--- Log opened Wed Aug 10 00:00:51 2011 | ||
-!- f-x [~user@117.192.210.94] has quit [Remote host closed the connection] | 03:23 | |
CIA-87 | shogun: Soeren Sonnenburg master * r9f7d9d7 / (11 files in 4 dirs): | 06:36 |
---|---|---|
CIA-87 | shogun: Merge pull request #277 from frx/streaming_vw | 06:36 |
CIA-87 | shogun: VW learners (+5 more commits...) - http://bit.ly/qiJ4Fh | 06:36 |
CIA-87 | shogun: Baozeng Ding master * rded1eb0 / (6 files): add some csharp examples - http://bit.ly/oP7sEb | 07:49 |
CIA-87 | shogun: Soeren Sonnenburg master * r7d75106 / (6 files): | 07:49 |
CIA-87 | shogun: Merge pull request #278 from sploving/master | 07:49 |
CIA-87 | shogun: add some csharp examples - http://bit.ly/p0n6H0 | 07:49 |
CIA-87 | shogun: Soeren Sonnenburg master * ra861dd4 / src/configure : remove illegal -Wimplicit from C++ compile flags - http://bit.ly/nm5IBM | 07:51 |
-!- blackburn [~blackburn@109.226.88.119] has joined #shogun | 09:42 | |
blackburn | sonney2k: thanks for the answer :) | 09:43 |
* blackburn got medicine from allergy and slept for ~13h long | 09:43 | |
-!- heiko [~heiko@134.91.55.45] has joined #shogun | 11:37 | |
-!- blackburn [~blackburn@109.226.88.119] has quit [Quit: Leaving.] | 12:01 | |
@sonney2k | :) | 12:15 |
-!- f-x [~user@117.192.208.105] has joined #shogun | 12:19 | |
heiko | sonney2k, are you there? | 12:35 |
@sonney2k | heiko, yes | 12:35 |
@sonney2k | heiko, how is it going? | 12:35 |
heiko | i am fine :) and in berlin? | 12:35 |
heiko | i have a question for loading parameters | 12:35 |
heiko | currently, calls the load method of a TParameter instance | 12:36 |
heiko | this instance knows its type and name | 12:36 |
heiko | but now i need to load ALL parameters of a file | 12:36 |
heiko | and I dont know the names/types of them | 12:36 |
heiko | is there a loading method that automatically loads all parameter data and the types of a file? | 12:37 |
heiko | or do I have to write it? | 12:37 |
heiko | sonney2, some parameters may have been deleted/renamed, so the TParameter instance of a class may be different than the on in the file | 12:38 |
heiko | sonney2k, do you understand what my problem is? | 12:39 |
@sonney2k | heiko, I think I understand - it is that the class registers different parameters (not the obsolete ones) | 12:39 |
heiko | yes | 12:39 |
@sonney2k | but I would try to solve the problem differently | 12:40 |
@sonney2k | couldn't we use this mapping thing to load obsolete parameters? | 12:40 |
@sonney2k | I mean we could register obsolete ones in some obsolete params thing | 12:40 |
@sonney2k | and load these maybe even into the mapping | 12:41 |
@sonney2k | heiko, am I making any sense? | 12:41 |
heiko | yes | 12:41 |
heiko | that might work | 12:41 |
heiko | good idea | 12:42 |
@sonney2k | this way you could just load all obsolete ones | 12:42 |
heiko | so we need the inverse mapping methods | 12:42 |
@sonney2k | and then in the mapping have the mapping + the obsolete value stored | 12:42 |
@sonney2k | inverse mapping? | 12:42 |
heiko | yes | 12:42 |
heiko | to get from current parameter version to obsolete one | 12:43 |
heiko | other way round than from obsolete to current | 12:43 |
@sonney2k | why that? | 12:43 |
@sonney2k | don't we always ever only need to obsolete -> current? | 12:43 |
heiko | yes | 12:43 |
@sonney2k | ahh but there is one problem | 12:43 |
heiko | but when we want to load, we only know the current one | 12:43 |
@sonney2k | when a type just changes from say int -> double | 12:44 |
heiko | yes? | 12:45 |
@sonney2k | how do we load this? | 12:46 |
@sonney2k | I mean it is then in obsolete *and* params | 12:46 |
@sonney2k | how do we know wheter 'variable' is int or double? | 12:47 |
@sonney2k | i.e. whether we call the obsolete or the params load routine | 12:47 |
heiko | mmh | 12:47 |
heiko | but i mean, | 12:47 |
heiko | we know the current is double | 12:48 |
heiko | then we check map for obsolete | 12:48 |
heiko | tells us sanem name and int | 12:48 |
heiko | we load int | 12:48 |
heiko | and call conversion method | 12:48 |
heiko | have double | 12:48 |
@sonney2k | ahh you mean because file version < version where double exists | 12:48 |
@sonney2k | ok | 12:48 |
heiko | yes | 12:49 |
@sonney2k | and for multiple type changes, like double -> float -> int? | 12:49 |
heiko | well, the inverse mapping is done until the file version is reached | 12:49 |
heiko | and then for each mapping, the conversion method is called | 12:49 |
@sonney2k | but how does the mapping look like? I mean we still have variable appearing 3 times with differnt types etc | 12:50 |
heiko | sonney2k, I dont understand | 12:51 |
@sonney2k | yeah ok if we would recursively traverse the mapping with ascending file version and then bottom up to the loading - it would work | 12:51 |
@sonney2k | heiko, the mapping currently contains varialbe_name, typestuff | 12:52 |
@sonney2k | -> variable_name, typestuff right? | 12:52 |
@sonney2k | and version | 12:52 |
heiko | yes | 12:52 |
heiko | version is not in there yet | 12:52 |
heiko | but ok | 12:52 |
heiko | lets assume | 12:52 |
@sonney2k | so if we have now 'v1',type1 -> 'v1', type2 and also 'v1',type2 -> 'v1', type3 | 12:53 |
heiko | yes | 12:53 |
@sonney2k | v1 (being a variable name) | 12:53 |
@sonney2k | we would need to load everything for v1 we could load that is <current_version into obsolete | 12:54 |
heiko | which one is the oldest one? | 12:55 |
CIA-87 | shogun: Shashwat Lal Das master * rab6daeb / src/shogun/classifier/vw/VwLearner.h : Added SG_ADD for VW learner base class. - http://bit.ly/qEDJTK | 12:56 |
CIA-87 | shogun: Soeren Sonnenburg master * rb80677d / src/shogun/classifier/vw/VwLearner.h : | 12:56 |
CIA-87 | shogun: Merge pull request #279 from frx/streaming_vw | 12:56 |
CIA-87 | shogun: Added SG_ADD for VW learner base class. - http://bit.ly/q6N4Qg | 12:56 |
@sonney2k | heiko, the one with typ1 | 12:56 |
@sonney2k | type1 | 12:56 |
heiko | ok then the class which wants to load is the one that contains v1, type3 | 12:57 |
heiko | in this class, the mapping is checked | 12:57 |
@sonney2k | yes | 12:57 |
heiko | returns v1, type2 , mapping is checked again | 12:57 |
heiko | returns v1, typ1 | 12:57 |
heiko | mapping is checked - nothing found | 12:57 |
heiko | so a TParameter is constructed and loaded with type1 | 12:57 |
heiko | and then, for every recursion step, the translation is called | 12:58 |
heiko | concrete: type1->type2 | 12:58 |
heiko | and afterwards type2->type3 | 12:58 |
@sonney2k | wait the translation can only be called after all variables are loaded | 12:58 |
heiko | yes | 12:59 |
@sonney2k | (not just 'v1') | 12:59 |
heiko | oh | 12:59 |
heiko | mmh true | 12:59 |
heiko | because it may depend on others | 12:59 |
@sonney2k | yes | 12:59 |
@sonney2k | this is not really a problem | 12:59 |
@sonney2k | I mean we could recursively load 'v1's into the respective mapping data ptr | 13:00 |
heiko | yes | 13:00 |
heiko | dont you think it works that way? | 13:00 |
@sonney2k | and then later on iterate over all mapping objects and call a free_data() | 13:01 |
@sonney2k | that will work | 13:01 |
@sonney2k | I am OK again :) | 13:01 |
heiko | alright :) | 13:01 |
heiko | implementation probably will be a bit more tricky ;) | 13:01 |
@sonney2k | that's some nasty beast | 13:01 |
@sonney2k | but it is super cool to have - I mean we can change whatever variable and still are able to load the old versions | 13:02 |
heiko | well ok, i will cut the best into smaller parts and see what will happen | 13:02 |
heiko | yes | 13:02 |
heiko | thats really a cool features | 13:02 |
@sonney2k | this also means we can significnatly simplify the testsuite | 13:02 |
@sonney2k | i.e. we can write out the objects once in python_modular | 13:02 |
@sonney2k | (as regression test reference) | 13:03 |
heiko | and use these as long as there is eniugh memory | 13:03 |
@sonney2k | then we can run things in e.g. java and write down the object, load the object from python (still within java) and write out the (updated!) object and compare | 13:03 |
@sonney2k | things should be 100% same | 13:03 |
heiko | oh | 13:04 |
heiko | yeah that is nice | 13:04 |
heiko | for testing all the language bindings | 13:04 |
@sonney2k | so we can do exact tests for all swig based languages :) | 13:04 |
heiko | in particular for numeric accuracy | 13:04 |
heiko | so, will create a test class now for translation stuff | 13:05 |
@sonney2k | heiko, apologies that I am screwing up your modelselection schedule but this really would be a very very cool feature... | 13:06 |
@sonney2k | heiko, btw what else did you intend to do modelselection wise afterwards? | 13:07 |
heiko | sonney2k, well ok, will do more modelselection stuff afterwards | 13:07 |
heiko | ah | 13:07 |
@sonney2k | heiko, I guess I can help you with the python typemaps for the syntactic sugar | 13:07 |
@sonney2k | (nicer way of specifying x-val params) | 13:08 |
heiko | yes | 13:08 |
heiko | that has to be done | 13:08 |
@sonney2k | what else is on your list? | 13:08 |
heiko | but more important first: the unimplemented methods for some more feature classes | 13:08 |
heiko | to make modelselectoin really work in more contexts | 13:08 |
heiko | perhaps some real life example will reveal important places where methods are missing | 13:09 |
heiko | and then: more search strategies | 13:09 |
heiko | this perhaps will cause some design changes in the model-selection class | 13:09 |
heiko | but ok | 13:09 |
@sonney2k | OK | 13:09 |
heiko | this coarse-fine grid-search would be cool | 13:09 |
heiko | at a first point because everyone uses it | 13:09 |
heiko | then, pausing grid-search and continue later on as you suggested | 13:10 |
heiko | some random sampling stuff perhaps --- | 13:10 |
@sonney2k | heiko, problem of course is that this becomes unrealisitic until Aug 22nd | 13:10 |
heiko | yes | 13:10 |
heiko | i know :) | 13:10 |
@sonney2k | I don't know if you will have time afterwards | 13:10 |
@sonney2k | would be great but I cannot force you in any way :D | 13:11 |
heiko | I will go to London in September, | 13:11 |
@bettyboo | <:*) | 13:11 |
heiko | but I will be studying, so there will probably be some time | 13:11 |
heiko | Well i really would like to continue in some matter :) | 13:11 |
@sonney2k | it would be great if you could still continue - even if much more slowly | 13:11 |
heiko | yes, I think so | 13:12 |
@sonney2k | ok heiko so lets prioritize | 13:12 |
heiko | September will be full - moving to London, start studying there, orientation stuff | 13:12 |
heiko | but afterwards will probably some space | 13:12 |
@sonney2k | sure | 13:12 |
@sonney2k | you finish this translation business | 13:12 |
heiko | yes | 13:13 |
heiko | I hope this will go straight from now | 13:13 |
@sonney2k | and then we should discuss about some syntactic sugar for python - I guess I can do the typemaps (with some guidance from you) | 13:13 |
heiko | ah, all the examples I wrote should be tranlsated to python | 13:13 |
heiko | yes ok | 13:13 |
@sonney2k | and you do the remaining stuff in feature classes and maybe 1 more strategy | 13:13 |
@sonney2k | I don't think more is realistic - there will be issues and documentation needs some update and... | 13:14 |
heiko | what about pencils down, writing documentation, clean up etc? | 13:14 |
@sonney2k | yes exactly | 13:14 |
heiko | thats still a lot of stuff | 13:14 |
@sonney2k | I think you can do the translation in 5 days? | 13:14 |
heiko | Hope so, hope to finish it until monday... | 13:15 |
@sonney2k | then 7 more days for feature classes and one modsel strategy | 13:15 |
@sonney2k | (or 5 more days) | 13:15 |
@sonney2k | ahh and docuemntation... | 13:15 |
heiko | yes | 13:15 |
heiko | i am a bit unsure as for the modsel strategy, perhaps the coarse-fine grid-search would be good, since its only a combination of the one already there, and its the most widely used one | 13:17 |
@sonney2k | heiko, it's ok - I mean we if other modsel strategies are not possible within these time limits this won't be the end of the world. There is still time to do these int the future | 13:17 |
heiko | yes true | 13:17 |
@sonney2k | the most important thing is that the framework itself is there | 13:18 |
heiko | yes | 13:18 |
@sonney2k | that was the most difficult part - and you did it already | 13:18 |
@sonney2k | so please continue with peace of mind :) | 13:18 |
heiko | I think it works pretty well - real life tests will probably reveal some bugs but ok, | 13:18 |
heiko | yes, good idea :) | 13:18 |
@bettyboo | <:*) | 13:19 |
heiko | sonney2k, is current git compiling for you? | 13:28 |
@sonney2k | heiko, where does it fail for you? | 13:28 |
heiko | math.h | 13:29 |
heiko | CMath.h | 13:29 |
heiko | ::log not defined for example | 13:29 |
@sonney2k | :: log not defined? | 13:31 |
@sonney2k | heiko, libshogun or which interface? | 13:31 |
f-x | an error popped up for me too, when i called the max() function (which should call the one in C's math.h) | 13:32 |
f-x | but now it compiles with libshogun at least | 13:32 |
heiko | yes, but I want to know if its my fault, so does it compile for you or not? | 13:32 |
heiko | mmh | 13:32 |
f-x | but surprisingly the code was working until yesterday | 13:33 |
heiko | sonney2k, it was my fault, nevermind | 13:36 |
f-x | hmm and mine too i guess | 13:37 |
@sonney2k | argh | 13:38 |
@sonney2k | I am still messing with this buildbot | 13:38 |
@sonney2k | while it build when manually triggered | 13:38 |
@sonney2k | it doesn't automagically | 13:38 |
@sonney2k | I don't understand why not... | 13:38 |
-!- heiko [~heiko@134.91.55.45] has quit [Ping timeout: 258 seconds] | 14:04 | |
CIA-87 | shogun: Shashwat Lal Das master * r46b44a0 / (5 files in 2 dirs): | 14:14 |
CIA-87 | shogun: Added VW native cache read/write classes. | 14:14 |
CIA-87 | shogun: Modified v_array to use shogun functions. - http://bit.ly/oCL8CM | 14:14 |
CIA-87 | shogun: Shashwat Lal Das master * r3bb29e2 / src/shogun/mathematics/Math.h : Made invsqrt inline. - http://bit.ly/q0w2sA | 14:14 |
CIA-87 | shogun: Soeren Sonnenburg master * rded0001 / (6 files in 3 dirs): | 14:14 |
CIA-87 | shogun: Merge pull request #280 from frx/streaming_vw | 14:14 |
CIA-87 | shogun: Added VW native cache read/write classes. - http://bit.ly/qxw7Ve | 14:14 |
serialhex | sonney2k: kick it! | 14:27 |
serialhex | just not too hard or you might hurt yourself | 14:27 |
@sonney2k | serialhex, I'd like to kick the buildbot hard enough to get it to work | 14:28 |
@sonney2k | but it is resisting all my attempts | 14:28 |
serialhex | :-/ | 14:28 |
serialhex | so you had a Q last night?? | 14:28 |
@sonney2k | serialhex, I wanted to know the ruby status | 14:29 |
@sonney2k | and get the configure stuff to work also at your machine | 14:29 |
serialhex | as far as the examples go i'm working on them but it seems that everything will work once i clean up those files... | 14:29 |
serialhex | that has been a pain and will probably need some 1-on-1 time with you or someone else who knows how to deal with shell scripting well | 14:30 |
serialhex | i know a bit but not a lot | 14:30 |
@sonney2k | so examples is something you will take care of right? | 14:30 |
@sonney2k | but the configure stuff is a pain | 14:31 |
serialhex | yes | 14:31 |
@sonney2k | so do you have your 'old' machine still running? | 14:31 |
@sonney2k | I mean if we have to run tests? | 14:31 |
serialhex | i actually installed debian just so i can test everything :) | 14:31 |
@sonney2k | I read that - good choice anyways ;) | 14:31 |
serialhex | that's what i'm on right now... everything else is suspended | 14:31 |
@sonney2k | ok then just ask if you run into trouble with the examples | 14:32 |
@sonney2k | btw, matrices/vector related stuff should work | 14:32 |
@sonney2k | I don't know about strings | 14:32 |
@sonney2k | serialhex, so better try to create some string example | 14:33 |
serialhex | i like it but i'd rather have my packages be more up to date... but that's the tradeoff with great stability | 14:33 |
@sonney2k | if it doesn't work we have to bug sploving about it | 14:33 |
serialhex | does one of the python examples use strings?? | 14:33 |
serialhex | i would magine they would... | 14:33 |
@sonney2k | e.g. ../examples/undocumented/python_modular/features_string_char_modular.py | 14:34 |
@sonney2k | just do some ls *string* :) | 14:34 |
serialhex | ok i'll make sure that works & if it dosn't bud sploving | 14:34 |
@sonney2k | dammed buildbot finished with succes but 3017 warnings | 14:35 |
serialhex | hmm... i get a bunch of warnings when i build myself... | 14:35 |
serialhex | but i dont know about 3k of em | 14:35 |
@sonney2k | f-x, that is in the recently added invsqrt function: | 14:36 |
@sonney2k | ../shogun/mathematics/Math.h:309:28: warning: dereferencing type-punned pointer will break strict-aliasing rules [-Wstrict-aliasing] | 14:36 |
f-x | one sec.. let me see | 14:36 |
f-x | sonney2k: i guess that's a hack they use.. should i look for an alternate implementation? | 14:37 |
@sonney2k | serialhex, the buildbot is running on debian-sid aka unstable - so it has the newest gcc and problably creates more warnings | 14:37 |
@sonney2k | f-x, well lets first try to understand the warning msg | 14:38 |
@sonney2k | at least I don't get it | 14:38 |
serialhex | anyway,i got to get off to work... have fun! | 14:38 |
f-x | converting the float ptr to int - does that always show up as an error like this? | 14:38 |
@sonney2k | serialhex, cu! | 14:38 |
@sonney2k | and enjoy | 14:38 |
@sonney2k | f-x, they are not converting the float ptr to an int but a float to an int | 14:39 |
f-x | sonney2k: yeah, indirectly by dereferencing the int ptr right? | 14:39 |
f-x | (i mean float ptr to int ptr) | 14:39 |
f-x | *i meant | 14:40 |
@sonney2k | f-x, I don't know - I cannot reproduce the problem in some small test script | 14:44 |
@sonney2k | gtg now | 14:45 |
@sonney2k | will dig into things later | 14:45 |
f-x | ok, later | 14:45 |
-!- heiko [~heiko@134.91.55.45] has joined #shogun | 14:49 | |
-!- alesis-novik [~alesis@188.74.87.206] has joined #shogun | 15:48 | |
-!- blackburn [~blackburn@188.122.238.50] has joined #shogun | 17:03 | |
blackburn | sonney2k: around? need some idea how to %ignore not-sg-object-classes | 17:08 |
serialhex | blackburn: magic! | 17:11 |
serialhex | that's how to do it!!! | 17:11 |
blackburn | serialhex: exactly it is but I can't do that magic yet ;) | 17:11 |
@bettyboo | ^_^ | 17:11 |
serialhex | aww! | 17:12 |
serialhex | so how ya been blackburn?? | 17:12 |
blackburn | serialhex: not bad, and you? | 17:12 |
serialhex | pretty good, pretty good | 17:12 |
serialhex | did you see my awesome python -> ruby trancekoder?? :P | 17:13 |
@bettyboo | ^_^ | 17:13 |
blackburn | serialhex: not yet, I've seen you done that | 17:13 |
blackburn | is it working now? | 17:14 |
serialhex | some of the examples work but most do not, fortunately it's done 90% of my work for me so it's just a matter of cleaning things up & making sure it works | 17:15 |
serialhex | have you made any new floating-face pictures?? :D | 17:15 |
blackburn | I've been working on something similar for java, but decided to stop | 17:15 |
blackburn | hahah nope | 17:16 |
serialhex | really?? what were ya working on?? | 17:16 |
blackburn | recently chris sent some sequence data to me | 17:16 |
blackburn | I'll try to make some fancy dimreduction for promoters | 17:16 |
blackburn | let's hope it will have any success haha | 17:16 |
serialhex | awesome! i hope it turns out great!! | 17:17 |
blackburn | the only problem is I really don't know if it is possible to get some euclidian mapping for sequences | 17:17 |
blackburn | cause some of them aren't metric-based | 17:17 |
blackburn | LLE, essentially | 17:18 |
serialhex | hmm, ok | 17:18 |
blackburn | well of course it uses distances but only in euclidean space | 17:18 |
blackburn | serialhex: how is my english? :D | 17:18 |
serialhex | infinitely better than my russian :D | 17:19 |
serialhex | it's great, i was actually thinking that you havn't made any mistakes that i noticed | 17:19 |
blackburn | hoho | 17:19 |
blackburn | nice to hear that | 17:19 |
blackburn | hmm may be I'll try to manage to get some sequence based LLE hehe | 17:20 |
blackburn | but at first I should get java serialization to work | 17:20 |
serialhex | hpefully yuou'll get a chance to come to the states or england or something and practice speaking it! | 17:20 |
blackburn | yeah I hope to leave this country too :D | 17:20 |
serialhex | (and i need to practice spelling :P ) | 17:20 |
serialhex | so what are you trying to do? serialize shogun objects in java so you can carry them around or soemthing?? | 17:21 |
blackburn | yeah so you can create ObjectOutputStream in java | 17:22 |
blackburn | and just writeObject(someShogunnyObject) | 17:22 |
blackburn | the only trouble I've got - there are objects without save_serializable | 17:23 |
blackburn | I guess I should remove it from modular interfaces somehow | 17:23 |
blackburn | and I'm not sure I will be able to do that with e.g. %ignore SGIO | 17:24 |
serialhex | hmm, i don't know much about that, so i'll be of no help really... but it would be nice to get that to work b/c then we could do the same for ruby (and a bunch of other langs too probably) | 17:25 |
blackburn | I'm sure it is possible | 17:25 |
-!- heiko [~heiko@134.91.55.45] has quit [Ping timeout: 258 seconds] | 17:26 | |
serialhex | this is awesome!!! http://xkcd.com/ :D | 17:26 |
blackburn | what are the methods used by ruby when some serialilzation, | 17:26 |
blackburn | ? | 17:26 |
blackburn | haha | 17:27 |
@bettyboo | :) | 17:27 |
serialhex | you can marshal an object... (why it's called that IDFK) i can't remember off hand what they are but i know they're there... i just havn't had the need to use them yet | 17:28 |
serialhex | yeah, awesome is it not? | 17:28 |
blackburn | exactly it is :) | 17:28 |
blackburn | marshalling is used in java too | 17:28 |
blackburn | when someone want to XMLize object | 17:29 |
blackburn | so in python _setstate_ or so is used | 17:29 |
blackburn | if you manage to write similar code for ruby | 17:29 |
blackburn | (just call save_serializable and read the temporary file) | 17:30 |
blackburn | it would be here for sure | 17:30 |
serialhex | yeah, ruby has an internal marshalling & we can also use YAML... though i'm sure we can XML and JSON too if we wanted too | 17:30 |
serialhex | though i don't know if they save certain state things - like singleton methods and the like - i can't remember at the moment | 17:31 |
blackburn | afaik for now we don't have any XML serialization | 17:33 |
serialhex | hmm... maybe someone can tackle that next summer?? :D | 17:35 |
@bettyboo | :) | 17:35 |
blackburn | I think not really big task to work on | 17:36 |
serialhex | so did they not have the meeting??? afaik there was supposed to be a meeting this am | 17:40 |
blackburn | yeah it was cancelled | 17:43 |
blackburn | okay ignore didn't help | 18:31 |
blackburn | meoow | 18:48 |
blackburn | meoooow | 18:52 |
blackburn | sonney2k: ignores work but ..Matrix, ..String?? | 19:35 |
@sonney2k | serialhex, blackburn we have xml serialization and json and some other ascii and hdf5 | 20:08 |
blackburn | wow | 20:08 |
blackburn | :D | 20:08 |
@bettyboo | *g* | 20:08 |
blackburn | didn't know | 20:08 |
blackburn | okay | 20:08 |
blackburn | sonney2k: what to do with template structures | 20:08 |
@sonney2k | ? | 20:09 |
blackburn | I see two ways: make it inherited from SGObject or implement save_serializable | 20:09 |
blackburn | sonney2k: I can't call save_serializable in e.g. ByteMatrix or so | 20:09 |
blackburn | java modular won't compile | 20:09 |
blackburn | I created modshogun_ignores.i | 20:10 |
blackburn | for ignoring Parallel, SGIO, etc | 20:10 |
blackburn | and I can't ignore *Matrix, *String, ... | 20:11 |
@sonney2k | blackburn, why not? | 20:11 |
blackburn | typemaps won't work | 20:11 |
@sonney2k | that's all weird... | 20:11 |
@sonney2k | then really the only option is to use %extends or what it is called for all the classes that don't have a save_serializable | 20:12 |
@sonney2k | and put a dummy method in there | 20:12 |
@sonney2k | for java only that is | 20:12 |
blackburn | but what about ignores? | 20:13 |
blackburn | should it stay? | 20:13 |
blackburn | it could speed up compilation | 20:14 |
@sonney2k | I thought ignores will work but it seems not... might be that ignores work for python though | 20:14 |
@sonney2k | so try some #ifdef SWIGPYTHON #endif around them | 20:14 |
@sonney2k | if it still compiles -> all good | 20:14 |
blackburn | will *not* work for python? | 20:14 |
blackburn | and by the way | 20:14 |
blackburn | I found one thing could be interesting | 20:14 |
blackburn | we should ignore operator= | 20:14 |
blackburn | and operator[] | 20:15 |
blackburn | == !=, etc | 20:15 |
@sonney2k | yes | 20:16 |
@sonney2k | blackburn, are you saying that %ignores will *not* work for python? | 20:16 |
blackburn | sonney2k: I don't know yet | 20:17 |
@sonney2k | blackburn, then just compile and run the examples :) | 20:17 |
blackburn | I mean why sohuld I #ifdef SWIGPYTHON them? | 20:17 |
CIA-87 | shogun: Heiko Strathmann master * r495e76b / examples/undocumented/libshogun/serialization_ascii.cpp : added serialization example - http://bit.ly/nTUSfv | 20:18 |
CIA-87 | shogun: Heiko Strathmann master * r0be1f49 / examples/undocumented/libshogun/Makefile : added serialization example - http://bit.ly/oOiWjg | 20:18 |
CIA-87 | shogun: Soeren Sonnenburg master * r0e23718 / (2 files): | 20:18 |
CIA-87 | shogun: Merge pull request #281 from karlnapf/master | 20:18 |
CIA-87 | shogun: serialization example - http://bit.ly/oir4k6 | 20:18 |
@sonney2k | blackburn, because if it works for python then the wrapper will be much smaller and code compilation faster | 20:18 |
blackburn | I know it works for java, I'll check if it works for python | 20:19 |
blackburn | sonney2k: can't understand how to use %extend to add dummy save_serializable | 20:20 |
CIA-87 | shogun: Soeren Sonnenburg master * r6d51367 / (3 files): | 20:20 |
CIA-87 | shogun: Merge pull request #282 from frx/streaming_vw | 20:20 |
CIA-87 | shogun: Added VW parser class. - http://bit.ly/o8Smr2 | 20:20 |
blackburn | I can %ignore all the classes and add dummy to every template structure class | 20:20 |
blackburn | works for you? | 20:20 |
@sonney2k | blackburn, look at http://www.swig.org/Doc1.3/SWIG.html and search for %extend Vector | 20:21 |
@sonney2k | then it becomes clear | 20:21 |
@sonney2k | blackburn, of course remove all %template for these then too | 20:21 |
CIA-87 | shogun: Soeren Sonnenburg master * rc8e5d1c / src/Makefile : remove lib/versionstring.h on distclean - http://bit.ly/qeGUra | 20:25 |
CIA-87 | shogun: Soeren Sonnenburg master * r10e7699 / (11 files in 5 dirs): Merge branch 'master' of github.com:shogun-toolbox/shogun - http://bit.ly/nTHetq | 20:25 |
blackburn | sonney2k: do you know if #extend might be used for every class in include? | 20:27 |
@sonney2k | %extend | 20:28 |
@sonney2k | just define a macro | 20:28 |
@sonney2k | and then do the magic | 20:28 |
blackburn | %include <shogun/lib/DataType.h> | 20:28 |
blackburn | %extend * | 20:28 |
blackburn | { | 20:28 |
blackburn | bool save_serializable(SerializableFile* file, const char* rw) {} ; | 20:28 |
blackburn | }; | 20:28 |
blackburn | that's how I did it | 20:28 |
@sonney2k | I don't think that this is allowed | 20:28 |
blackburn | soo should I %extend every class?? | 20:28 |
@sonney2k | besides it would clash with the other SGObject derived classes | 20:28 |
@sonney2k | yes | 20:29 |
blackburn | uuuuh | 20:29 |
@sonney2k | blackburn, well it is easy - that will be all classes without C prefix | 20:29 |
blackburn | but templates? | 20:29 |
@sonney2k | so just grep 'class ' in header file | 20:29 |
@sonney2k | blackburn, no difference | 20:29 |
@sonney2k | for templates you also only have to do it once | 20:29 |
blackburn | I mean can I write extend Class<T>? | 20:29 |
@sonney2k | I would expect so | 20:30 |
@sonney2k | and you can use macro like | 20:30 |
blackburn | okay will try | 20:30 |
@sonney2k | EXTEND_SERIALIZABLE(SGVector) | 20:30 |
@sonney2k | so you only have to write that list | 20:30 |
blackburn | okay | 20:31 |
CIA-87 | shogun: Soeren Sonnenburg master * r70476d8 / src/Makefile : | 20:33 |
CIA-87 | shogun: Revert "remove lib/versionstring.h on distclean" | 20:33 |
CIA-87 | shogun: This reverts commit c8e5d1c767cd3f08adacdc84e204ed116e02d2ca. | 20:33 |
CIA-87 | shogun: Releases contain lib/versionstring.h and this file is just once auto | 20:33 |
CIA-87 | shogun: generated before releases. So removing it in distclean will require to | 20:33 |
CIA-87 | shogun: unpack the tarball again -> not a good idea. - http://bit.ly/pVybuW | 20:33 |
blackburn | what are you doing? :D | 20:35 |
@bettyboo | <:*) | 20:36 |
@sonney2k | that @@$&%&*@%!! buildbot refuses to recognize the changes... | 20:36 |
blackburn | he wants to do nothing? | 20:38 |
@sonney2k | f-x, | 20:38 |
@sonney2k | classifier/vw/VwParser.cpp: In constructor 'shogun::CVwParser::CVwParser()': | 20:38 |
@sonney2k | classifier/vw/VwParser.cpp:25:18: error: invalid conversion from 'uint32_t (*)(shogun::substring, uint32_t) {aka unsigned int (*)(shogun::substring, unsigned int)}' to 'shogun::hash_func_t {aka long unsigned int (*)(shogun::substring, unsigned int)}' [-fpermissive] | 20:38 |
@sonney2k | classifier/vw/VwParser.cpp: In constructor 'shogun::CVwParser::CVwParser(shogun::CVwEnvironment*)': | 20:38 |
@sonney2k | classifier/vw/VwParser.cpp:36:18: error: invalid conversion from 'uint32_t (*)(shogun::substring, uint32_t) {aka unsigned int (*)(shogun::substring, unsigned int)}' to 'shogun::hash_func_t {aka long unsigned int (*)(shogun::substring, unsigned int)}' [-fpermissive] | 20:38 |
@sonney2k | current git doesn't compile | 20:38 |
@sonney2k | blackburn, yeah and I even updated to the latest version | 20:39 |
f-x | sonney2k: i compiled that on my computer and it did.. i'll check again | 20:39 |
@sonney2k | blackburn, I don't understand why gitpoller seems to detect a new version but then nothing happens... | 20:40 |
@sonney2k | f-x, you probably have a 32bit machine? | 20:40 |
f-x | hmm yes | 20:40 |
f-x | the size_t problem? | 20:40 |
blackburn | shit! | 20:41 |
f-x | sonney2k: ok i'll replace size_t by uint32_t in the typedef | 20:41 |
f-x | that should work | 20:41 |
blackburn | deleted modshogun_ignores with git clean -dfx | 20:41 |
blackburn | :DD | 20:41 |
f-x | sonney2k: i've replaced it and sent the pull request.. i think it should be ok | 20:46 |
f-x | and now i see what you mean by the use of size_t being a problem | 20:46 |
blackburn | sonney2k: are parsers needed in modular? | 20:52 |
blackburn | it seems there are no parser used in python_modular | 20:56 |
CIA-87 | shogun: Shashwat Lal Das master * r960471e / src/shogun/classifier/vw/vw_common.h : Replaced size_t with uint32_t in hash func typedef. - http://bit.ly/o1stLA | 20:57 |
CIA-87 | shogun: Soeren Sonnenburg master * rb75a971 / src/shogun/classifier/vw/vw_common.h : | 20:57 |
CIA-87 | shogun: Merge pull request #283 from frx/streaming_vw | 20:57 |
CIA-87 | shogun: Replaced size_t with uint32_t in hash func typedef. - http://bit.ly/nb9Xhx | 20:57 |
f-x | sonney2k: does it compile now? | 20:57 |
-!- heiko [~heiko@134.91.55.45] has joined #shogun | 20:58 | |
@sonney2k | we might know in 20 minutes | 21:00 |
f-x | 20 minutes?? oh wow | 21:01 |
@sonney2k | that is how long it takes to build for *all* interfaces | 21:02 |
@sonney2k | f-x, did compile | 21:23 |
@sonney2k | only 3042 warnings | 21:23 |
@sonney2k | o_O | 21:23 |
f-x | and that's exactly 20 minutes later | 21:23 |
f-x | sonney2k: how come the warnings don't show up with make? what kind of warnings are these? | 21:24 |
@sonney2k | they show up with make | 21:24 |
f-x | or maybe it's since i have only been compiling with libshogun | 21:24 |
@sonney2k | which g++ version do you have? | 21:24 |
f-x | 4.6.1 | 21:24 |
@sonney2k | gcc version 4.6.1 (Debian 4.6.1-6) | 21:25 |
@sonney2k | here too | 21:25 |
f-x | but i only do libshogun | 21:25 |
@sonney2k | on 64bit ... | 21:25 |
f-x | 32 here | 21:25 |
@sonney2k | f-x, http://cellperformance.beyond3d.com/articles/2006/06/understanding-strict-aliasing.html | 21:30 |
@sonney2k | this is the thing that bites us | 21:30 |
f-x | sonney2k: treating the value first as float and then dereferencing as int? | 21:31 |
@sonney2k | it is having two pointers of different type pointing to the same memory location | 21:31 |
@sonney2k | illegal with -fstrict-aliasing | 21:32 |
f-x | yeah | 21:32 |
f-x | sonney2k: so do you think the invsqrt has to be changed? | 21:37 |
f-x | or can we do -fno-strict-aliasing and get away with it? | 21:37 |
@sonney2k | in the document above it is recommended to use a union to do the casting | 21:37 |
@sonney2k | union { float x; int y } | 21:37 |
blackburn | but 3042 warnings? | 21:38 |
@sonney2k | then one could access x / y w/o errors and get the conversions | 21:38 |
blackburn | the whole shogun is a one big warning? | 21:38 |
@sonney2k | blackburn, yeah that thing generates only say 100-200 | 21:38 |
@sonney2k | blackburn, every function/variable that is not documented will generate a warning | 21:39 |
blackburn | ah I see | 21:39 |
f-x | sonney2k: what flags do you use to get these extra warnings? | 21:39 |
@sonney2k | so way to go for shogun 1.0 | 21:40 |
@sonney2k | f-x, $ cat x.cpp | 21:40 |
@sonney2k | int main() | 21:40 |
@sonney2k | { | 21:40 |
@sonney2k | volatile float x=17.0; | 21:40 |
@sonney2k | volatile int i = ((int*)((void*)&x))[0]; | 21:40 |
@sonney2k | } | 21:40 |
@sonney2k | g++ -O2 -Wall -Wstrict-aliasing -fstrict-aliasing x.cpp | 21:40 |
f-x | @1025: 1:10AM$ g++ -O2 -Wall -Wstrict-aliasing -fstrict-aliasing alias.cpp | 21:42 |
f-x | alias.cpp: In function ‘int main()’: | 21:42 |
f-x | alias.cpp:4:48: warning: dereferencing type-punned pointer will break strict-aliasing rules [-Wstrict-aliasing] | 21:42 |
f-x | alias.cpp:4:23: warning: unused variable ‘i’ [-Wunused-variable] | 21:42 |
f-x | 21:42 | |
blackburn | I guess we have to have some doc-marathon | 21:42 |
blackburn | sonney2k: good news - no dummy is required | 21:44 |
@sonney2k | blackburn, how so? | 21:44 |
blackburn | removing template solve the problem | 21:44 |
@sonney2k | like I said :) | 21:44 |
blackburn | sonney2k: much less classes to compile | 21:49 |
blackburn | I guess it will compile faster now | 21:49 |
CIA-87 | shogun: Soeren Sonnenburg master * r072dabc / src/shogun/lib/DataType.cpp : fix uninitialized variable warning - http://bit.ly/nELciy | 21:51 |
blackburn | sonney2k: how can I hide some struct from SWIG in .h? | 21:51 |
@sonney2k | blackburn, so check if it still compiles | 21:51 |
@sonney2k | would be great | 21:51 |
@sonney2k | blackburn, ?! | 21:51 |
blackburn | struct SSKDoubleFeature | 21:52 |
blackburn | { | 21:52 |
blackburn | int feature1; | 21:52 |
blackburn | int feature2; | 21:52 |
blackburn | int group; | 21:52 |
blackburn | }; | 21:52 |
@sonney2k | the buildbot builds :))))) | 21:52 |
blackburn | %ignore can't help with this thing | 21:52 |
@sonney2k | %ignore struct SSKDoubleFeature; doesn't work? | 21:52 |
blackburn | tried without struct | 21:52 |
blackburn | will try with struct | 21:52 |
blackburn | sonney2k: compiles with python_modular | 21:53 |
@sonney2k | blackburn, yeah but try if it compiles with java - could very well be that java is more picky | 21:56 |
blackburn | the only thing to solve now is to 'hide' QP and SSK*Feature | 21:56 |
CIA-87 | shogun: Soeren Sonnenburg master * r81af90d / src/shogun/lib/DataType.cpp : fix another g++ uninited warning - http://bit.ly/pm5SBD | 21:58 |
blackburn | sonney2k: so buildbot works? | 21:58 |
-!- heiko [~heiko@134.91.55.45] has quit [Quit: Leaving.] | 21:58 | |
@sonney2k | seems so | 22:06 |
@sonney2k | blackburn, given the current commit rate it will build non-stop all day long | 22:08 |
blackburn | sonney2k: hehe | 22:08 |
CIA-87 | shogun: Shashwat Lal Das master * rcfb3f07 / (2 files): Added StreamingVwFile. - http://bit.ly/r8szQ3 | 22:08 |
CIA-87 | shogun: Shashwat Lal Das master * r057b70b / (2 files): Modified StreamingFile to include VwExample read functions. - http://bit.ly/pLkKcb | 22:08 |
CIA-87 | shogun: Shashwat Lal Das master * rc59c21f / src/shogun/io/StreamingVwFile.cpp : Made the parser type switch statement handle cases better. - http://bit.ly/o8S4n5 | 22:08 |
CIA-87 | shogun: Shashwat Lal Das master * rbc3f108 / src/shogun/classifier/vw/VwParser.cpp : Improved another switch-case block in VwParser. - http://bit.ly/nasouX | 22:08 |
CIA-87 | shogun: Soeren Sonnenburg master * r47bedcc / (5 files in 2 dirs): | 22:09 |
CIA-87 | shogun: Merge pull request #284 from frx/streaming_vw | 22:09 |
CIA-87 | shogun: Added StreamingVwFile. - http://bit.ly/rmN9Wd | 22:09 |
blackburn | 3 classes to %ignore! | 22:09 |
-!- alesis-novik [~alesis@188.74.87.206] has quit [Ping timeout: 252 seconds] | 22:12 | |
@sonney2k | blackburn, if you get things to compile in <2G I can run 4 buildslaves | 22:20 |
@sonney2k | now I can only afford 2 | 22:20 |
blackburn | sonney2k: how to change tabs in vim :D | 22:21 |
@bettyboo | grin! 8) | 22:21 |
blackburn | I ran nerdtree | 22:21 |
blackburn | AAA | 22:21 |
@sonney2k | :set ts=4 | 22:22 |
blackburn | what is ts? | 22:22 |
@sonney2k | or did you mean tabsize | 22:22 |
blackburn | noo | 22:22 |
blackburn | windows or so | 22:22 |
@sonney2k | or set et | 22:23 |
blackburn | windows/tabs/.. | 22:23 |
@sonney2k | expandtab -> spaces | 22:23 |
@sonney2k | ahh | 22:23 |
@sonney2k | gT | 22:23 |
@sonney2k | : | 22:23 |
blackburn | didn't help hmm | 22:24 |
f-x | sonney2k: tiny question - do we have to add the Yahoo part of the copyright in every file linked to VW? | 22:24 |
f-x | i mean in StreamingVwFile etc where no actual code is taken from there | 22:24 |
@sonney2k | f-x, then not | 22:25 |
f-x | well ok then | 22:25 |
@sonney2k | f-x, only in the files that have code from yahoo *not just call functions from them* | 22:25 |
-!- alesis-novik [~alesis@vpn2-170.vpn.net.ed.ac.uk] has joined #shogun | 22:25 | |
f-x | hmm ok thanks.. that's more clear. | 22:25 |
blackburn | sonney2k: changing buffer is what I wanted | 22:28 |
blackburn | Ctrl+WW is pretty bad combination hehe | 22:28 |
@sonney2k | blackburn, hehe | 22:29 |
@sonney2k | f-x, I have a fix for the strict-aliasing thing | 22:29 |
f-x | great! how? | 22:29 |
blackburn | closed two tabs in my Guake heh | 22:29 |
@sonney2k | f-x, I am defining union float_to_int | 22:30 |
@sonney2k | { | 22:30 |
@sonney2k | float32_t f; | 22:30 |
@sonney2k | int32_t i; | 22:30 |
@sonney2k | }; | 22:30 |
@sonney2k | and then do float_to_int tmp; tmp.f=x; and then one can access tmp.i | 22:30 |
f-x | sonney2k: neat.. and the article says the compiler can optimize this pretty well too | 22:34 |
@sonney2k | f-x, ahh btw I thing I understand now why you don't see the warnings | 22:36 |
@sonney2k | one needs to compile with optimizations on | 22:36 |
@sonney2k | because only then stric-aliasing is turned on | 22:36 |
f-x | sonney2k: ahh.. there should be a -W flag which does this without -O too shouldn't there? | 22:37 |
@sonney2k | f-x, no you need to do -fstrict-aliasing and turn the warnings on | 22:37 |
f-x | ok, i'll add that to my configure command too | 22:38 |
CIA-87 | shogun: Soeren Sonnenburg master * r4eb1ae9 / src/shogun/mathematics/Math.h : use unions to fix warnings w/ strict-aliasing on - http://bit.ly/nLkXtw | 22:38 |
CIA-87 | shogun: Soeren Sonnenburg master * r4c44f6b / (5 files in 2 dirs): Merge branch 'master' of github.com:shogun-toolbox/shogun - http://bit.ly/rbpzvj | 22:38 |
@sonney2k | f-x, I committed the fix | 22:38 |
blackburn | wtf is QP | 22:39 |
@sonney2k | blackburn, quadratic program | 22:39 |
f-x | sonney2k: great, thanks.. | 22:39 |
blackburn | sonney2k: I can't find where it is declared | 22:39 |
@sonney2k | blackburn, do you mean the one in shogun/classifier/svm/SVMLight.h | 22:40 |
@sonney2k | ? | 22:40 |
blackburn | yeah | 22:41 |
blackburn | seems that | 22:41 |
@sonney2k | it is declared in place... | 22:41 |
@sonney2k | I hate this syntax | 22:41 |
blackburn | ignore doesn't work for this struct | 22:41 |
@sonney2k | blackburn, we don't need this to be a typedef | 22:42 |
CIA-87 | shogun: Soeren Sonnenburg master * r0796878 / (4 files in 4 dirs): fix further uninited warnings - http://bit.ly/pyaqpn | 22:42 |
blackburn | so how to ignore that? | 22:42 |
@sonney2k | remove the typedef; | 22:42 |
@sonney2k | alternative ignore quadratic_program and QP | 22:44 |
blackburn | I would not modify code | 22:44 |
@sonney2k | why not? | 22:44 |
@sonney2k | with heikos soon to come ueber serilaization stuff we will be able to do testing throughout all modular interfaces | 22:45 |
@sonney2k | blackburn, that will solve our test suite issues | 22:45 |
blackburn | nice | 22:45 |
@sonney2k | blackburn, so we can use the tests from python_modular | 22:45 |
@sonney2k | and just load these serilazed objects in other languages and compare | 22:46 |
blackburn | java modular should work now | 22:46 |
@sonney2k | with you %ignores'? | 22:46 |
blackburn | yes | 22:47 |
blackburn | I did some code for writeObject | 22:47 |
@sonney2k | and the other languages too? | 22:47 |
blackburn | for python too | 22:47 |
blackburn | don't know about any other | 22:47 |
@sonney2k | the buildbot will tell | 22:47 |
blackburn | may be it could be faster to test it on buildbot | 22:47 |
@sonney2k | how big are the wrappers now? | 22:47 |
blackburn | .so? | 22:47 |
@sonney2k | no *_wrap.cxx | 22:48 |
blackburn | I'll let you know in a minute or so | 22:49 |
@sonney2k | I think I should set up ccache for the buildbot | 22:49 |
@sonney2k | it is too slow otherwise if there are just minimal changes | 22:49 |
CIA-87 | shogun: Shashwat Lal Das master * rc73dbc8 / (2 files): Added StreamingVwCacheFile. - http://bit.ly/ot3DPh | 22:49 |
CIA-87 | shogun: Soeren Sonnenburg master * r7cca7ff / (2 files): | 22:49 |
CIA-87 | shogun: Merge pull request #285 from frx/streaming_vw | 22:49 |
CIA-87 | shogun: Added StreamingVwCacheFile. - http://bit.ly/r5MtKN | 22:49 |
blackburn | hmm took ~600mb on my machine | 22:50 |
@sonney2k | blackburn, the good news is that it can send out emails to people complaining about code breakage | 22:50 |
@sonney2k | blackburn, yay! | 22:50 |
@sonney2k | that would be great? | 22:50 |
@sonney2k | do the scripts all still work? | 22:50 |
@sonney2k | blackburn, and how big are the *wrap* files? | 22:50 |
blackburn | sonney2k: remind me a command to get size of file? | 22:51 |
@sonney2k | ls -lh | 22:51 |
blackburn | 5.3M 2011-08-11 00:49 interfaces/java_modular/modshogun_wrap.cxx | 22:51 |
blackburn | java | 22:51 |
blackburn | I'll check python now | 22:51 |
@sonney2k | it was 41M here | 22:52 |
blackburn | wow why it changed so much? | 22:52 |
@sonney2k | useless wrappers removed | 22:52 |
blackburn | I should check if all examples work in python | 22:52 |
@sonney2k | but blackburn please check if python modular examples still work | 22:53 |
blackburn | I will | 22:53 |
@sonney2k | otherwise this might be useless | 22:53 |
blackburn | no, should work | 22:54 |
@sonney2k | blackburn, that would be soooo great | 22:56 |
@sonney2k | then the buildbot here could use all 4 cores | 22:56 |
blackburn | 12M 2011-08-11 00:55 interfaces/python_modular/modshogun_wrap.cxx | 22:57 |
@sonney2k | the cpu is still at 29 C even when building - so just parallelizing things would be super cool | 22:57 |
blackburn | sonney2k: no, not working :( | 22:58 |
@sonney2k | blackburn, what happens? | 22:58 |
blackburn | NotImplementedError: Wrong number of arguments for overloaded function 'new_RealFeatu | 22:58 |
@sonney2k | blackburn, but java did work? | 22:59 |
blackburn | examples? don't know | 22:59 |
@sonney2k | of course examples... | 22:59 |
blackburn | I checked only compilation | 22:59 |
@sonney2k | compiling usually works | 22:59 |
@sonney2k | too bad | 22:59 |
blackburn | sonney2k: the problem is all the %template things are used in python typemaps | 23:01 |
blackburn | hmm no | 23:02 |
@sonney2k | I don't understand | 23:02 |
blackburn | nevermind | 23:02 |
blackburn | I'll try to get it work in python | 23:03 |
blackburn | restored template things for matrices and vectors | 23:03 |
@sonney2k | yeah it could be that it will work when these are there | 23:04 |
@sonney2k | that shouldn't be too much bloat | 23:04 |
blackburn | but I've ignored not too much | 23:05 |
blackburn | ExamplesRing | 23:05 |
blackburn | Parser | 23:05 |
blackburn | both sparse and * | 23:05 |
blackburn | and some ~15 ignores | 23:05 |
blackburn | and operators[], etc | 23:06 |
f-x | blackburn: i think the ExamplesRing and Parser can be safely ignored - one doesn't need to call any of their functions explicitly | 23:08 |
f-x | (if that's what %ignore is for, i don't know) | 23:08 |
@sonney2k | f-x, %ignore stuff is for hiding them from the modular interfaces | 23:08 |
blackburn | f-x: well so I just removed %template things | 23:08 |
@sonney2k | like python /java etc | 23:08 |
blackburn | sonney2k: OK now | 23:09 |
blackburn | ehhh | 23:09 |
blackburn | some segfaults?? | 23:09 |
@sonney2k | blackburn, wrapper size? | 23:09 |
blackburn | sonney2k: 15m | 23:09 |
@sonney2k | blackburn, no idea if it did segfault before | 23:10 |
blackburn | sonney2k: please run classifier_svmsgd_modular.py | 23:10 |
@sonney2k | we didn't have a test / buildbot before ... | 23:10 |
blackburn | and classifier_subgradientsvm_modular.py | 23:10 |
@sonney2k | blackburn, crashes with double free | 23:11 |
blackburn | boom | 23:11 |
blackburn | okay except this everything is working | 23:11 |
@sonney2k | blackburn, then better try all java_modular examples | 23:11 |
f-x | sonney2k: where is the double free? the only major change was the introduction of the LossFunction class in SVMSGD | 23:11 |
blackburn | sonney2k: will try to remove DynamicArray things | 23:12 |
@sonney2k | f-x, shogun::SGSparseVector<double>::free_vector() (DataType.h:340) | 23:13 |
f-x | sonney2k: ok.. but it doesn't make sense if it was working until now | 23:14 |
@sonney2k | f-x, very likely it wasn't working for some time | 23:15 |
blackburn | uh it was wrong to ignore parallel | 23:19 |
@sonney2k | we now have twice the build slaves :) | 23:28 |
blackburn | restored parallel | 23:28 |
blackburn | trying without svmsgd | 23:28 |
blackburn | and subgradient | 23:28 |
blackburn | uhoh | 23:28 |
blackburn | distance_sparseeuclidean_modular.py | 23:29 |
blackburn | svmlin | 23:29 |
blackburn | svmocas | 23:29 |
@sonney2k | so everything that uses sparse | 23:29 |
@sonney2k | easy to fix then :) | 23:29 |
blackburn | running features_string_sliding_window_modular.py .. ERROR | 23:29 |
blackburn | running features_io_modular.py .. ERROR | 23:29 |
blackburn | worried about io | 23:29 |
blackburn | hope I didn't cause it | 23:30 |
blackburn | ah | 23:30 |
blackburn | HDF5 | 23:30 |
blackburn | okat | 23:30 |
blackburn | okay | 23:30 |
blackburn | sonney2k: could you please fix that sparse bug? | 23:30 |
@sonney2k | blackburn, the fix is not that easy though... more involved than I thought. hmm | 23:39 |
blackburn | okay | 23:39 |
blackburn | I'll try to write extends for data types that should stay | 23:40 |
@sonney2k | yeah! | 23:41 |
@sonney2k | we are down to 2543 warnings! | 23:41 |
@sonney2k | and only about 50 of them are not 'documentation missing' | 23:42 |
@sonney2k | blackburn, don't forget to use a macro for that | 23:43 |
blackburn | sonney2k: yeah currently writing it | 23:44 |
@sonney2k | kk | 23:44 |
--- Log closed Thu Aug 11 00:00:59 2011 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!