--- Log opened Thu May 17 00:00:40 2012 | ||
CIA-113 | shogun: Sergey Lisitsyn master * r6580ef2 / src/shogun/features/Labels.cpp : Fixed function introduced in the last commit - http://git.io/6vHSBQ | 00:18 |
---|---|---|
blackburn | hehe stupid coding | 00:18 |
-!- blackburn [~qdrgsm@188.122.250.167] has quit [Ping timeout: 252 seconds] | 01:34 | |
-!- emrecelikten [~Anubis@217.131.4.22] has quit [Quit: Leaving.] | 01:54 | |
-!- Francis_Chan [~Adium@58.194.224.108] has joined #shogun | 02:17 | |
-!- av3ngr [av3ngr@nat/redhat/x-qkezysfryayimbjt] has joined #shogun | 02:53 | |
-!- Francis_Chan [~Adium@58.194.224.108] has quit [Quit: Leaving.] | 03:47 | |
-!- abn_ [av3ngr@nat/redhat/x-luuwxqrxbhzoojli] has joined #shogun | 03:57 | |
-!- av3ngr [av3ngr@nat/redhat/x-qkezysfryayimbjt] has quit [Ping timeout: 272 seconds] | 04:02 | |
-!- av3ngr [av3ngr@nat/redhat/x-jkdqgjkzryrnmdft] has joined #shogun | 05:10 | |
-!- abn_ [av3ngr@nat/redhat/x-luuwxqrxbhzoojli] has quit [Ping timeout: 240 seconds] | 05:14 | |
-!- av3ngr [av3ngr@nat/redhat/x-jkdqgjkzryrnmdft] has quit [Read error: Connection reset by peer] | 06:37 | |
-!- av3ngr [av3ngr@nat/redhat/x-zzweqaftcracouxr] has joined #shogun | 06:38 | |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking] | 06:49 | |
-!- abn_ [av3ngr@nat/redhat/x-lsiqhomuwggtathm] has joined #shogun | 08:16 | |
-!- av3ngr [av3ngr@nat/redhat/x-zzweqaftcracouxr] has quit [Ping timeout: 245 seconds] | 08:20 | |
-!- uricamic [~uricamic@2001:718:2:1634:a8ca:b8c8:8e49:b04c] has joined #shogun | 08:56 | |
-!- blackburn [~qdrgsm@188.122.250.167] has joined #shogun | 09:10 | |
-!- Francis_Chan [~Adium@59.108.60.58] has joined #shogun | 09:41 | |
-!- abn_ [av3ngr@nat/redhat/x-lsiqhomuwggtathm] has quit [Quit: That's all folks!] | 10:33 | |
-!- Marty28 [~marty@cable-158-181-77-81.cust.telecolumbus.net] has joined #shogun | 11:00 | |
-!- Marty28 [~marty@cable-158-181-77-81.cust.telecolumbus.net] has quit [Quit: Colloquy for iPad - http://colloquy.mobi] | 11:10 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 11:41 | |
-!- blackburn [~qdrgsm@188.122.250.167] has quit [Quit: Leaving.] | 12:31 | |
-!- wiking [~wiking@c-98-231-171-175.hsd1.md.comcast.net] has joined #shogun | 13:09 | |
-!- wiking [~wiking@c-98-231-171-175.hsd1.md.comcast.net] has quit [Changing host] | 13:09 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 13:09 | |
-!- Francis_Chan [~Adium@59.108.60.58] has quit [Quit: Leaving.] | 13:35 | |
-!- vikram360 [~vikram360@117.192.186.253] has joined #shogun | 14:47 | |
-!- pluskid [~pluskid@111.120.28.216] has joined #shogun | 14:51 | |
-!- Francis_Chan [~Adium@58.194.224.108] has joined #shogun | 16:11 | |
-!- vikram360 [~vikram360@117.192.186.253] has quit [Ping timeout: 260 seconds] | 17:16 | |
-!- pluskid [~pluskid@111.120.28.216] has quit [Quit: Leaving] | 17:40 | |
-!- blackburn [~qdrgsm@188.122.250.167] has joined #shogun | 17:46 | |
n4nd0 | is there any reason why CList is not made with template? | 18:03 |
blackburn | n4nd0: some serialization issue I am afraid | 18:04 |
blackburn | but I am not sure | 18:04 |
n4nd0 | aham | 18:04 |
n4nd0 | I think it would be more comfortable with templates | 18:05 |
n4nd0 | not so many casts when returning elements for example | 18:05 |
blackburn | n4nd0: try to ask soeren once he is online | 18:11 |
n4nd0 | blackburn: I'll do that, thanks ;) | 18:11 |
-!- Francis_Chan [~Adium@58.194.224.108] has quit [Ping timeout: 252 seconds] | 18:29 | |
-!- gsomix [~gsomix@37.61.181.74] has quit [Quit: Ex-Chat] | 18:37 | |
-!- blackburn [~qdrgsm@188.122.250.167] has quit [Quit: Leaving.] | 18:48 | |
-!- blackburn [~blackburn@188.122.250.167] has joined #shogun | 18:50 | |
n4nd0 | hey wiking, how is it going? | 19:05 |
-!- uricamic [~uricamic@2001:718:2:1634:a8ca:b8c8:8e49:b04c] has quit [Quit: Leaving.] | 19:10 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 272 seconds] | 19:17 | |
-!- vikram360 [~vikram360@117.192.186.253] has joined #shogun | 19:27 | |
blackburn | students are getting crazy waiting for welcome packages | 20:00 |
-!- Netsplit *.net <-> *.split quits: @sonney2k | 20:10 | |
-!- Netsplit over, joins: sonney2k | 20:11 | |
-!- emrecelikten [~Anubis@82.222.148.100] has joined #shogun | 20:29 | |
emrecelikten | Hello all | 20:30 |
blackburn | hi | 20:31 |
emrecelikten | Stumbled upon this: http://groups.csail.mit.edu/netmit/sFFT/code.html | 20:34 |
blackburn | hah yeah that cat is cute | 20:34 |
-!- blackburn [~blackburn@188.122.250.167] has quit [Quit: Leaving.] | 20:52 | |
-!- blackburn [~blackburn@188.122.250.167] has joined #shogun | 20:54 | |
sonney2k | emrecelikten, heh :) | 21:06 |
emrecelikten | Might be useful in future I guess | 21:07 |
emrecelikten | Not the cat :P | 21:07 |
sonney2k | not until you did a fft^-1 | 21:07 |
emrecelikten | sonney2k: :D | 21:08 |
blackburn | is anybody aware of fourier features usage in CV? | 21:08 |
sonney2k | n4nd0, YMMV. it used to be template based but IMHO it is much more clean to use SGObject as the required base, since you know you can do SG_REF/UNREF | 21:09 |
sonney2k | but I agree lots of casting... | 21:10 |
blackburn | sonney2k: when did you plan next meeting? | 21:13 |
sonney2k | check the logs 2x.05. | 21:19 |
-!- mode/#shogun [+o sonney2k] by ChanServ | 21:21 | |
@sonney2k | blackburn, I've changed all the label stuff - but one question pops up. what shall classifiers output? | 22:02 |
@sonney2k | I mean BinaryLabels, MulticlassLabels, Regressionlabels | 22:03 |
@sonney2k | are all specific | 22:03 |
blackburn | sonney2k: hehe depends whether you did add some base class there | 22:03 |
@sonney2k | no | 22:03 |
blackburn | iirc it is ok to specialize class in descendant | 22:03 |
blackburn | then you will have to make apply class specific.. | 22:04 |
@sonney2k | well we have CLabels as base class | 22:04 |
@sonney2k | but it won't help us | 22:04 |
blackburn | why not? | 22:04 |
@sonney2k | because from python labels will be an abstract class | 22:04 |
@sonney2k | like features | 22:04 |
blackburn | sonney2k: as I can see the only way is to move apply to specific machines | 22:05 |
@sonney2k | so I only see two solutions: RealvaluedLabels or OutputLabels | 22:05 |
@sonney2k | blackburn, you mean binarymachine? | 22:06 |
@sonney2k | multiclassmachine? | 22:06 |
@sonney2k | regressionmachine? | 22:06 |
blackburn | sonney2k: yes binary machine, multiclass machine all has they own apply | 22:06 |
@sonney2k | makes sense | 22:06 |
@sonney2k | so Machine would loose its 'apply' interface function | 22:06 |
@sonney2k | only problem I see is that we have kernel machine | 22:06 |
@sonney2k | and that can be binary.mc, regression | 22:07 |
@sonney2k | any of it | 22:07 |
blackburn | KernelBinaryMachine.. | 22:07 |
@sonney2k | same w/ linear | 22:07 |
blackburn | yeah.. | 22:07 |
@sonney2k | blackburn, not so easy because e.g. SVRLight derives from SVMLight | 22:08 |
blackburn | heh | 22:09 |
@sonney2k | so I would rather suggest to remove apply() from CMachine | 22:09 |
@sonney2k | and then introduce some type specific apply to the machines we have | 22:10 |
blackburn | sonney2k: there is the diff? | 22:10 |
blackburn | where* | 22:10 |
blackburn | sonney2k: SVR/SVM can be solved I think - we can merge regression and binary labels | 22:11 |
@sonney2k | blackburn, which name do you prefer: DenseOutputLabels, RealvaluedLabels, ... | 22:15 |
blackburn | sonney2k: dense? | 22:16 |
blackburn | no not dense | 22:16 |
blackburn | RealLabels is ok I think | 22:16 |
blackburn | sonney2k: we should think about multiple outputs here.. | 22:16 |
@sonney2k | blackburn, I currently have regression labels - which would be the same as RealLabels | 22:17 |
@sonney2k | I think I would keep them though | 22:17 |
blackburn | sonney2k: yes, why? | 22:17 |
@sonney2k | because it is obvious that you want regression labels for regression, mc labels of mc, binary for binary classifiers... | 22:18 |
blackburn | true | 22:18 |
blackburn | but I am ok with real labels and multiclass labels | 22:19 |
blackburn | not binary, regression and mc I mean | 22:19 |
@sonney2k | blackburn, we need binary... I mean otherwise there is no way to check if labels are just +1/-1 | 22:19 |
blackburn | sonney2k: there is an advantage of real as binary - we can train on outputs of machine | 22:20 |
@sonney2k | blackburn, I am talking about inputs | 22:20 |
@sonney2k | outputs will only be RealLabels | 22:21 |
blackburn | oh different classes? | 22:21 |
blackburn | input/output? | 22:21 |
@sonney2k | yes of course | 22:23 |
@sonney2k | for binary we really prefer real valued outputs | 22:24 |
@sonney2k | we can add a constructor to binary labels that converts from real valued ones though | 22:24 |
@sonney2k | splitting at default threshold 0 | 22:25 |
@sonney2k | btw for multiclass I would also want to have some function that returns the outputs of all the sub-machines | 22:26 |
@sonney2k | real-valued... | 22:26 |
blackburn | yeap | 22:27 |
blackburn | sonney2k: I have used some OvR ROC/PRC/F1/Recall/Precision measures - do you think it should be added to evals too? | 22:27 |
@sonney2k | blackburn, for multi-output labels we should have a separate classes ... | 22:28 |
@sonney2k | sure | 22:28 |
@sonney2k | blackburn, btw what was the issue with auROC? | 22:28 |
blackburn | sonney2k: hmm? | 22:29 |
@sonney2k | blackburn, you had auROC=1.0 but accuracy <1 | 22:29 |
blackburn | sonney2k: it is possible | 22:29 |
@sonney2k | how? | 22:30 |
@sonney2k | ahh | 22:30 |
@sonney2k | bia | 22:30 |
@sonney2k | s | 22:30 |
@sonney2k | so no bug? | 22:30 |
blackburn | sonney2k: hmm no idea but it is | 22:30 |
blackburn | no no bug I compared with other impl | 22:30 |
blackburn | of you I guess? | 22:30 |
blackburn | in mldata utils | 22:30 |
@sonney2k | blackburn, yes | 22:30 |
blackburn | sonney2k: what is bias here? | 22:31 |
@sonney2k | simple example is add + 1000 to all classifications | 22:31 |
@sonney2k | auROC will stay correct | 22:31 |
@sonney2k | 100% | 22:31 |
@sonney2k | accuracy will drop | 22:31 |
blackburn | hmm why? | 22:31 |
@sonney2k | to 0 I guess | 22:31 |
blackburn | I admit I do not fully understand role of ROC | 22:31 |
@sonney2k | because sign(output+100000) -> +1 for everything | 22:31 |
blackburn | okay so ROC is TPR/FPR | 22:32 |
@sonney2k | for all possible bias terms | 22:32 |
@sonney2k | so independent of bias | 22:32 |
@sonney2k | auROC | 22:32 |
blackburn | sonney2k: ah so neither TPR nor FPR is not changed, right? | 22:33 |
blackburn | sonney2k: will the graph change at all? | 22:34 |
@sonney2k | no | 22:35 |
@sonney2k | same thing | 22:35 |
blackburn | sonney2k: sounds pretty useless | 22:35 |
@sonney2k | what sounds useless? | 22:37 |
blackburn | sonney2k: roc | 22:37 |
@sonney2k | why? | 22:37 |
@sonney2k | imho accuracy is useless | 22:37 |
@sonney2k | because it depends on bias | 22:37 |
blackburn | sonney2k: quality is pretty bad but auROC is ok.. | 22:37 |
@sonney2k | quality is actually superb | 22:38 |
@sonney2k | you should never trust the bias a machine outputs | 22:38 |
@sonney2k | but adjust it to what you want later | 22:39 |
@sonney2k | (wanted fpr or tpr) | 22:39 |
blackburn | sonney2k: so only b term is to be adjusted then? | 22:39 |
blackburn | sonney2k: cool wisdom hehe | 22:40 |
@sonney2k | yes | 22:40 |
blackburn | sonney2k: my ROC/PRC analysis led to some interesting results in my multiclass task | 22:42 |
blackburn | I found that one class really sucks hehe | 22:42 |
blackburn | http://dl.dropbox.com/u/10139213/share/00000_00001.png | 22:44 |
blackburn | sonney2k: what is the sign? :D :D | 22:44 |
@sonney2k | no cows :D | 22:45 |
blackburn | sonney2k: traffic light ahead | 22:45 |
blackburn | I have absolutely no idea how to recognize something what I can't recognize w/o prior :D | 22:46 |
@sonney2k | yeah no... | 22:48 |
@sonney2k | too tough | 22:48 |
@sonney2k | blackburn, CSupervisedKernelLocalTangentSpaceAlignment - does it need binary or MC features? | 23:09 |
@sonney2k | err labels | 23:09 |
blackburn | sonney2k: it needs to be removed | 23:09 |
@sonney2k | ? | 23:09 |
blackburn | sonney2k: *really* useless | 23:09 |
blackburn | and makes no sense at all | 23:10 |
@sonney2k | that doesn't answer the question | 23:10 |
@sonney2k | binary or mc? | 23:10 |
blackburn | sonney2k: mc but I will remove it anyway | 23:10 |
blackburn | sonney2k: let me better remove it now | 23:10 |
blackburn | sonney2k: I mean dim reducition is generally crappy | 23:11 |
blackburn | but it is really really useless shit | 23:11 |
blackburn | I just wanted to remove it once I am done with libedrt things | 23:12 |
blackburn | done | 23:13 |
CIA-113 | shogun: Sergey Lisitsyn master * rf1d5d22 / (4 files in 2 dirs): Removed SKLTSA algorithm - http://git.io/k40L9g | 23:13 |
blackburn | in the end I believe it was mistake to add it :D | 23:14 |
blackburn | same effect can be obtained with custom distance - but it is still totally useless to do that in that tangent mean | 23:15 |
blackburn | where are students? | 23:18 |
wiking | where | 23:34 |
blackburn | heh | 23:34 |
shogun-buildbot | build #836 of python_static is complete: Failure [failed test test_1] Build details are at http://www.shogun-toolbox.org/buildbot/builders/python_static/builds/836 blamelist: blackburn91@gmail.com | 23:39 |
blackburn | no space left? cool | 23:41 |
--- Log closed Fri May 18 00:00:40 2012 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!