)11:55
kanzurefundamentally i want to complain about my experienced differences in prototyping time11:55
kanzurebetween software and hardware11:55
kanzurefor software, i can easily whip up lots of bullshit in under an hour11:55
kanzurefor machining, i find it much harder.. sourcing materials and such.11:55
lepton_it helps a lot to have inertia to go off of11:56
kanzure:)11:56
lepton_As in, a stock of material already on site11:56
kanzurehave you heard of "mental inertia"11:56
kanzureright11:56
lepton_sure11:56
kanzurebut it costs money to accumulate stock material11:56
kanzurehttp://improvingoneself.blogspot.com/2007/12/mental-inertia.html11:56
kanzurehttp://www.autistics.org/library/inertia.html11:56
lepton_aluminum recyclers are a great low cost material source11:56
kanzureanother issue is that i started off assuming i'd be near broke forever, so i was always trying to think in terms of avoiding money :)11:57
lepton_and the more projects tend to just in more material scape accumlating11:57
kanzureso.. calculations for how many coke cans i'd have to melt down for a 2x8x1 inch aluminum cast11:57
lepton_^oh my, my bad grammer11:57
kanzurebut now that i have some money, that has changed a bit11:57
kanzurebut it's still perplexing to me11:58
kanzurei also notice that others at the hackerspace take a pretty long time for turn around too11:58
kanzurewe've had a half-finished mechmate for 1.5 years now11:58
kanzurethe project is measured in years! that's unacceptable, imho11:58
lepton_Yeah, it can be tricky to get up and running11:58
lepton_Took us about a month at full time to got our shop functional11:59
kanzurewhat does functional mean11:59
lepton_We can make parts11:59
lepton_:)11:59
kanzureso the reason i asked about lowest turn around time is because i'd like to go meet the "fastest machinist"12:00
kanzureif that makes sense12:00
kanzuremachining operations take the same amount of time no matter where you go12:00
kanzurebut it's all the other stuff that matters too12:00
lepton_not always, different machines do have different feed / speed / load capabilities12:01
kanzurelike material sourcing, organization, getting stuff done12:01
kanzureheh12:01
lepton_A part it takes me 10 mineuts to do some machines could do in 112:01
lepton_or 0.512:01
kanzureyes but usually the order of magnitude is not significant12:01
lepton_True12:01
kanzureanyway, 1.5 years for a mechmate project wouldn't be changed much by a 1min or 10min machining time for parts12:01
lepton_And often setup is the biggest time component anyway12:01
lepton_or finshing work12:01
lepton_^Very true12:01
lepton_In my experience one of the more time consuming aspects of having a a well setup CNC system is getting setup with different kinds of tooling12:03
lepton_straight end mills, ball-nose end mills, chamfer mills, fillet cutters, surfacers, keyseat cutters, etc etc12:04
kanzuremaybe i have a warped perspective since the hackerspace isn't a fully equipped machine shop yet12:04
lepton_all with different kinds of materials, geometries, and what not12:04
kanzureperhaps things get smoother once you actually have functioning tools?12:04
lepton_It takes time to get things programmed in well12:04
lepton_sure!12:04
kanzurei can't quite run down at midnight at the moment and cut up some chunks of metal into funny shapes for the hell of it12:04
lepton_I think it gets progressivly easier12:04
lepton_Oh yeah, fixturing12:05
kanzurei should send you an inventory list12:05
kanzurethe inventory hasn't changed much in a year's time (guh)12:05
lepton_that's another huge element of CNC work and time consumption12:05
kanzurefenn: how long did your gingery lathe take?12:05
kanzurefenn's been working on hextatic, a stewart platform cnc machine, for at least half a decade now :(12:05
-!- lepton42 [~john@2.sub-75-231-3.myvzw.com] has joined #hplusroadmap12:07
kanzurelepton_: did you get my last two messages re: fenn?12:07
lepton42I don't think so12:07
kanzurefenn: how long did your gingery lathe take?12:07
kanzurefenn's been working on hextatic, a stewart platform cnc machine, for at least half a decade now :(12:07
lepton423G is not so great today12:07
lepton42Ah yes, I did get that12:07
lepton42I've been attracted to that mechanical design for a while12:08
-!- lepton_ [~john@84.sub-75-231-133.myvzw.com] has quit [Ping timeout: 260 seconds]12:09
-!- lepton_ [~john@0.sub-75-220-101.myvzw.com] has joined #hplusroadmap12:10
lepton_Damn these telecomm networks12:10
-!- superkuh [hukrepus@unaffiliated/superkuh] has quit [Ping timeout: 265 seconds]12:11
-!- lepton42 [~john@2.sub-75-231-3.myvzw.com] has quit [Ping timeout: 240 seconds]12:12
lepton_kanzure: Yesterday you said I should bug you about getting instructions for compiling GEOM for PythonOCC in a fresh Ubuntu 10.04 install...12:15
-!- Redeemer [~lorddeeme@c-75-72-218-226.hsd1.mn.comcast.net] has joined #hplusroadmap12:21
lepton_CNC Post Processor testing time!12:39
fennomg kanzure when you say it like that i feel old :(13:07
fenngingery took about a year elapsed, 3 months continuous13:07
fennwith breaks for meals and being lazy etc of course13:08
fenni was in college at the time13:08
-!- superkuh [hukrepus@unaffiliated/superkuh] has joined #hplusroadmap13:14
kanzurefenn: do you want to take a stab at pythonocc 0.4 + GEOM + SMESH?13:30
-!- Redeemer [~lorddeeme@c-75-72-218-226.hsd1.mn.comcast.net] has quit [Ping timeout: 240 seconds]13:33
kanzureit's okay if you say no, of course13:33
kanzurejust if you have time now, it will be quicker than lepton_ waiting until later today when i get it hacked out13:33
-!- Dustbin [~grumble@adsl-71-145-162-129.dsl.austtx.sbcglobal.net] has quit [Remote host closed the connection]13:37
lepton_I'm not in a rush today13:53
lepton_I'm gonna be working on our post processor for the rest of the foreseeable day/evening13:53
-!- lepton_ [~john@0.sub-75-220-101.myvzw.com] has quit [Ping timeout: 268 seconds]14:00
-!- lepton_ [~john@135.sub-75-231-10.myvzw.com] has joined #hplusroadmap14:04
fennno thanks14:12
-!- Redeemer [~lorddeeme@c-75-72-237-18.hsd1.mn.comcast.net] has joined #hplusroadmap14:16
-!- Redeemer [~lorddeeme@c-75-72-237-18.hsd1.mn.comcast.net] has quit [Ping timeout: 248 seconds]14:31
-!- Dustbin [~grumble@71.22.109.98] has joined #hplusroadmap15:09
-!- Redeemer [~lorddeeme@c-75-72-237-18.hsd1.mn.comcast.net] has joined #hplusroadmap15:09
-!- lepton_ [~john@135.sub-75-231-10.myvzw.com] has quit [Ping timeout: 276 seconds]15:12
-!- lepton_ [~john@84.sub-75-231-74.myvzw.com] has joined #hplusroadmap15:15
-!- klafka [~klafka@129.21.70.151] has quit [Quit: This computer has gone to sleep]15:29
-!- lepton_ [~john@84.sub-75-231-74.myvzw.com] has quit [Read error: Connection reset by peer]15:53
fennman, wikipedia is way better than these stanford class notes15:55
fennwhy did it take me so long to figure this out?15:56
-!- lepton_ [~john@84.sub-75-231-74.myvzw.com] has joined #hplusroadmap15:59
-!- lepton_ [~john@84.sub-75-231-74.myvzw.com] has quit [Ping timeout: 260 seconds]16:04
-!- wolfspraul [~wolfsprau@elphelut.fttp.xmission.com] has quit [Quit: Lost terminal]16:05
-!- lepton_ [~john@112.sub-75-220-119.myvzw.com] has joined #hplusroadmap16:08
-!- ybit [~quassel@unaffiliated/ybit] has quit [Read error: Operation timed out]16:08
-!- wolfspraul [~wolfsprau@lucia.q-ag.de] has joined #hplusroadmap16:11
-!- wolfspra1l [~wolfsprau@elphelut.fttp.xmission.com] has joined #hplusroadmap16:11
-!- wolfspra1l [~wolfsprau@elphelut.fttp.xmission.com] has quit [Read error: Connection reset by peer]16:11
-!- wolfspraul [~wolfsprau@lucia.q-ag.de] has quit [Ping timeout: 276 seconds]16:15
-!- wolfspraul [~wolfsprau@elphelut.fttp.xmission.com] has joined #hplusroadmap16:16
fennwonder if this was coded up from scratch: 16:20
kanzure..16:37
kanzurefenn: ?16:38
fennhttp://en.wikipedia.org/wiki/File:Airy-3d.svg16:39
fennsorry, connection is bunk16:40
fennsays "made in chipmunk basic"16:40
kanzureheh16:40
kanzuregnuplot has an svg export i think, btw16:40
fenndoes anything do curves instead of line segments?16:44
fennok this is annoying16:44
Dustbincurves instead of line segments in what16:45
fennin the plot16:48
Dustbinin exporting a plot?16:48
Dustbindon't know anything about that16:49
fennlike IGES instead of a quad mesh16:49
fennit doesnt matter, nevermind16:49
kanzurealright, i have 3 extra tickets to h+ summit 201016:54
kanzureplus a discount code for 20% off16:54
drazakwhere is it?16:54
kanzureboston16:54
drazakwhen is it?16:55
kanzurejune 12/1316:55
fennum.. how is it?16:55
fennwhy is it?16:55
drazakmhm16:55
kanzurehttp://hplussummit.com/speakers.html16:56
drazakI might be in boston around then16:56
kanzurei should have opted for a million trillion tickets16:56
drazakdepends on my class schedule16:56
kanzurehahah ticket scalping for h+16:56
kanzureYES16:56
fennyou FOOL16:56
kanzurehm?16:56
drazakyou're speaking?16:56
kanzureyes16:56
drazakbitchin16:56
drazakif I don't have classes then I might go16:57
kanzurealso demoing a facebook app via andrew hessel's presentation16:57
drazakand if my aunt/uncle have a couch16:57
kanzurei think my talk is right before ray's16:57
-!- Incarnation [~lorem@66.51.248.231] has joined #hplusroadmap16:57
kanzureoh boy16:57
kanzurei want joseph jackson to go on after ray though16:57
kanzureit's a nice compare/contrast16:57
drazakyou've got a grammar mistake on your bio16:58
-!- Dustbin [~grumble@71.22.109.98] has quit [Remote host closed the connection]16:58
fennjeebus that's a lot of speakers16:59
kanzuredrazak: the last sentence?16:59
drazakyeah16:59
drazakfenn: that's what I fuckin' said16:59
kanzure"you can find him or call him"16:59
drazakno16:59
drazakthe other last sentence16:59
kanzure..16:59
* kanzure slaps drazak with a trout16:59
drazak"lately he spends his waking hours at the recently new hackerspace..."17:00
drazakthat's the last sentence of the body17:00
kanzurehm17:00
kanzureso, what's wrong with it?17:00
drazak"recently new"17:00
kanzureare you sure? what rule does that fuck17:01
drazakit's awkward, and new doesn't match because recently is an adverb17:01
kanzureit's definitely awkward17:01
drazakrecently is an adverb, no noun17:01
drazaklately is on spends, recently is on ??17:01
fennnew is only a verb in c++17:01
drazakoh man, you guys are at harvard?17:01
kanzureyeah17:01
kanzureharvard, bitch!17:02
fenni thought drazak was at harvard17:02
kanzurebrown?17:02
drazakI've totally been where you're going to be17:02
drazakme?17:02
drazakfucking community college for 2 semesters17:02
drazakmy grades suck17:02
drazakbut that's ok17:02
fenndidnt you do some summer internship or something?17:02
drazakyeah17:02
kanzurenepotism at its finest17:02
drazakfuck yeah17:03
kanzurehm the speakers.html page seems to be randomly displaying the speakers17:03
fennthis is a beautifully rendered portable toilet or something like that.. anyway http://www.cadsoftsolutions.co.uk/store/software/shark/shark/files/page124_6.jpg17:03
kanzureexcept for the first few rows17:03
fenngah17:04
kanzurehey since when is mark hatch the ceo of techshop?17:04
kanzurewhen did that happen17:04
kanzureand why didn't i know about it17:04
drazakI think it is17:04
fennwhy do you care?17:04
kanzurebecause jim newton hit me up for money for his precious techshop17:04
kanzurebut where did that money go if.. uh17:04
fennjim newton hit you up for money?17:04
kanzureyes17:05
fenni hope you didnt give it to him17:05
kanzurei did not.17:06
kanzureso what's wrong with scalping, again?17:08
fennit may be illegal17:09
kanzurebesides that17:09
fennnothing17:09
kanzure"bryan's latest get rich quick scheme" haha17:09
fenndoes anybody actually understand the laplace transform?17:11
fennand not in a trivial "it's like the fourier transform" sense17:11
fenni'm trying to figure out what this means: "any LTI system can be characterized in the frequency domain by the system's transfer function, which is the Laplace transform of the system's impulse response"17:11
kanzureyay babble17:12
fennLTI=linear time invariant17:12
fenngod i hate math17:13
-!- klafka [~klafka@129.21.70.151] has joined #hplusroadmap17:15
-!- wolfspraul [~wolfsprau@elphelut.fttp.xmission.com] has quit [Quit: leaving]17:17
fennespecially shit like this: http://en.wikipedia.org/wiki/Dirac_delta_function17:17
fenna 'function' ?(x) that has the value zero everywhere except at x = 0 where its value is infinitely large in such a way that its total integral is 117:18
fennnot just any infinity, a particular value of infinity17:18
fennwhy? "because the math works out better that way"17:18
klafkait's a pretty weird function as functions go17:19
fennhow come i'm not allowed to just make shit up like that17:19
klafkayou are if you can prove something awesome with it17:19
klafkalike the ending to a futurama episode17:20
fenn"find the integral of f(x)" "1" "why" "because I said so"17:20
klafkait's weird that it's the limit of of the sequence of gaussians17:21
klafkabut kinda makes sense17:21
kanzurewhy isn't this working?17:42
kanzuredatetime.datetime.strptime("2010-05-07 21:32:46 -0500", "%Y-%m-%d %H:%M:%S %z")17:42
kanzureValueError: 'z' is a bad directive in format '%Y-%m-%d %H:%M:%S %z'17:42
kanzurehttp://docs.python.org/library/datetime.html#strftime-strptime-behavior17:43
kanzureaccording to the documentation, %z is valid17:43
fennwell this is rather nifty: 17:46
fennhttp://en.wikipedia.org/wiki/Automatic_differentiation17:46
kanzurehttp://bugs.python.org/issue6641 hm.. 17:46
kanzuregod damn it17:46
kanzureit's 2010, we should support #$!S@ZXcfefeq timezones17:47
fennhm i could use this machine learning stuff for discovering the parameters of a stewart platform17:50
klafkaoh yeah fenn i just learned about that in my statistical machine learning class a few weeks ago17:50
klafkaauto diff is very much like backprop17:50
kanzuresymbolic regression is hot17:50
klafkayeah17:51
klafkathere is some haskell package setup for autodiff i saw recently17:51
kanzurealso, hod lipson did a symbolic regression paper17:51
kanzurehttp://ccsl.mae.cornell.edu/eureqa_download17:51
fennklafka you're taking a class?17:51
fennwhat have you covered so far?17:51
klafkahttp://people.rit.edu/~jcdicsa/courses/SML/17:52
kanzurehttp://ccsl.mae.cornell.edu/sites/default/files/Schmidt-Final-2009-06-05.pdf17:52
klafkaclass just ended this week17:52
kanzureheh, SVMs can be fun :)17:52
fennrawr /me spears the lag monster dead17:52
fennyeah i was thinking earlier tha haskell would be a good fit for this stuff17:52
klafkakanzure, SVMs were a pain17:53
klafkaactually boosted decision stumps were very fun17:53
klafkai like that shit a lot17:53
fenneveryone says "SVMs are vun" but what's fun about it?17:54
kanzurethat's fenn's russian showing through17:54
fennit's just a fitting problem like every other regressionan17:54
fennargh17:54
kanzurei mean more what they are used for, more than anything else17:54
fenni havent learned any russian cuss words yet17:55
kanzurehttp://www.sourcemap.org/ from kristianpaul re: copyleft hardware17:55
fennbah it's all just "x kg CO2"17:57
fennlike i fucking care17:57
fennthe new dollar17:59
fenna totally useless number yet valued for its shared understandability17:59
* kanzure clicks18:00
kanzureomg this looks terrible18:00
kanzurebut people are going to eat this up18:00
fenni'm surprised at how new all of these machine learning algorithms are.. seems like all this should have been figured out in the 1930's 18:01
klafkaSVMs are basically a form of template methods18:01
klafkafenn, not really18:01
klafkaa lot of ML methods are inherently not probabilistic18:02
klafkaalso the 30's was when modern probability hteory was founded basically18:02
klafka20's/30's18:02
klafkaSVMs seem to be useful because they work with very little tweaking, to some extent, but they have a ton of drawbacks IMO18:02
klafkawell that and they _can_ have massive computational benefits when doing complex and expensive basis expansions 18:04
fenni'm still learning about it.. basically they are n-dimensional linear discriminators right?18:04
fennwe havent even talked about computational cost, which seems like a fucktard move t18:05
klafkaum i'm not sure about linear... ugh the term linear has taken on more confusing connotations since i've learned more about ML18:06
fenn"hey cs people, it's not math!"18:06
klafkawhat an SVM is is minimizing the hinge loss18:06
klafkaor what it does rather18:06
klafkais a technique for minimizing hinge loss18:06
klafkamy advisor told me hinge loss with ridge regularization, but i've had someone dispute that w/ me so i'm not sure18:07
klafkathe reason it's nice for minimizing hinge loss, is hinge loss is non-differentiable 18:07
klafkaso things like GD don't really work18:07
fennyeah i agree, i used to think linear meant "line" :(18:07
klafkaheh yeah18:07
-!- lepton_ [~john@112.sub-75-220-119.myvzw.com] has quit [Ping timeout: 276 seconds]18:08
-!- lepton_ [~john@6.sub-75-220-161.myvzw.com] has joined #hplusroadmap18:11
lepton_fenn: You still having Lapace space trouble?18:13
lepton_Laplace*18:13
fenni guess. it's one of those things i look at every ytear and can't vfigure out18:13
fennalso i'm terribly lagged.18:14
lepton_I didn't like it until my last year of school18:14
lepton_and then I realized how much easier it made things, and all the sudden it seemed easy18:14
lepton_What are you working on?18:14
fennmachine learning homework via wikipedia ramble18:15
klafkalol18:15
klafkamy final in it is on weds18:15
klafkaugh it's intense18:15
fenni realized the explanations in the course notes were even worse than wikipedia so i'd do myself a favor18:15
klafkawow18:16
klafkathat sucks18:16
klafkathe notes in the above link are pretty good18:16
klafkai think18:16
klafkaalso andrew ng's notes are also good18:16
fenn~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~yes i'm reading the backprop notes and it actually makes sense the first time around (for once)18:18
klafkahaha my advisor will be happy ot hear that18:18
lepton_:)18:18
fennklafka: andrew ng's notes are what i'm trying to avoid in the first place18:21
klafkalol are you taking his class?18:22
klafkaor does your teacher just use them18:23
fennng tends to give less than the minimum amount of information required to solve the problem18:23
klafkaaah18:23
klafkaidk i found some of his notes useful18:23
fenni'm constantly wondering "wtf is L" or similar18:23
klafkaLOL18:23
fenni'm taking the hacker dojo remix of cs22918:24
fennwhich is basically "do the homework ahead of time and we'll wank off for an hour"18:24
klafkalol18:24
kanzurehow is that different from the actual class18:24
fennbut i never seem to get around to actually doing the homework :\18:24
fennkanzure: the difference is i dont have to do the homework :)18:25
fennit's fucking stupid problems anyway18:25
fenn"manually calculate the desecond derivative of this arbitrary thing we arent going to tell you why it's importantn"18:25
kanzurewait18:25
kanzureso a few things you have said tonight have thrown me off18:26
kanzuresecond derivatives and simple integrals of continuous, smooth functions and such18:26
kanzureand then you complain about fourier transforms and laplace transforms in a complex number plane or something18:26
kanzureso which is it? a simple calculus class or what18:26
fennyes, doesn't sound much like a discrete computer problem does it18:26
kanzurethere are discrete methods of approximating integrals with polynomials i guess18:27
fennmost of it has been about calculus and matrix algebra so far, which doesn't make sense to me18:27
fennwtf am i supposed to do with this knowledge?18:27
fenni've been out of school too long to just suck it up and hope they 18:27
fennbestow enlightenment on us at theend of the class18:28
klafkayou mean the hessian18:28
klafka hessians are actually really important for newton's methods and approximate newtons methods like BFGS 18:28
klafkaalso for proving convexity18:29
fennkanzure: above "automatic differentiation" link would have been useful to know about, can bypass all we've learned in class so far pretty much18:29
kanzureif i had ops, i'd kickban acroynm users18:29
klafkawhich is important for a large number of algorithms18:29
kanzurefenn: i didn't click it, but is it symbolic differentiation stuff?18:29
klafkakanzure, it's basically forward prop/back prop18:29
kanzureokay.18:29
fennklafka yeah, see http://www.stanford.edu/class/cs229/ps1.pdf18:29
fennno18:30
fenngah18:30
fennthey dont explain what i'm supposed to do with knowledge of calculus that has anything to do with millions of data points18:31
klafkasee it makes sense to find the hessian there though man, by finding the hessian and proving it's negative semi-definite you've shown that the solution you arrive at via some optimization method is the global optimum18:31
lepton_Done with CNC post programming for the day, heading back home. Good luck with Lady Math, she's a harsh but rewarding mistress18:32
klafkafenn I contend that understanding the math behind the algorithms is important to understand why they work and allows you to know when you can apply the algorithm/methods to those millions of data points18:33
fennso i have all this data, and i'm trying to approximate it with a linear function, and somehow i'm proving the gradient of changing parameters is bowl shaped?18:33
fennthe bowl-shapedness of the parameter space is not under my control anyway so who cares if it's  is ior not?18:33
klafkait matters because if it wasn't then the approximation could be arbitrarily bad18:33
klafkathat's actually why we use logistic loss18:33
klafkaas opposed to the 0-1 loss or the hinge loss18:33
klafkathe logistic loss is the closest approximation to the 0-1 loss that is differentiable18:34
-!- lepton_ [~john@6.sub-75-220-161.myvzw.com] has quit []18:34
fennklafka: the prooblem is that was the first homework problem, and we're totally unprepared for what it was or why we're doing it18:34
klafkathe hinge loss is the closest approximation to the 0-1 loss that is um continous (I believe)18:34
fennin fact i still dont understand why it's important to know hot w to do that18:34
fenni hate typing blind18:34
fennmaybe i should connect locally then i can just timeout a bunch instead18:34
klafkalol18:34
klafkait's important because it gives you a deeper understanding of why peopledo these things18:35
klafkai mean you could just learn algo x does y, algo z does k but you're just memorizing shit18:35
klafkayou're not gaining deeper understanding18:35
klafkaalthough I can see how it should have been presented differently18:35
klafkai contend that these things are still useful to know18:36
fennalso wikipedia is html and not pdf which is a huge bonus in terms of speed18:37
fennklafka here's an example how bad it is at explaining stuff18:38
fenni went through the whole chapter on logistic regression 18:38
fennwithout realizing that the "log" in logistic was for "logical" not "logarithm"18:38
klafkalol18:38
klafkathe real kick in the pants is that logistic regression is for CLASSIFICATION not regression18:39
kanzurewhy do math students put up with this?18:39
klafkakanzure, most math students never touch this stuff18:39
klafkathis is pretty strictly the purview of CS 18:39
kanzurewhy do any students put up with this, then18:39
fennklafka it's ok because nobody seems to know what regression ameans anyway18:39
klafkalol18:40
klafkatrue true18:40
fennanother example of badness: i thought locally weighted linear regression returned a linear function (y=mx+b)18:42
fennturns out it's just some kind of spatial filte18:42
fennring like a gaussian blur18:42
klafkayeah18:45
fenni dont even know why it was included in the same chapter18:45
klafkalinear regression is a confusing idea imo 18:45
klafkain the same chapter as?18:45
klafkalogistic regression?18:46
klafkayeah i didn't like that either18:46
fennas linear regression18:46
klafkaah18:46
fenn(hint: it's not regression!)18:46
fennthere really arent enough pictures in ng's notes or lectures18:48
fennhe sprinkles in a few greaphs that all look the same18:48
klafkayeah18:48
klafkayou should really check out my advisors notes18:48
klafkaparticularly his linear methods notes18:48
klafkahe talks about linear regression, also talks about logistic regression18:48
klafkahas a lot of really cool graphs actually18:48
klafkahttp://people.rit.edu/~jcdicsa/courses/SML/03linear.pdf18:48
klafkaalternatively the elements of statistical learning does too18:49
fennok will do18:49
klafkabut some of his notes are pretty unique 18:50
klafkaugh you totally distracted me from this terrible paper i'm writing18:51
klafkaat least now it's running18:51
klafkaerr the code is running18:51
-!- Redeemer [~lorddeeme@c-75-72-237-18.hsd1.mn.comcast.net] has quit [Read error: Connection reset by peer]19:10
-!- Redeemer [~lorddeeme@c-75-72-237-18.hsd1.mn.comcast.net] has joined #hplusroadmap19:11
kristianpaulkanzure: yeah is not so good but the code is avaliable if i'm not wrong19:18
kristianpaulat least for making something really touching important aspects to think about  when stuff is made19:20
* kristianpaul off bed19:20
-!- genehacker [~genehacke@cpe-66-68-104-134.austin.res.rr.com] has joined #hplusroadmap19:25
kanzurehm where did lepton go19:30
kanzurei was about to ask him on his upload progress19:30
kanzureand the path location :/19:30
-!- Redeemer [~lorddeeme@c-75-72-237-18.hsd1.mn.comcast.net] has quit [Ping timeout: 268 seconds]20:15
-!- genehacker [~genehacke@cpe-66-68-104-134.austin.res.rr.com] has quit [Ping timeout: 260 seconds]20:24
-!- JayDugger [~duggerj@pool-173-57-16-175.dllstx.fios.verizon.net] has quit [Quit: Leaving.]20:41
-!- genehacker [genehacker@wireless-128-62-150-8.public.utexas.edu] has joined #hplusroadmap20:57
-!- Redeemer [~lorddeeme@c-75-72-218-226.hsd1.mn.comcast.net] has joined #hplusroadmap20:58
-!- klafka [~klafka@129.21.70.151] has quit [Quit: This computer has gone to sleep]21:13
-!- genehacker [genehacker@wireless-128-62-150-8.public.utexas.edu] has quit [Ping timeout: 258 seconds]21:13
-!- heath [~quassel@unaffiliated/ybit] has joined #hplusroadmap21:23
-!- genehacker [genehacker@wireless-128-62-150-8.public.utexas.edu] has joined #hplusroadmap21:23
-!- shepazu [~schepers@adsl-69-155-247.rmo.bellsouth.net] has quit [Ping timeout: 240 seconds]21:24
-!- shepazu [~schepers@adsl-242-202-103.rmo.bellsouth.net] has joined #hplusroadmap21:29
-!- genehacker [genehacker@wireless-128-62-150-8.public.utexas.edu] has quit [Ping timeout: 265 seconds]21:30
-!- quantumkat [~kat@ip72-196-96-148.ga.at.cox.net] has quit [Ping timeout: 276 seconds]21:43
-!- klafka [~klafka@cpe-66-66-5-254.rochester.res.rr.com] has joined #hplusroadmap21:44
-!- quantumkat [~kat@ip72-196-96-148.ga.at.cox.net] has joined #hplusroadmap21:44
-!- parolang [~user@8e4a01246100775874c4f448e9887093.oregonrd-wifi-1261.amplex.net] has quit [Remote host closed the connection]21:44
-!- genehacker [genehacker@wireless-128-62-150-8.public.utexas.edu] has joined #hplusroadmap21:44
-!- heath is now known as ybit22:07
-!- klafka [~klafka@cpe-66-66-5-254.rochester.res.rr.com] has quit [Quit: This computer has gone to sleep]22:27
ybitthanks kanzure for installing uzbl on davinci22:35
-!- klafka [~klafka@cpe-66-66-5-254.rochester.res.rr.com] has joined #hplusroadmap22:45
-!- shepazu [~schepers@adsl-242-202-103.rmo.bellsouth.net] has quit [Read error: Connection reset by peer]22:46
-!- shepazu [~schepers@adsl-242-202-103.rmo.bellsouth.net] has joined #hplusroadmap22:46
-!- klafka [~klafka@cpe-66-66-5-254.rochester.res.rr.com] has quit [Quit: This computer has gone to sleep]23:20
fennybit: you're welcome23:22
kanzureybit: you're welcome23:24
kanzureoh fuck23:24
fennlulz23:38

Generated by irclog2html.py 2.15.0.dev0 by Marius Gedminas - find it at mg.pov.lt!