--- Log opened Thu Jun 21 00:00:41 2012 | ||
-!- emrecelikten [~emre@176.41.10.108] has quit [Ping timeout: 246 seconds] | 03:11 | |
-!- emrecelikten [~emre@176.40.230.182] has joined #shogun | 03:27 | |
-!- emrecelikten [~emre@176.40.230.182] has quit [Quit: Leaving.] | 04:38 | |
-!- emrecelikten [~emre@176.40.230.182] has joined #shogun | 04:38 | |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking] | 05:18 | |
-!- emrecelikten [~emre@176.40.230.182] has quit [Quit: Leaving.] | 05:23 | |
-!- uricamic [~uricamic@2001:718:2:1634:5cfb:a4b3:604d:78a] has joined #shogun | 08:48 | |
-!- emrecelikten [~emre@176.40.230.182] has joined #shogun | 08:54 | |
@sonney2k | uricamic: could n4nd0 directly compare his .mc-so code with your example ? | 09:03 |
---|---|---|
uricamic | I don't know if he tried, but it should be possible now | 09:04 |
@sonney2k | I mean you do mc via the so framework but with bmrm right,? | 09:04 |
uricamic | yes | 09:04 |
uricamic | I told him that as soon as it will be finished, he can try to compare | 09:04 |
uricamic | btw I have some problems with that pull request | 09:04 |
@sonney2k | I assume you cannot have a true bias term? | 09:05 |
uricamic | is it possible to select only some commits for the PR? | 09:05 |
uricamic | no | 09:05 |
@sonney2k | no what? | 09:05 |
uricamic | the true bias term | 09:06 |
@sonney2k | well you need to create a new branch | 09:06 |
@sonney2k | so no bias term? | 09:06 |
uricamic | I see, ok, and then it is possible to move commits to the new branch? | 09:06 |
uricamic | yes, we are not using the bias term | 09:07 |
@sonney2k | in the worst case with git format-patch | 09:07 |
uricamic | ok, thanks, I will look on that | 09:08 |
@sonney2k | alright gtg | 09:08 |
-!- pluskid [~pluskid@202.130.113.141] has joined #shogun | 09:51 | |
pluskid | hi sonney2k, sorry, just see the email | 09:52 |
-!- emrecelikten [~emre@176.40.230.182] has quit [Quit: Leaving.] | 09:55 | |
-!- emrecelikten [~emre@176.40.230.182] has joined #shogun | 10:10 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 10:25 | |
@sonney2k | uricamic, did you manage to create a new pull request? | 12:21 |
@sonney2k | pluskid, I just wanted to say that it is Ok to merge - but we should use the shogun core functions to not have duplications everywhere | 12:22 |
@sonney2k | pluskid, but I guess it is better you have some working example before refactoring the code... | 12:23 |
pluskid | sonney2k: you mean memory allocation? | 12:26 |
@sonney2k | pluskid, yes and the vector functions like \alpha x etc | 12:26 |
@sonney2k | pluskid, I would want to see if the sse optimized stuff brings any gain (which I don't expect) | 12:27 |
@sonney2k | the compiler usually does the same | 12:27 |
pluskid | sonney2k: the lbfgs doc require that we use lbfgs_malloc to allocate memory, I still haven't look into its implementation to see what it has done, if it is just plain malloc, I think we can replace that with SG_MALLOC | 12:28 |
@sonney2k | pluskid, I looked at the code - they have their own malloc function and it uses malloc | 12:28 |
pluskid | sonney2k: and by "\alpha x" you mean which code? | 12:28 |
@sonney2k | actually it does malloc + zeros the stuff | 12:28 |
@sonney2k | so SG_CALLOC would be the replacement | 12:29 |
pluskid | sonney2k: OK, thanks! | 12:29 |
@sonney2k | https://github.com/shogun-toolbox/shogun/pull/590/files | 12:29 |
@sonney2k | look at arithmetic ansi | 12:29 |
pluskid | sonney2k: ah, I see, you mean we replace those with SGVector stuff? | 12:31 |
@sonney2k | pluskid, yes | 12:31 |
pluskid | I might be able to try to do this after I made ShareBoost working | 12:31 |
@sonney2k | or take their code if theirs is better | 12:31 |
@sonney2k | pluskid, it is very little amount of work | 12:32 |
@sonney2k | but yes | 12:32 |
@sonney2k | first some reference | 12:32 |
@sonney2k | otherwise it is hard to detect if one breaks things | 12:32 |
pluskid | but I guess keeping the original code untouched will be easier for future upgrade (to newer version of liblbfgs), what do you think? | 12:32 |
@sonney2k | pluskid, the last update is years back | 12:35 |
@sonney2k | so no | 12:35 |
@sonney2k | this lib is basically unmaintained | 12:35 |
pluskid | ok | 12:35 |
uricamic | sonney2k: I have done it right now, had some troubles with moving the commits, but hope now it is ok | 12:45 |
@sonney2k | uricamic, "This pull request cannot be automatically merged." | 12:45 |
@sonney2k | so no it is not... | 12:46 |
uricamic | hmm, I have done git cherry-pick, could this be because of that? | 12:46 |
@sonney2k | uricamic, my suggestion would be to update to the state of master | 12:46 |
@sonney2k | and then in a branch just apply the missing patches | 12:46 |
@sonney2k | uricamic, no idea - you have files in there that are already merged | 12:47 |
@sonney2k | for example I moved things from so -> structure a day ago or so | 12:47 |
@sonney2k | seems like you do the same thing now | 12:47 |
uricamic | I know, I have done some commit and then before PR it introduced the conflict, so I resolved it, but it seems it was not good | 12:48 |
uricamic | I will try to fix it | 12:49 |
@sonney2k | uricamic, look at https://github.com/shogun-toolbox/shogun/pull/593/commits | 12:50 |
@sonney2k | there is 'indention fixes' | 12:50 |
@sonney2k | which is long merged | 12:50 |
@sonney2k | uricamic, as I said I would suggest to do a git pull --rebase on your master branch | 12:56 |
@sonney2k | then resolve the conflicts | 12:56 |
@sonney2k | (git add the conflicting files afterwards) | 12:56 |
@sonney2k | and continue the merge | 12:57 |
n4nd0 | sonney2k: is there any reason why DynamicObjectArray is not includid in Library.i for the modular interfaces? | 13:17 |
n4nd0 | sonney2k: I have found it in ModelSelection.i but for some reason to do get_name from python fails with it, I want to include in Library and do the rename and so | 13:18 |
@sonney2k | n4nd0, yes please fix | 13:20 |
n4nd0 | ok | 13:20 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: Lost terminal] | 13:49 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 13:50 | |
@sonney2k | uricamic, did you get it to work? | 13:51 |
uricamic | sonney2k: working on it, I think I have found the source of the problem, so I have decided to introduce new commits from scratch | 13:55 |
uricamic | soon there will be a PR | 13:56 |
@sonney2k | uricamic, ok | 13:56 |
uricamic | sonney2k: done. I hope now it will work | 14:02 |
CIA-18 | shogun: Michal Uricar master * r2cbb216 / (2 files): libbmrm fixes - http://git.io/IVx0MA | 14:04 |
CIA-18 | shogun: Michal Uricar master * r6e12559 / (3 files): minor RiskFunction change + MulticlassRiskFunction implementation - http://git.io/mh_HQg | 14:04 |
CIA-18 | shogun: Michal Uricar master * rd8c48bf / (2 files): DualLibQPBMSOSVM fixes due to changes in libbmrm and RiskFunction - http://git.io/J8BFaw | 14:04 |
CIA-18 | shogun: Soeren Sonnenburg master * ra53beb2 / (7 files): Merge pull request #594 from uricamic/BM_SOL2 - http://git.io/yWso-w | 14:04 |
@sonney2k | uricamic, alright | 14:04 |
@sonney2k | thanks | 14:04 |
@sonney2k | uricamic, so could n4nd0 use this for comparison to his MC-SO mosek based thing? | 14:04 |
uricamic | I thank you for patience :) | 14:04 |
uricamic | I think he can | 14:04 |
@sonney2k | uricamic, btw do you have the value objective at the end? | 14:04 |
@sonney2k | then you could both compare nicely... | 14:05 |
uricamic | well now it is only written on output, I think, but it is there | 14:05 |
uricamic | oh, I just realized, I haven't post the example to this PR | 14:06 |
n4nd0 | yes yes, I want to test my stuff! testing testing :D | 14:06 |
n4nd0 | sonney2k: btw, is there any way to compile and just show the warnings? | 14:09 |
n4nd0 | any flag or so | 14:09 |
n4nd0 | otherwise I will redirect the output and grep or something like that | 14:10 |
@sonney2k | n4nd0 ... I use vim and :mak it will directly stop at all warnings/errors | 14:10 |
n4nd0 | ok, thank you! | 14:11 |
shogun-buildbot | build #1082 of libshogun is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/libshogun/builds/1082 blamelist: uricar.michal@gmail.com | 14:11 |
@sonney2k | uricamic, http://www.shogun-toolbox.org/buildbot/builders/libshogun/builds/1082/steps/compile/logs/stdio ? | 14:15 |
@sonney2k | seen the error? | 14:15 |
uricamic | this is because ot that changes, I will fix it | 14:16 |
shogun-buildbot | build #1083 of libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/libshogun/builds/1083 | 14:16 |
uricamic | but this is strange | 14:19 |
uricamic | I just compiled it on my computer and everything works | 14:20 |
@sonney2k | uricamic, well you committed something that doesn't compile | 14:22 |
@sonney2k | and then later fixed it | 14:22 |
@sonney2k | and committed again | 14:22 |
@sonney2k | uricamic, next time please use git commit --amend | 14:23 |
@sonney2k | to create just one *but working* commit | 14:23 |
uricamic | so it is ok now? | 14:23 |
@sonney2k | uricamic, yes ^ buildbot above said so | 14:34 |
uricamic | aha, ok, thanks, next time I will be more careful | 14:35 |
uricamic | I am trying to push also the example I forgot to push it into the last PR, but it seems something is wrong with my connection | 14:36 |
uricamic | I have sent the PR with the example | 14:56 |
-!- uricamic [~uricamic@2001:718:2:1634:5cfb:a4b3:604d:78a] has quit [Quit: Leaving.] | 15:25 | |
-!- pluskid [~pluskid@202.130.113.141] has quit [Ping timeout: 246 seconds] | 15:57 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 16:10 | |
-!- pluskid [~pluskid@111.120.60.54] has joined #shogun | 16:16 | |
-!- blackburn [d557894d@gateway/web/freenode/ip.213.87.137.77] has joined #shogun | 16:25 | |
blackburn | hey there | 16:25 |
blackburn | got bsc :) | 16:26 |
-!- ckwidmer [8ca3fe9d@gateway/web/freenode/ip.140.163.254.157] has joined #shogun | 16:27 | |
-!- pluskid [~pluskid@111.120.60.54] has quit [Ping timeout: 264 seconds] | 16:49 | |
-!- pluskid [~pluskid@202.130.113.141] has joined #shogun | 16:50 | |
-!- blackburn [d557894d@gateway/web/freenode/ip.213.87.137.77] has quit [Ping timeout: 245 seconds] | 17:05 | |
-!- pluskid [~pluskid@202.130.113.141] has quit [Quit: Leaving] | 17:15 | |
-!- gsomix [~gsomix@188.168.14.46] has joined #shogun | 17:40 | |
-!- alexlovesdata [c25faea9@gateway/web/freenode/ip.194.95.174.169] has joined #shogun | 18:29 | |
@sonney2k | gsomix, how is it going? recovered from the tornado? | 18:48 |
gsomix | sonney2k, huh. yes. minimal destruction. | 18:48 |
@sonney2k | gsomix, any updated regarding gsoc besides reconstruction works? | 18:49 |
gsomix | sonney2k, not many, only drafts of examples. I'm little busy by last exams now. However, I'll finish with examples this week. | 18:51 |
@sonney2k | gsomix, when are your exams? | 18:52 |
gsomix | sonney2k, last on 25 June. is there some job related to director classes except examples for machines and dotfeatures? | 18:53 |
gsomix | and I have passed optics exam at 19 June. | 18:56 |
@sonney2k | gsomix, well I can think of sparse features, string features, multiclass machines, preprocessors, structured output machines, converters, loss, labels | 18:58 |
gsomix | wow! | 18:58 |
@sonney2k | before midterm ... | 18:58 |
@sonney2k | then after that the other cool feature | 18:59 |
@sonney2k | python buffer protocol | 18:59 |
@sonney2k | that is not so easy to get to work | 18:59 |
@sonney2k | anyway, what it does is it will enable you to work with shogun densefeatures etc as if they are numpy matrices | 19:00 |
@sonney2k | so one can from python do x + y (with x being a shogun dense feature obj and y being a numpy array) | 19:00 |
@sonney2k | and even x+=y | 19:00 |
@sonney2k | such stuff | 19:01 |
@sonney2k | gsomix, for that you will need to write a simply python extension (stand-a-alone like for directors) | 19:02 |
@sonney2k | that just is some double* x, int len | 19:02 |
@sonney2k | some vector thingy | 19:02 |
@sonney2k | and then register a new pyobject for that type implementing the buffer protocol | 19:03 |
@sonney2k | if this works | 19:03 |
@sonney2k | the next step is to create a simple swig example which does the same | 19:03 |
@sonney2k | that will require a very recent version of swig though. | 19:04 |
@sonney2k | if that works - do this inside shogun and add some convenience operations to shogun features | 19:05 |
alexlovesdata | is it possible that MKL is broken in 1.1.0?? | 19:05 |
gsomix | sonney2k, swig2? | 19:05 |
@sonney2k | maybe 2.0.6 or so | 19:06 |
@sonney2k | not sure | 19:06 |
gsomix | ok | 19:06 |
gsomix | >> with shogun densefeatures etc as if they are numpy matrices | it's cool feature. I think that cpp+python+swig is one of best "platform" for programming | 19:07 |
gsomix | after lisp, haskell... huh, joke :) | 19:08 |
CIA-18 | shogun: Michal Uricar master * ra5c9046 / examples/undocumented/libshogun/so_multiclass_BMRM.cpp : BMRM SOL Multiclass example - http://git.io/urAKog | 19:08 |
CIA-18 | shogun: Soeren Sonnenburg master * r9c2be58 / examples/undocumented/libshogun/so_multiclass_BMRM.cpp : Merge pull request #595 from uricamic/BM_SOL_EXAMPLE - http://git.io/RVDxgw | 19:08 |
@sonney2k | gsomix, the director stuff is not a lot of work now that we know how it works | 19:09 |
@sonney2k | only dotfeatures is involving | 19:09 |
gsomix | sonney2k, get it. so, thanks for further plan. you made my work easier. :) | 19:16 |
-!- ckwidmer [8ca3fe9d@gateway/web/freenode/ip.140.163.254.157] has quit [Ping timeout: 245 seconds] | 19:34 | |
-!- gsomix [~gsomix@188.168.14.46] has quit [Remote host closed the connection] | 21:08 | |
-!- blackburn [d557894d@gateway/web/freenode/ip.213.87.137.77] has joined #shogun | 23:06 | |
blackburn | hey there | 23:07 |
blackburn | sonney2k: around? | 23:11 |
@sonney2k | blackburn, yes | 23:12 |
@sonney2k | blackburn, seen that you broke the tests with chris' stuff? | 23:12 |
@sonney2k | blackburn, ahh and congrats! | 23:12 |
blackburn | sonney2k: yes yes yes | 23:12 |
blackburn | thanks | 23:12 |
blackburn | sonney2k: I'll fix tomorrow | 23:12 |
@sonney2k | too much vodka already? | 23:13 |
blackburn | no had some headache after all | 23:13 |
blackburn | so had to get some sleep :) | 23:13 |
@sonney2k | naah sleep ;-) | 23:14 |
blackburn | I wish I had some vodka and had headache because of that :) | 23:14 |
blackburn | sonney2k: ok my plans tomorrow include fixing that | 23:15 |
blackburn | and sparsevector too | 23:16 |
blackburn | the last thing to change for referenced data | 23:16 |
@sonney2k | string stuff yes | 23:16 |
blackburn | sonney2k: what is purpose of cached vector? | 23:16 |
blackburn | I mean how do we get any impact on performance | 23:17 |
@sonney2k | blackburn, simple err dense features don't need to be in memory | 23:17 |
blackburn | that is the thing I am getting confused | 23:17 |
blackburn | sonney2k: but they have matrix aren't they? | 23:17 |
@sonney2k | and preprocessors can apply things just one by one | 23:17 |
@sonney2k | blackburn, no | 23:17 |
@sonney2k | matrix is optional | 23:17 |
@sonney2k | so things can be cached | 23:18 |
blackburn | how can we get densefeatures with no matrix but some contents inside? | 23:18 |
@sonney2k | that is why we have this get_feature_vector / free_feature_vector business | 23:18 |
blackburn | for example in python | 23:18 |
@sonney2k | with these functions | 23:18 |
@sonney2k | things are fetched/preprocessed/computed on-demand | 23:18 |
blackburn | sonney2k: did guys make some great progress on SO already? | 23:22 |
blackburn | a lot of commits I am lost in | 23:22 |
@sonney2k | blackburn, well at least some progress - framework is merged and uricamics bmrm with application to MC is in there | 23:23 |
@sonney2k | n4nd0 needs to compare against it and make some progress | 23:23 |
@sonney2k | blackburn, nothing from wiking yet - no idea where he is stuck... | 23:24 |
blackburn | sonney2k: isn't it easy to get basically anything to work after getting MC example to work? | 23:24 |
blackburn | I do not understand that point | 23:24 |
@sonney2k | the GP stuff seems also to be progressing well | 23:24 |
@sonney2k | and gsomix you know probably better | 23:24 |
blackburn | yes no news from wiking | 23:24 |
@sonney2k | and pluskid is working like 10 of us as usual | 23:25 |
@sonney2k | blackburn, yes MC is just the PoC | 23:25 |
blackburn | PoC? | 23:26 |
@sonney2k | then one 'only' needs to change the argmax | 23:26 |
@sonney2k | proof-of-concept | 23:26 |
blackburn | aham | 23:26 |
@sonney2k | but the argmax can be *very difficult* depending on so problem | 23:26 |
blackburn | does that mean they can integrate some HMM or anything pretty easy | 23:26 |
blackburn | ? | 23:26 |
blackburn | ahh | 23:26 |
@sonney2k | HM-SVM yes | 23:27 |
blackburn | so argmax is the most hard to code thing? | 23:27 |
blackburn | got it | 23:27 |
@sonney2k | blackburn, well HM-SVM is not so difficult | 23:27 |
blackburn | I actually checked HM-SVM code | 23:27 |
@sonney2k | after all it is just viterbi algorithm | 23:27 |
blackburn | I do not like it pretty much hehe | 23:27 |
@sonney2k | much more difficult is Semi HM-SVM | 23:27 |
blackburn | your expertise :) | 23:27 |
@sonney2k | non-markovian behavior on segments | 23:28 |
@sonney2k | backtracking from hell | 23:28 |
blackburn | semi-supervised? | 23:28 |
@sonney2k | no | 23:28 |
blackburn | semi what then? | 23:28 |
@sonney2k | the markov property is relaxed | 23:28 |
blackburn | how 'much'? | 23:28 |
@sonney2k | you can have arbitrary look back | 23:29 |
@sonney2k | so no longer p(x_t| x_{t-1}) | 23:29 |
blackburn | not limited? | 23:29 |
@sonney2k | limited! | 23:29 |
blackburn | with? | 23:29 |
@sonney2k | but up to whole segment length | 23:29 |
blackburn | ahh | 23:29 |
@sonney2k | so you can model things like length distribution of segment | 23:29 |
blackburn | okay nice | 23:29 |
blackburn | pretty complex thing | 23:30 |
@sonney2k | or frequencies of characters in segements | 23:30 |
@sonney2k | etc | 23:30 |
@sonney2k | whatever you want to compute for a whole segment | 23:30 |
blackburn | is it called semi markov? | 23:30 |
@sonney2k | argmax on graphs is probably even worse | 23:30 |
@sonney2k | yes | 23:30 |
blackburn | I see | 23:30 |
@sonney2k | sometimes generalized HM * | 23:30 |
blackburn | it looks like you have some passion with that :) | 23:30 |
@sonney2k | blackburn, recall that I wrote the HMM code | 23:31 |
blackburn | sonney2k: I recall you have master thesis on markov stuff | 23:31 |
@sonney2k | ...which I no longer can parse since gunnar parallelized it | 23:31 |
@sonney2k | blackburn, no | 23:31 |
@sonney2k | student research project only | 23:32 |
blackburn | am I wrong? | 23:32 |
blackburn | ahhh | 23:32 |
blackburn | what is your master thesis about then? | 23:32 |
@sonney2k | something before master thesis | 23:32 |
@sonney2k | hmm + top / fisher kernel + svm | 23:32 |
@sonney2k | for splice site recognition | 23:32 |
blackburn | ah more non-understable stuff :) | 23:32 |
@sonney2k | so I tried a couple of things to detect splice sites | 23:32 |
@sonney2k | hmms | 23:32 |
@sonney2k | locality improved kernel on splice sites | 23:33 |
@sonney2k | then fisher/top kernel on hmms | 23:33 |
@sonney2k | and this in the end evolved into the weighted degree kernel :) | 23:33 |
blackburn | it actually looks like some over-complexity for me | 23:33 |
blackburn | isn't it? | 23:33 |
@sonney2k | blackburn, no | 23:33 |
@sonney2k | sometimes you don't know how to compare things | 23:34 |
@sonney2k | but you have some good probabilistic models | 23:34 |
@sonney2k | so you derive the fisher/top kernel from it | 23:34 |
@sonney2k | and use an svm and get much better results... | 23:34 |
@sonney2k | this *can* be very complex | 23:34 |
blackburn | why to use svm if you have good prob model? | 23:34 |
@sonney2k | blackburn, because it is just some OKish model | 23:35 |
blackburn | I see | 23:36 |
@sonney2k | blackburn, for example GNB is probably very close to poly kernel | 23:37 |
blackburn | sonney2k: btw I want to know your opinion on semisupervised and transductive stuff | 23:37 |
@sonney2k | I mean the fisher kernel from GNB | 23:37 |
blackburn | (btw because you said semi word and I am activated now :D) | 23:37 |
@sonney2k | blackburn, not really a success story so far | 23:38 |
blackburn | sonney2k: I remember there is some relation between GNB and logit regression | 23:38 |
@sonney2k | blackburn, fisher kernel is just gradient of log(p(x)) | 23:39 |
@sonney2k | (except for some normalization that no one uses) | 23:40 |
blackburn | sonney2k: what do you consider as success story then? | 23:41 |
@sonney2k | well SVMs for classification | 23:42 |
@sonney2k | GPs for regression | 23:42 |
@sonney2k | linear methods in general | 23:42 |
@sonney2k | LDA, linear SVM ... | 23:42 |
@sonney2k | SO learning | 23:42 |
@sonney2k | not even true multiclass | 23:42 |
blackburn | sonney2k: I want to push my gf to research GPs | 23:42 |
blackburn | sonney2k: true multiclass is not kind of success for you? | 23:43 |
@sonney2k | well KNN is | 23:43 |
@sonney2k | and trees :D | 23:44 |
@sonney2k | blackburn, almost no gain compared to OvR | 23:44 |
blackburn | sonney2k: yes, I actually gain really small % of accuracy on my road sign data.. | 23:44 |
@sonney2k | ahh and PCA :) | 23:47 |
blackburn | sonney2k: what is so cool about PCA? | 23:48 |
@sonney2k | that it is often relevant in practise | 23:48 |
@sonney2k | there is lots of stuff out there that is rarely useful but overly complicated | 23:48 |
blackburn | sonney2k: I never managed to use it properly.. | 23:48 |
@sonney2k | blackburn, to get an idea/ visualize data only I mean | 23:49 |
blackburn | ah | 23:49 |
blackburn | LLE is cool for that too I believe | 23:49 |
blackburn | and isomap/mds | 23:49 |
blackburn | ltsa, hlle, blabla is useless but these ones are cool | 23:50 |
@sonney2k | I like stuff that works *often* | 23:50 |
@sonney2k | and is not so difficult and potentially even fast | 23:50 |
blackburn | mds and isomap work always | 23:50 |
@sonney2k | blackburn, work as in give useful results? | 23:51 |
@sonney2k | don't you need some distance for that? | 23:51 |
blackburn | sonney2k: yes always constructs some embedding | 23:51 |
blackburn | yes | 23:51 |
blackburn | sonney2k: I will implement t-SNE soon in shogun I think | 23:52 |
blackburn | it is probably the best 'visualizer' | 23:52 |
@sonney2k | oh well whatever method you choose - experience tells that you can gain much more by finding the right features | 23:52 |
@sonney2k | and this is actually the hard part | 23:53 |
blackburn | sonney2k: http://homepage.tudelft.nl/19j49/t-SNE_files/mnist_large.jpg | 23:53 |
blackburn | that's how t-SNE works on MNIST | 23:53 |
@sonney2k | impressive | 23:53 |
blackburn | yes, very impressive | 23:54 |
blackburn | sonney2k: http://homepage.tudelft.nl/19j49/t-SNE_files/vis_caltech101_small.jpg one more example | 23:54 |
blackburn | but actually LLE can produce similar results | 23:55 |
@sonney2k | I think the distance function matters a lot here | 23:56 |
-!- zxtx [~zv@cpe-75-83-151-252.socal.res.rr.com] has quit [Ping timeout: 246 seconds] | 23:56 | |
blackburn | exactly | 23:56 |
-!- shogun-buildbot [~shogun-bu@7nn.de] has quit [Quit: buildmaster reconfigured: bot disconnecting] | 23:56 | |
blackburn | but still method matters too | 23:56 |
-!- shogun-buildbot [~shogun-bu@7nn.de] has joined #shogun | 23:57 | |
@sonney2k | blackburn, the new buildbot will create documentation on nightly_default | 23:57 |
@sonney2k | if it doesn't fail of course | 23:57 |
blackburn | cool | 23:57 |
@sonney2k | blackburn, so highest prio is to get the build back to green | 23:57 |
blackburn | will do | 23:58 |
@sonney2k | blackburn, hmmhh I think I know some people on that image | 23:58 |
@sonney2k | blackburn, it will be under http://www.shogun-toolbox.org/doc/en/latest/ then | 23:58 |
@sonney2k | (if I didn't make a mistake :D | 23:59 |
blackburn | sonney2k: know? really? | 23:59 |
blackburn | it is caltech-101 dataset | 23:59 |
--- Log closed Fri Jun 22 00:00:20 2012 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!