IRC logs of #shogun for Wednesday, 2018-05-30

« Tuesday, 2018-05-29 Index Thursday, 2018-05-31 »
--- Log opened Wed May 30 00:00:28 2018
-!- durovo [~durovo@31.8f.559e.ip4.static.sl-reverse.com] has joined #shogun02:02
-!- durovo2 [~durovo@31.8f.559e.ip4.static.sl-reverse.com] has quit [Read error: Connection reset by peer]02:03
-!- Netsplit *.net <-> *.split quits: @ChanServ02:13
-!- Netsplit over, joins: @ChanServ02:34
-!- wuwei1 [wuweilinma@gateway/shell/matrix.org/x-lwvcxvbgbdmrbwiy] has quit [Ping timeout: 255 seconds]02:38
-!- wuwei [wuweilinma@gateway/shell/matrix.org/x-yjobwnmvmkogacxz] has joined #shogun03:00
-!- shogun-buildbot [~shogun-bu@7nn.de] has quit [Ping timeout: 248 seconds]05:56
-!- shogun-buildbot [~shogun-bu@7nn.de] has joined #shogun05:59
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/4314 opened by abhishek-199106:44
wuweiwiking; very weird that renaming make travis dying and i can reproduce that error locally07:50
@wikingwuwei, if you want i can restart that job07:59
@wikingsee if it's just a glitch07:59
wuweiwell.. i think that's actually a bug08:01
wuweiit's string preprocessor08:01
wuweiin python: feats = transformer.transform(feats)08:02
wuweiin stringpreproc::transform, in this case, the refcount is one when entering and exiting08:02
wuweiand then it's unrefed in python08:02
wuweibecause of '=', the original feats is unrefed08:02
wuweihowever in this case they are the same object08:03
wuweimeaning that, changing the line to: transformer.transform(feats)08:03
wuweiit works08:03
wuweibut i don't see any points that this should be happening08:03
wuweiah.. maybe that's related to %newobject in swig, let me check08:08
-!- sukey1 [~nodebot@ks312251.kimsufi.com] has joined #shogun08:27
-!- sukey [~nodebot@ks312251.kimsufi.com] has quit [Read error: Connection reset by peer]08:28
wuweidoes this line apply to things outside CMachine?09:11
wuweihttps://github.com/shogun-toolbox/shogun/blob/7e9699e5a1c7df75271e8dd26fc71e615e8138b4/src/interfaces/swig/Machine.i#L1109:11
sukey1[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/4285 synchronized by vinx1309:40
@wikingwuwei, imo yes10:02
wuweiyeah10:03
wuweitravis is green now :)10:03
sukey1[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/4285 merged by vigsterkr10:08
sukey1[https://github.com/shogun-toolbox/shogun] vigsterkr pushed 61 commits:10:08
sukey1https://github.com/shogun-toolbox/shogun/commit/6029e94a3cdad6ce20c9a420a11a273bda3d22c210:08
sukey1https://github.com/shogun-toolbox/shogun/commit/8a7dcabb7c6b40abcc69b3978e40e2b2378221d910:08
sukey1https://github.com/shogun-toolbox/shogun/commit/0d43482a9ee7542851e01baa6d936212cd0b48a110:08
@wikingboom10:08
sukey1https://github.com/shogun-toolbox/shogun/commit/9a0dcc067336e8d15cd0ebc8df1d650b0d76cff010:08
sukey1https://github.com/shogun-toolbox/shogun/commit/b9c7067241e1bb432a513f3393a71c14f187c18010:08
@wiking:D10:08
sukey1https://github.com/shogun-toolbox/shogun/commit/7acd42e985bd85eb686fd204da7a2ae76e8c5c9510:08
sukey1https://github.com/shogun-toolbox/shogun/commit/c26877aac1abb4735ec0f30c21d07397bdf9158310:08
sukey1https://github.com/shogun-toolbox/shogun/commit/5c87fdb38feee82f1258649deea8576e88a6890e10:08
sukey1https://github.com/shogun-toolbox/shogun/commit/4840a8caa8300687b06a85ea14e76e3f8b22ae6210:08
sukey1https://github.com/shogun-toolbox/shogun/commit/f93b8525cd58542c69b815a8a9bc562940299f1410:09
sukey1https://github.com/shogun-toolbox/shogun/commit/0155355f09e5395f2a638e2b132d85d4056b6b8110:09
sukey1https://github.com/shogun-toolbox/shogun/commit/f2d9b85d2870da7f61129fb3e64d9d8443d736aa10:09
sukey1https://github.com/shogun-toolbox/shogun/commit/556cb6ec51cb71b89260f7706fd9c8dc532812d110:09
sukey1https://github.com/shogun-toolbox/shogun/commit/c8fd10c75f42477502c7e7c47ffa814a71f6ff1f10:09
sukey1https://github.com/shogun-toolbox/shogun/commit/363d54aae91e810e75e1f9c8814057805739767b10:09
sukey1https://github.com/shogun-toolbox/shogun/commit/d2df975c05e94fe0d35e8da3ebd8f04957ba583510:09
sukey1https://github.com/shogun-toolbox/shogun/commit/582721c46c6ac141ea37064b99d1e3357e94259910:09
sukey1https://github.com/shogun-toolbox/shogun/commit/f32f61621299c86fd7cef627416abf176dd7833910:09
sukey1https://github.com/shogun-toolbox/shogun/commit/5f2ac96fe87275d5c26d941b75a6a7161839d7f510:09
sukey1https://github.com/shogun-toolbox/shogun/commit/6b098585028a139bb128e5877a505b3c052a397d10:09
sukey1https://github.com/shogun-toolbox/shogun/commit/67f02c4f5208d8e8308e8ecaadbeeae59e52856a10:09
sukey1https://github.com/shogun-toolbox/shogun/commit/66015f2fff7f4e707456f1dc4a68ae0dd9cd66d410:09
sukey1https://github.com/shogun-toolbox/shogun/commit/b9a95dd8212650b4ecd6d253d314e59d6aef711f10:09
sukey1https://github.com/shogun-toolbox/shogun/commit/cd798503ae19141bcccf8f69a9c19f998f53c23d10:09
sukey1https://github.com/shogun-toolbox/shogun/commit/ec66d1d6f3c12ef170097f67c7693f250be6132a10:09
sukey1https://github.com/shogun-toolbox/shogun/commit/f765b525534067a9db7baf826222c0e6dc46dc4d10:09
sukey1https://github.com/shogun-toolbox/shogun/commit/801633caac31ac898e32161fa5fda7409bc8326210:09
sukey1https://github.com/shogun-toolbox/shogun/commit/b383892c73fab74b3d3458ad3d3b9a923f613c2810:09
-!- sukey1 [~nodebot@ks312251.kimsufi.com] has quit [Remote host closed the connection]10:09
@wikingwuwei, great stuff!!!10:16
wuwei:D10:16
@wikingwuwei, ok so next one?10:17
@wikingpipeline?10:17
wuweiyeah10:17
@wikingccool10:17
@wikingjust asec10:17
@wikingin a meeting10:17
wuweishould i refactor out those old preproc api from features10:19
-!- HeikoS [~heiko@host81-153-166-11.range81-153.btcentralplus.com] has joined #shogun10:56
-!- mode/#shogun [+o HeikoS] by ChanServ10:56
-!- travis-ci [~travis-ci@ec2-184-73-117-198.compute-1.amazonaws.com] has joined #shogun10:59
travis-ciit's Wuwei Lin's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: https://travis-ci.org/shogun-toolbox/shogun/builds/38556988510:59
-!- travis-ci [~travis-ci@ec2-184-73-117-198.compute-1.amazonaws.com] has left #shogun []10:59
wuweiwiking: ready for today's meeting? :)11:06
-!- witness_ [uid10044@gateway/web/irccloud.com/x-smarjtpaswqoslpr] has joined #shogun11:07
-!- HeikoS [~heiko@host81-153-166-11.range81-153.btcentralplus.com] has quit [Ping timeout: 240 seconds]11:37
@wikingwuwei, yes11:49
@wikingi am11:49
@wikingi was in a meeting till now11:49
@wiking:DDD11:49
@wikingwuwei, i'm soooo bad :(11:49
@wikingsorry aboutit11:49
@wikingstill here?11:49
wuweisure no problem11:50
@wikingok so yeah11:50
@wikingactually maybe that would be better now11:50
@wikingbefore jumping into pipelines11:50
@wikingto clear the air about preprocessors in features11:50
wuweiapply preproc , I think this can be removed11:51
wuweiBut there are on the fly things that should be kept11:51
wuweilike apply to vector11:52
@wikingye11:52
-!- travis-ci [~travis-ci@ec2-54-224-19-64.compute-1.amazonaws.com] has joined #shogun11:55
travis-ciit's Wuwei Lin's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: https://travis-ci.org/shogun-toolbox/shogun/builds/38557046511:55
-!- travis-ci [~travis-ci@ec2-54-224-19-64.compute-1.amazonaws.com] has left #shogun []11:55
wuweibtw about pipeline, if in linear case it should be zero or more transformers, +1 machine as the final stage right ?11:56
@wikingyeah that's the idea11:57
wuweithen actually we don't need variant :)11:58
@wikingwell11:58
@wikingcoz you wanna store them separately? :)11:58
-!- sukey [~nodebot@ks312251.kimsufi.com] has joined #shogun11:59
-!- mode/#shogun [+o sukey] by ChanServ12:00
@wikingsukey, flip12:00
@sukey(????)??????/12:00
wuweiyeah12:00
wuweiOtherwise we have a vector of variant, and only the last element is machine12:01
@wikingyeah12:02
@wikingnow of course12:02
@wikingwe should keep in mind12:02
@wikingthe dag case right? :)12:02
@wikingwhen you can have 2 machines :P12:02
@wikingor more12:02
wuweioh yeah12:03
@wikinglisitsyn, you're back? :)12:03
@wikingwuwei, i mean for the time being we'll just work on linear stuff12:03
@wikingbut let's do things more flexible way in the bakcend12:04
@wikingso we dont need to completely change things :)12:04
wuweisure I see12:04
-!- travis-ci [~travis-ci@ec2-54-224-19-64.compute-1.amazonaws.com] has joined #shogun12:17
travis-ciit's Wuwei Lin's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: https://travis-ci.org/shogun-toolbox/shogun/builds/38557046512:17
-!- travis-ci [~travis-ci@ec2-54-224-19-64.compute-1.amazonaws.com] has left #shogun []12:17
-!- HeikoS [~heiko@eduroam-int-pat-8-0.ucl.ac.uk] has joined #shogun12:23
-!- mode/#shogun [+o HeikoS] by ChanServ12:23
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/4313 synchronized by karlnapf12:27
-!- HeikoS [~heiko@eduroam-int-pat-8-0.ucl.ac.uk] has quit [Ping timeout: 248 seconds]12:27
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/4313 synchronized by karlnapf12:32
@sukey[https://github.com/shogun-toolbox/shogun] Issue https://github.com/shogun-toolbox/shogun/issues/4315 opened by karlnapf12:36
@sukey[https://github.com/shogun-toolbox/shogun] Issue https://github.com/shogun-toolbox/shogun/issues/4315 karlnapf added label: "good first issue"12:36
@sukey[https://github.com/shogun-toolbox/shogun] Issue https://github.com/shogun-toolbox/shogun/issues/4315 karlnapf added label: "Cleanups"12:36
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has joined #shogun12:42
-!- mode/#shogun [+o HeikoS] by ChanServ12:42
@HeikoSlisitsyn: you around?13:22
@HeikoSwiking: can you send me the reimbursement infos for the hackathon in belgrade?13:23
@HeikoSI already put in the airbnb, but anything else pls send13:24
@wikingsure i'll try to send out asap13:24
@HeikoSjust need numbers13:24
@HeikoShave everything else13:24
@HeikoSfrom previous13:24
@wikingsure13:24
@wikingmeeting13:24
@wikingsorry13:24
@HeikoSkk13:24
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/4313 merged by karlnapf13:27
@sukey[https://github.com/shogun-toolbox/shogun] New commit https://github.com/shogun-toolbox/shogun/commit/bf2f54543425a62f36d02061da7d8b8f6005847c by karlnapf13:27
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/4310 merged by karlnapf13:55
@sukey[https://github.com/shogun-toolbox/shogun] New commit https://github.com/shogun-toolbox/shogun/commit/35449565ac35d72f91a79d60606e7cc43966af08 by karlnapf13:55
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/4316 opened by vinx1314:52
-!- travis-ci [~travis-ci@ec2-54-161-181-102.compute-1.amazonaws.com] has joined #shogun14:56
travis-ciit's Heiko Strathmann's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: https://travis-ci.org/shogun-toolbox/shogun/builds/38564334814:56
-!- travis-ci [~travis-ci@ec2-54-161-181-102.compute-1.amazonaws.com] has left #shogun []14:56
@wikingwuwei, i'm wondering what happens if you change this to CFeatures* https://github.com/vinx13/shogun/blob/d7d61d67aa21c1c12be3d101b19054515550a188/tests/unit/classifier/GaussianProcessClassification_unittest.cc#L15214:57
@wikingsame for features_test14:57
@wikingis there anything in that unit test that it actually needs that feature type? :)14:58
@wikingspecifically14:58
@wikingi would hope not14:58
@wikingthis would allow us to drop the bit ugly ->as<CDense...>() part :P15:01
wuweiin this case we can actually use CFeatures*, imo15:04
@wikingyeah that would be better15:04
@wikingcould you add those changes into that pr plz15:04
@wikingthnx heaps!15:04
@wikingHeikoS, here?15:05
wuweias for others, we should either cast to dense after transforming or cast to dense when get_feature_matrix cuz we need to compare the matrix in unit test15:06
@wikingwhere. ?:)15:06
@wikingplz copy me a link15:06
@wikingsorry my head is all over atm15:06
wuweifor example you can check tests for dense preproc, like log plus one15:07
-!- travis-ci [~travis-ci@ec2-54-161-208-147.compute-1.amazonaws.com] has joined #shogun15:07
travis-ciit's Heiko Strathmann's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: https://travis-ci.org/shogun-toolbox/shogun/builds/38564334815:07
-!- travis-ci [~travis-ci@ec2-54-161-208-147.compute-1.amazonaws.com] has left #shogun []15:07
@wikingjust a sec15:08
@wikinghehe15:09
@wikingDotIterator!15:09
@wiking:)15:09
wuweiyeah :)15:09
wuweicool!15:09
@wikingyou can either test with dot product :)15:10
@wikingor add15:10
@wiking:>15:10
-!- travis-ci [~travis-ci@ec2-54-161-208-147.compute-1.amazonaws.com] has joined #shogun15:16
travis-ciit's Shubham Shukla's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: https://travis-ci.org/shogun-toolbox/shogun/builds/38565304615:16
-!- travis-ci [~travis-ci@ec2-54-161-208-147.compute-1.amazonaws.com] has left #shogun []15:16
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/4316 synchronized by vinx1315:16
@wikingcool15:16
@wikinglets see CI but when it's ready i'll merge15:17
@wikingok?15:17
@wikingand then lets try to use dotiterator :)15:17
wuweiseems i broke integration test15:18
@wiking:D15:18
wuweiwe need update integration data15:18
wuweicuz 'processed' in features is dropped15:18
@wikingah ok15:20
@wikingsure thng15:20
@wikingjust prepare a patch for shogun-data15:20
wuweishould i send a pr to a new branch in shogun-data, otherwise that will break develop15:23
@wikingyeye15:26
@wikingnew pr to shogun-data15:26
wuweii'm wondering why there's only one integration test broken locally15:27
@wiking: ?\_(?)_/?15:28
@wikingsukey, flip15:28
@sukey?(????)? ???15:28
-!- Farouk [9cc2876f@gateway/web/freenode/ip.156.194.135.111] has joined #shogun15:35
@wikingFarouk, hellooooo15:35
@wikingwhatsup15:35
Faroukwiking hey15:35
@wikingFarouk, can you do a rebase over this https://github.com/shogun-toolbox/shogun/pull/430915:36
@wikingusing latest develop15:37
@wikingand then you should be able to use the new api everywhere15:37
@wikinghttps://git-scm.com/book/en/v2/Git-Branching-Rebasing15:37
@wikingfyi15:37
@wikingthnx15:37
@wikinglemme know if you are having troulbes15:37
@wiking*troubles15:37
FaroukYea, I saw Heiko's PR yesterday. was waiting for it to be merged. Just woke up and found it merged so will get to it now15:37
@wikingkk15:38
@wikingcool15:38
@wikingjust push when ready15:38
Farouksure (y)15:38
@wikingFarouk, what's with the other one15:38
@wiking?15:38
@wikingthe cost function15:38
@wikingwhat's the blocker there atm?15:38
@wikingor is there any blocker?15:38
FaroukThe cost function me and Fernando are meeting tomorrow at 10pm to discuss some design things15:38
FaroukCEST time15:38
Faroukso he told me to pause on it for now. But no blockers15:39
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/4314 merged by karlnapf15:45
@sukey[https://github.com/shogun-toolbox/shogun] karlnapf pushed 2 commits:15:45
@sukeyhttps://github.com/shogun-toolbox/shogun/commit/dac796c2f47ac2321483e488df096afd2067ad9315:45
@sukeyhttps://github.com/shogun-toolbox/shogun/commit/363c0dc853108c25c02d3743536f690bf27d344415:45
wuweiwiking: actually no problem with integration test :) i checked out the latest shogun-data and tests passed15:45
wuweilet's see how ci goes15:45
@wikingFarouk, if you want we can talk now15:46
@wikingabout that15:46
@wikingwuwei, ah cool15:46
Faroukwiking, yea sure we can talk now.15:47
@wikingshoot15:47
@wikingwhat's the design issue there atm15:48
@wiking?15:48
@wikingi mean fist of all15:48
@wikingwhat i wanna ask is15:48
@wikingwhy that may pointers?15:48
@wikingi.e. why everything is a pointer :)15:48
Faroukso Fernando said he has a couple of comments on the patterns I used (that some of them is anti-shogun) and also some comments on pointers :D15:48
Faroukhahaha thats one of the things Fernando commented on too :-D15:48
FaroukI am not sure, I guess I've mostly been using C-like syntax when it comes to these things15:49
Faroukand C has lots of pointers15:49
Faroukalso he wanted to talk about whether the cost function should hold the data (i.e. X, y)15:50
Faroukor we should supplement it15:50
@wikingfirst answer would be that the cost function shouldnt hold it but then again15:50
@wikingi have to see the context15:50
@wikingbut first about the pointers15:50
@wikinglemme open the pr15:50
@wikingand paste you here liens15:50
Farouksure15:50
@wikinghttps://github.com/shogun-toolbox/shogun/pull/4294/files#diff-09a10f92947e82f417b2b175687c3e6eR1915:52
@wikingdoes this link bring u to a line?15:52
Faroukyes15:53
@wikingso the point being is that StanVector* m_trainable_parameters15:53
@wikingwhy?15:53
@wikingdoes it have to be a pointer?15:53
@wikingwhy couldn't you just pass a ref ?15:53
@wikingMatrix<function<var(int32_t)>, Dynamic, 1>& cost_for_ith_point15:53
@wikingsame goes for the cost fucntion15:54
@wikingtotal_cost15:54
FaroukI agree with you. I think all of them should be by reference to be honest. I didn't realise that except when writing the unittest15:54
FaroukI wasn't creating any objects on the heap, so a reference would've been fine15:55
@sukey[https://github.com/shogun-toolbox/shogun] New commit https://github.com/shogun-toolbox/shogun/commit/4a1299e4b4ed6e05fe467ce063ebebe119e12681 by vigsterkr15:56
@wikingk15:57
@wikingso go with that change right away15:57
@wikingi'm just scrolling around again15:57
Faroukalright, and about the data? Should the data be provided when I want to calculate the cost/gradient, or should I keep a reference of the data?15:57
@wikingjust a sec16:00
@HeikoSFarouk: I commented on a few things in your PRs16:10
FaroukHey Heiko, you mean just now?16:10
FaroukOhh year just saw them16:11
Faroukwill finish cookbook and go there16:11
@wikingFarouk, ok so16:12
@wikingi would really not like to store data in such a method16:13
@wikingbut then again16:13
FaroukBut im not sure how I would keep track of the current sample that I am dealing with16:21
Faroukit would have to fall on the user to keep track of that16:21
wuweiwiking: I've made some initial implementation of pipeline, should I make the pr to a new branch, or the one for transformer16:27
@wikingwelll16:27
@wikingbecause we wanna use the transformer api16:27
@wikingit would make sense to use the transformer branch16:27
@wikingi know that the name of the branch is a bit mixing16:27
@wikingbut whatever16:27
@wikingjust go with that branch plz :)16:27
@wikingthnx16:27
@wikingwuwei, and great!16:28
wuweisure16:28
@wikingFarouk, REQUIRE(!X_new.equals(SGMatrix<float64_t>()), "Empty X provided"); :)16:32
@wikingso i know this is not the main issue16:32
@wikingbut if you wanna check whether an SGMatrix is emptuy16:32
@wikingjust call16:32
@wikingsgmatrix.size()16:32
@wikingthis one actually constructs an matrix16:32
@wikingobject etc etc16:32
@wikingFarouk, in general as fernando mentioned16:33
@wikingif you use the word Interface16:33
@wikingplz just really create an interface16:33
@wikingmeaning barely any implementation16:33
@wikingor just drop the Interface naming16:34
@wikingFarouk, btw the usecase of FirstOrderSAGCostFunctionInterface16:34
@wikingdo we really need the support16:35
@wikingfor chaning a cost function's member variables16:35
@wikingafter the object being created?16:35
@wikingso for example if you have a new train data16:35
@wikingwouldn't you actually create a new cost function16:35
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has quit [Ping timeout: 260 seconds]16:35
@wiking?16:35
@wikingsame goes for set_trainable_parameters16:36
@wikingetc?16:36
@wikingFarouk, see what i mean?16:40
@wikingbut again the best would be that actual;lyu16:41
@wikingyou draft me a code that uses this cost function16:41
@wikingin a 'real-life' setup16:41
@wikingthen we could a littlebit discuss this16:41
@wikingpseudo code is fine16:41
@wikingcurrently i'm just checking it from the unit test perspective16:41
Faroukwiking, sorry my irc was buggy, I didn't get a notification alert16:44
@wikingsure16:45
@wikingFarouk, you can use gitter.im16:45
@wikingthe messages are being relayed by shog16:45
@wikingshogitter,16:45
@wikinganyhow16:45
@wikingdo you see my comments/messages?16:45
Faroukyea reading them now16:45
@wikingkk16:45
FaroukHmm I agree, if a user wants to change the cost function, just create a new cost function, so maybe the setter of cost functions isn't really needed.16:46
@wikingyep16:46
@wikingbut yeah16:47
@wikingwould be great if you could give me a pseudo code16:47
@wikinghow you would imagine this now with the current api16:47
Faroukso the way I envisioned it to be used is similar to how its used in the unit test16:47
Faroukbasically you define X, y16:47
@wikingso say we have now a NN16:47
@wikingi guess this is your usecase right?16:47
Faroukthen write a function that defines your cost regardless16:47
Faroukyes16:47
@wikingok so lets have the big picutre now16:47
@wiking:)16:47
@wikinglet's have a pseudo code for nn16:48
@wikingcould you write this in a paastebin16:48
@wikingor whatever sharing tool16:48
@wikingso we can discuss it16:48
@wikingand maybe we refactor a bit the whole cost function story?16:48
Faroukyea sure, so you mean how the NN will use the cost function?16:48
Faroukin a big picture16:48
Farouk?16:48
@wikingyeah16:48
Farouksure16:49
@wikingi mean we have our current design of NN16:49
@wikingbest would be to use that in yoru examplke16:49
@wikingbut it's ok if it's all imaginary16:49
@wikingjsut wanna see how this would exactly play out16:49
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/4317 opened by vinx1316:49
@wikingwuwei, is on fire!16:50
@wiking\o/16:50
@wikingwuwei, ehhehe16:51
@wikingtrain_require_labels16:51
@wiking:>16:51
@wikingyeah the shitty design of train(data)16:51
@wikingbut not actually having train(data, label)16:51
@wiking:(16:51
wuweiwell16:51
-!- HeikoS [~heiko@eduroam-int-pat-8-0.ucl.ac.uk] has joined #shogun16:52
-!- mode/#shogun [+o HeikoS] by ChanServ16:52
@wikingwuwei, maybe tiiiiime for refactor? :D16:52
@wikinglol16:52
wuweiu mean fit + predict? :)16:53
@wikingyes :DDD16:53
@wikingbut maybe that's a bit heavy heavy :)16:53
@wikingbut yeah this is quite shitty :)16:53
@wikingtrain_require_labels16:54
@wikingi guess you agree :)16:54
wuweisure16:54
wuweiyeah16:54
wuweibut how can we know which one to call16:55
wuweifit(features*) or fit (features,labels)16:55
@wikingyeah16:55
@wikingcan we actually use type for this?16:56
@wikingi mean both preprocessors16:56
@wikingand machines has types16:56
@wikingEPreprocessorType16:56
@wikingall machines needs labels right?16:57
wuweilook up a table by type?16:58
wuweialmost16:58
@wikingmmmm clustering doesn16:58
wuweiexcept linear latent machine16:58
@wikingmmm16:58
@wikingvirtual bool train_require_labels() const { return false; }16:59
@wikingthis thing is around since when? :)16:59
@wikingok16:59
@wikingnevermind my comments16:59
@wiking:)))16:59
@wikingcool16:59
@wiking:D16:59
wuweii made it public16:59
@wikingsome minor stuff17:02
@wiking;)17:02
@wikingi've just sent i17:03
@wikingu17:03
@wikingbrb17:03
@wikingFarouk, i'll be here in 10, just paste me stuff when u hacce it17:03
@wiking*have17:03
FaroukSure wiking17:05
-!- Farouk [9cc2876f@gateway/web/freenode/ip.156.194.135.111] has quit [Ping timeout: 260 seconds]17:09
-!- travis-ci [~travis-ci@ec2-54-161-208-147.compute-1.amazonaws.com] has joined #shogun17:25
travis-ciit's Heiko Strathmann's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: https://travis-ci.org/shogun-toolbox/shogun/builds/38569890017:25
-!- travis-ci [~travis-ci@ec2-54-161-208-147.compute-1.amazonaws.com] has left #shogun []17:25
-!- HeikoS [~heiko@eduroam-int-pat-8-0.ucl.ac.uk] has quit [Ping timeout: 256 seconds]17:38
-shogun-buildbot:#shogun- Build deb4 - python3 #437 is complete: Success [build successful] - http://buildbot.shogun-toolbox.org:8080/#builders/28/builds/43717:45
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has joined #shogun17:53
-!- mode/#shogun [+o HeikoS] by ChanServ17:53
-!- TwofTp [67d51504@gateway/web/freenode/ip.103.213.21.4] has joined #shogun17:54
-!- witness_ [uid10044@gateway/web/irccloud.com/x-smarjtpaswqoslpr] has quit [Quit: Connection closed for inactivity]17:57
-!- TwofTp [67d51504@gateway/web/freenode/ip.103.213.21.4] has quit [Client Quit]17:58
-!- travis-ci [~travis-ci@ec2-54-161-208-147.compute-1.amazonaws.com] has joined #shogun18:07
travis-ciit's Viktor Gal's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: https://travis-ci.org/shogun-toolbox/shogun/builds/38570429518:07
-!- travis-ci [~travis-ci@ec2-54-161-208-147.compute-1.amazonaws.com] has left #shogun []18:07
@sukey[https://github.com/shogun-toolbox/shogun] New commit https://github.com/shogun-toolbox/shogun/commit/15624ba7941b174f6ba999012ca6154e95707d9c by vigsterkr18:09
-shogun-buildbot:#shogun- Build deb4 - python3 #438 is complete: Failure [failed python (failure)] - http://buildbot.shogun-toolbox.org:8080/#builders/28/builds/43818:35
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/4206 synchronized by FaroukY18:40
-!- Farouk [33267174@gateway/web/freenode/ip.51.38.113.116] has joined #shogun18:42
-!- HeikoS1 [~heiko@eduroam-int-pat-8-0.ucl.ac.uk] has joined #shogun19:01
-!- HeikoS1 [~heiko@eduroam-int-pat-8-0.ucl.ac.uk] has quit [Client Quit]19:01
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has quit [Ping timeout: 248 seconds]19:03
-!- travis-ci [~travis-ci@ec2-54-161-208-147.compute-1.amazonaws.com] has joined #shogun19:34
travis-ciit's Viktor Gal's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: https://travis-ci.org/shogun-toolbox/shogun/builds/38576844019:34
-!- travis-ci [~travis-ci@ec2-54-161-208-147.compute-1.amazonaws.com] has left #shogun []19:34
-!- Farouk [33267174@gateway/web/freenode/ip.51.38.113.116] has quit [Ping timeout: 260 seconds]19:42
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/4309 synchronized by FaroukY19:59
-!- Farouk [33267174@gateway/web/freenode/ip.51.38.113.116] has joined #shogun20:01
-!- travis-ci [~travis-ci@ec2-54-224-19-64.compute-1.amazonaws.com] has joined #shogun20:13
travis-ciit's Viktor Gal's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: https://travis-ci.org/shogun-toolbox/shogun/builds/38576844020:13
-!- travis-ci [~travis-ci@ec2-54-224-19-64.compute-1.amazonaws.com] has left #shogun []20:13
Faroukwiking, you here?20:59
@wikingye20:59
FaroukOhh, I have the pseudocode20:59
Farouksend it?20:59
@wikingye21:00
@wikingemail whatever is good21:00
Faroukhttps://pastebin.com/Yxf6sebY21:00
@wikingperfect21:00
@wikinglemme check21:00
@wikingok21:01
@wikingso the problem is here for me is kind of like21:01
@wikingthe adhoc dropins21:01
@wikingof the train features and labels21:02
@wikingmeaning21:02
@wikingi mean i'm now cirticising the whole not your part per se21:02
@wikingfirst of all how the NN and optimizer being connected?21:03
Faroukwhat do you mean by adhoc dropins?21:03
FaroukSo the optimizer will operate on the cost function, which has the **weights** of the NN21:03
Faroukso it's directly acting on the NN weights21:04
@wikingwhat i mean is21:04
@wikingthat some way i'd like to have this encapsulated within nn21:04
@wikingright?21:04
FaroukHmm, yea, but I was kind of thinking of it like the way in Tensorflow21:05
Faroukso in tf the backprogation is not part of the nn itself21:05
@wikingok21:05
@wikingi would do this compile time21:06
@wikingsorry21:06
@wikingin ctor21:06
@wikingline 21-2321:06
@wikingi mean the problem is that actually21:06
@wikingthis is an sgobject (FirstOrderSAGCostFunctionInterface)21:06
@wikingso if you would use this from modular (python etc)21:07
@wikingthen with the new api21:07
@wikingit would look like this21:07
@wikingloss.put("features", train_features)21:07
@wikingloss.put("labels", labels_test)21:07
@wikingetc21:07
@wikingwhich is super combersome :D21:07
@wiking*cumbersome21:08
@wikingbut yeah21:08
@wikingthere's gonna be the problem21:08
@wikingof21:08
@wikingith_cost_function and total_cost_function21:08
@wikingwhich is in a way impossible to be mapped into any interface atm21:08
@wikingbut lets keep that on ice for the time being21:08
FaroukOhh I see where you're coming with this. There isn't a particularly obvious way of definint the total_cost_function and ith_cost_function21:09
Farouklike using .put() ?21:09
Faroukadding them into the loss that is21:09
@wikingyeah21:11
@wikingbut ok lets put that in the back log21:11
FaroukAlright I'll keep that in mind21:11
Faroukare you joining us tomorrow?21:12
Faroukor is it too late?21:12
@wikingi mean one way or another one could say that21:12
FaroukI'll make sure to discuss it with Fernando21:12
@wikingit'd be great to have all this auto loss = some<FirstOrderSAGCostFunctionInterface>();21:12
@wikingin the ctor21:12
@wikingbecause i dont see the point changing these values21:12
@wikingafter construction21:12
@wikingbut yeah21:12
@wikingneed to think a bit21:13
FaroukHmmm yea I think it makes sense. training data shouldn't change, and cost function shouldn't (we just create a new one if we want)21:13
@wikingFarouk, when's the meeting tomo?21:36
@wikingFarouk, have you updated in this https://github.com/shogun-toolbox/shogun/pull/4206 the new data?21:37
@wikingas the integration test seems to be failing21:37
@wikingor idk if that needs change... but definitely the integration test seems to be failing :(21:39
Faroukwiking, sorry I am downloading an irc client, it keeps disconnecting/not alerting me so hopefully after I download the client it will be better22:39
Faroukthe meeting is at 10pm22:40
@wikingok22:40
@wikingwill try to make it22:40
@wikingsent u some feedbak22:40
FaroukYea, saw it and working on it22:40
FaroukI think this was before you told me that i need to use obtain_from_generic22:41
Faroukso I am changing it22:41
FaroukAlso regarding: labs? please rename labs to labels otherwise this will fail... as it fails now22:41
Faroukif I rename to labels, then if I do: Labels labels= labels("...")22:42
Faroukthen it causes an error22:42
Faroukthat I am using labels before using it, so all labels have to have a different name22:42
@wikingok22:44
@wikingthen use labs22:44
@wikingjust be consistent22:44
@wikingas this way its failing :)22:44
FaroukAlright22:45
-!- witness_ [uid10044@gateway/web/irccloud.com/x-koiqjutynwflbjuq] has joined #shogun22:57
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/4206 synchronized by FaroukY23:11
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/4309 synchronized by FaroukY23:19
-!- Farouk [33267174@gateway/web/freenode/ip.51.38.113.116] has quit [Quit: Page closed]23:35
--- Log closed Thu May 31 00:00:29 2018
« Tuesday, 2018-05-29 Index Thursday, 2018-05-31 »

Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!