--- Log opened Thu Mar 22 00:00:19 2012 | ||
-!- romi_ [~mizobe@187.57.1.50] has quit [Ping timeout: 246 seconds] | 00:33 | |
-!- romi_ [~mizobe@187.57.1.50] has joined #shogun | 00:45 | |
-!- romi_ [~mizobe@187.57.1.50] has quit [Ping timeout: 246 seconds] | 00:52 | |
n4nd0 | sonney2k: I just sent a pull request for QDA but appears to me like none has been assigned to it, can you tell me if you have been notified? | 00:53 |
---|---|---|
-!- vikram360 [~vikram360@117.192.164.158] has joined #shogun | 01:15 | |
-!- PhilTillet [~Philippe@tillet-p42154.maisel.int-evry.fr] has quit [Remote host closed the connection] | 01:40 | |
-!- mayanks43 [~Marcus@115.248.130.148] has joined #shogun | 01:57 | |
-!- l0nr4n [~l0nr4n@unaffiliated/l0nr4n] has quit [Quit: This computer (or maybe me) has gone to sleep] | 02:06 | |
-!- canonind [d2198538@gateway/web/freenode/ip.210.25.133.56] has joined #shogun | 02:15 | |
-!- canonind [d2198538@gateway/web/freenode/ip.210.25.133.56] has quit [Quit: Page closed] | 02:33 | |
-!- canonind [d2198538@gateway/web/freenode/ip.210.25.133.56] has joined #shogun | 02:33 | |
-!- canonind [d2198538@gateway/web/freenode/ip.210.25.133.56] has quit [Client Quit] | 02:34 | |
-!- Canonind [d2198538@gateway/web/freenode/ip.210.25.133.56] has joined #shogun | 02:35 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 02:50 | |
-!- vikram360 [~vikram360@117.192.164.158] has quit [Ping timeout: 252 seconds] | 03:20 | |
-!- vikram360 [~vikram360@117.192.164.158] has joined #shogun | 03:23 | |
-!- vikram360 [~vikram360@117.192.164.158] has quit [Ping timeout: 252 seconds] | 03:35 | |
-!- yash [46ab0416@gateway/web/freenode/ip.70.171.4.22] has joined #shogun | 04:07 | |
-!- romi_ [~mizobe@187.57.1.50] has joined #shogun | 04:43 | |
-!- yash [46ab0416@gateway/web/freenode/ip.70.171.4.22] has quit [Quit: Page closed] | 04:43 | |
-!- Miggy [~piggy@14.139.82.6] has quit [Ping timeout: 260 seconds] | 04:56 | |
-!- mayanks43 [~Marcus@115.248.130.148] has left #shogun [] | 05:26 | |
-!- harshit_ [~harshit@182.68.142.173] has joined #shogun | 05:30 | |
-!- in3xes [~in3xes@180.149.49.227] has joined #shogun | 06:10 | |
-!- in3xes [~in3xes@180.149.49.227] has quit [Ping timeout: 244 seconds] | 06:32 | |
-!- in3xes [~in3xes@210.212.58.111] has joined #shogun | 06:45 | |
-!- harshit_ [~harshit@182.68.142.173] has quit [Ping timeout: 252 seconds] | 07:11 | |
-!- gsomix [~gsomix@85.26.232.68] has joined #shogun | 07:23 | |
gsomix | hi | 07:23 |
-!- harshit_ [~harshit@182.68.142.173] has joined #shogun | 07:24 | |
-!- in3xes_ [~in3xes@59.163.196.121] has joined #shogun | 07:27 | |
-!- in3xes_ [~in3xes@59.163.196.121] has quit [Client Quit] | 07:27 | |
-!- in3xes_ [~in3xes@59.163.196.121] has joined #shogun | 07:28 | |
-!- in3xes__ [~in3xes@59.163.196.121] has joined #shogun | 07:37 | |
-!- in3xes__ [~in3xes@59.163.196.121] has quit [Remote host closed the connection] | 07:40 | |
-!- in3xes [~in3xes@210.212.58.111] has quit [Ping timeout: 276 seconds] | 07:41 | |
-!- in3xes_ [~in3xes@59.163.196.121] has quit [Ping timeout: 265 seconds] | 07:41 | |
-!- gsomix [~gsomix@85.26.232.68] has quit [Ping timeout: 246 seconds] | 07:47 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 07:49 | |
harshit_ | hey @n4nd0 | 07:52 |
n4nd0 | harshit_: hi there! | 07:52 |
n4nd0 | harshit_: how is your stuff going? | 07:53 |
harshit_ | one more problem now | 07:53 |
harshit_ | my ./configure dont recognize lapack library | 07:53 |
harshit_ | i have installed it and also its c interface | 07:54 |
n4nd0 | ok | 07:54 |
n4nd0 | so how have you installed it? | 07:54 |
harshit_ | by making some changes in make.inc file | 07:55 |
harshit_ | and then installing fortran library | 07:55 |
harshit_ | and then installed the lapacke | 07:55 |
n4nd0 | mmm | 07:55 |
n4nd0 | what OS are you with? | 07:55 |
harshit_ | ubuntu 11.10 | 07:55 |
n4nd0 | apt-get install liblapack-dev? | 07:56 |
n4nd0 | I normally install everything with apt-get | 07:56 |
n4nd0 | I think I did this also this way | 07:56 |
harshit_ | okay wait i am doing that | 07:57 |
harshit_ | Is lapack example in shogun's examples, Running on your machine | 07:58 |
harshit_ | ? | 07:58 |
n4nd0 | which example in particular? | 07:59 |
harshit_ | mathematics_lapack.cpp | 07:59 |
n4nd0 | compiling ... | 08:00 |
harshit_ | hey thanks ./configure now recognized the lapack | 08:00 |
n4nd0 | yes, it looks like it runs without problems | 08:00 |
harshit_ | thanks apt-get is great | 08:01 |
n4nd0 | yes | 08:01 |
n4nd0 | if you are fond of more graphical stuff I recommend you synaptic | 08:01 |
n4nd0 | apt-get is good enough though | 08:01 |
harshit_ | also one more thing if say suppose i didht have any library installed at time i installed shogun | 08:02 |
n4nd0 | the only problem is that you need to know the name of the package to install / have intuition for the naming conventation | 08:02 |
n4nd0 | conventions* | 08:02 |
harshit_ | and then later i install that library | 08:02 |
harshit_ | then do i need to recompile whole shogun ? | 08:02 |
n4nd0 | yes, probably | 08:03 |
n4nd0 | I think there's no general answer for that (there may be some exceptions) but in general yes | 08:03 |
n4nd0 | to build shogun faster I recommend you to do | 08:04 |
n4nd0 | ./configure --disable-optimization [the other options you want to use] | 08:04 |
n4nd0 | and later run make for several threads | 08:04 |
n4nd0 | make -j2 or make -j4 | 08:04 |
harshit_ | that would be useful sometimes | 08:04 |
n4nd0 | yes | 08:05 |
-!- Canonind [d2198538@gateway/web/freenode/ip.210.25.133.56] has left #shogun [] | 08:07 | |
harshit_ | hey, my shogun compiled and is installed successfully but still when i run mathematics_lapack.cpp it doesn't show anything | 08:08 |
-!- Jiuding [d2198538@gateway/web/freenode/ip.210.25.133.56] has joined #shogun | 08:08 | |
Jiuding | hi feeling great to join you | 08:09 |
n4nd0 | harshit_: well, have you checked the code? | 08:10 |
harshit_ | yes .. in my case HAVE_LAPACK is not defined | 08:10 |
harshit_ | so ifdef evaluates to be false | 08:11 |
harshit_ | and doesnt show anything | 08:11 |
harshit_ | and HAVE_LAPACK is supposed to be defined in config.h | 08:11 |
n4nd0 | I think that there's no nothing to show in that example apart from error messages | 08:11 |
-!- gsomix [~gsomix@85.26.232.68] has joined #shogun | 08:12 | |
n4nd0 | there's nothing* ... fuck, my fingers are clumsy in the morning | 08:12 |
harshit_ | but still i dont knw why my machine is doing that | 08:14 |
harshit_ | i think i should redo the entire procedure | 08:15 |
n4nd0 | how do you know that HAVE_LAPACK is evaluated to false? | 08:16 |
n4nd0 | have you tried it writing some printfs inside? | 08:16 |
harshit_ | becoz if i printf inside it doesnt print | 08:16 |
harshit_ | yes | 08:16 |
n4nd0 | ok | 08:16 |
n4nd0 | have you built shogun after installing lapack? | 08:17 |
harshit_ | yes | 08:17 |
n4nd0 | you should run configure, check that lapack is detected | 08:17 |
n4nd0 | run make, and finally make install | 08:17 |
n4nd0 | have you done all the steps? | 08:17 |
harshit_ | yes after you told me to install via apt-get | 08:17 |
harshit_ | it was detected by configure | 08:17 |
n4nd0 | so if you did later make and make install I guess it should work | 08:18 |
n4nd0 | however, it might be that there are other dependencies in order for lapack to work, I don't come up with other explanation | 08:18 |
harshit_ | yes i did | 08:18 |
harshit_ | i think only way out is to reinstall | 08:19 |
harshit_ | no ? | 08:19 |
harshit_ | reinstall everything | 08:19 |
harshit_ | also do you knw how to uninstall any library | 08:19 |
n4nd0 | mmm | 08:20 |
n4nd0 | I don't know if I recommend you to do that, one should be careful | 08:20 |
n4nd0 | yes, you can uninstall libraries with apt-get remove | 08:21 |
n4nd0 | easier with synaptic | 08:21 |
n4nd0 | I'd recommend you to take a look here first http://www.shogun-toolbox.org/doc/en/current/installation.html | 08:21 |
n4nd0 | in the requirements, it might be that you're still missing sth | 08:22 |
harshit_ | what is sth ? | 08:22 |
n4nd0 | it looks like that atlas library could be necessary | 08:22 |
n4nd0 | ups sorry, sth = something | 08:23 |
n4nd0 | :) | 08:23 |
harshit_ | i thought sth is some program | 08:24 |
harshit_ | also i tried to apt-get it :) | 08:24 |
n4nd0 | haha sorry | 08:24 |
harshit_ | okay thanks n4nd0 ,g2g will do it and | 08:25 |
harshit_ | will ask you some more questions .lol | 08:25 |
n4nd0 | apt-cache show is useful to read package descriptions before installation | 08:26 |
n4nd0 | try if you want for example | 08:26 |
n4nd0 | apt-cache show liblapack-dev | 08:26 |
n4nd0 | you'll see the description of the package you just installed | 08:26 |
harshit_ | yes it works. nice | 08:27 |
harshit_ | and also when we download documentation through synaptic, how can we access that | 08:28 |
-!- gsomix [~gsomix@85.26.232.68] has quit [Ping timeout: 246 seconds] | 08:31 | |
n4nd0 | afaik the package documentation is usually man pages | 08:32 |
n4nd0 | e.g. now you should be able to do man lapack | 08:32 |
n4nd0 | if you want to know more info about a topic apropos | 08:32 |
harshit_ | but when i do man lapack it says : | 08:33 |
harshit_ | No manual entry for lapack | 08:33 |
harshit_ | same for every other documentation i download | 08:33 |
n4nd0 | probably you need the package liblapack-doc | 08:33 |
harshit_ | yeah get it.. | 08:36 |
harshit_ | thanks | 08:36 |
harshit_ | bye | 08:36 |
-!- harshit_ [~harshit@182.68.142.173] has quit [Quit: Leaving] | 08:37 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 08:55 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 09:37 | |
-!- blackburn [5bdfb203@gateway/web/freenode/ip.91.223.178.3] has joined #shogun | 09:43 | |
blackburn | n4nd0: yes we were notified | 09:44 |
n4nd0 | blackburn: cool :) | 09:45 |
blackburn | sonne|work: ping | 09:49 |
blackburn | This program is SG_FREE software | 09:56 |
blackburn | oh lol auto replace | 09:56 |
n4nd0 | haha | 09:56 |
blackburn | n4nd0: so did you fix this mess in your repo? | 10:05 |
n4nd0 | blackburn: yeah, it is looking way better now :) | 10:05 |
n4nd0 | blackburn: I have to ask you one thing about one vs. one training, not repo related | 10:05 |
blackburn | n4nd0: just create new branches for each new pull | 10:06 |
blackburn | and use rebase not merge | 10:06 |
blackburn | n4nd0: ask | 10:06 |
n4nd0 | blackburn: I didn't manage to get to work the trick using the same vector all the time, I don't know if you remember | 10:07 |
n4nd0 | blackburn: the thing was to allocate an array big enough and pass to CSubset on each iteration the adequate lenght of this array | 10:07 |
n4nd0 | blackburn: probably not the complete one | 10:07 |
n4nd0 | blackburn: the thing is that later, when one does remove subset both in feature and labels, the vector is also de-allocated so in the next second iteration it explodes | 10:08 |
n4nd0 | blackburn: let me show you in code | 10:08 |
n4nd0 | https://github.com/iglesias/shogun/blob/multiclass_1vs1/src/shogun/machine/MulticlassMachine.cpp | 10:10 |
n4nd0 | the nice idea would be to have lines 147 and 148 out of the for loop | 10:10 |
blackburn | n4nd0: let it stay for now | 10:12 |
blackburn | not really high demanding thing | 10:12 |
n4nd0 | blackburn: ok :) | 10:12 |
n4nd0 | blackburn: about the repo, I read about rebase yesterday and I think I got the difference wrt merge | 10:13 |
n4nd0 | blackburn: but why do you think it is better to do it that way? | 10:13 |
blackburn | n4nd0: yeah you just put your commits on top of it | 10:13 |
blackburn | no 'merged blabla from blabla' commits | 10:13 |
n4nd0 | ok | 10:14 |
n4nd0 | but in any case I should not worry about merging my branches into master when doing a pull request | 10:14 |
blackburn | n4nd0: the approach is to keep master updated and rebase your branches on top of it | 10:15 |
blackburn | in this way everything looks clearer | 10:16 |
blackburn | http://book.git-scm.com/assets/images/figure/rebase5.png | 10:16 |
blackburn | http://book.git-scm.com/assets/images/figure/rebase5.png | 10:16 |
blackburn | ehm | 10:16 |
blackburn | one picture heh | 10:16 |
blackburn | n4nd0: commit graph would not look like some social graph :D | 10:17 |
n4nd0 | hehe ok :) | 10:17 |
blackburn | n4nd0: train 1-v-1 looks ready | 10:21 |
blackburn | what u'd need now is insert classify 1-v-1 from multiclasssvm | 10:21 |
-!- muddo [~muddo@gateway/tor-sasl/muddo] has quit [Ping timeout: 276 seconds] | 10:25 | |
n4nd0 | blackburn: yes | 10:26 |
n4nd0 | blackburn: the simple example I created with LibLinear was working at least | 10:27 |
-!- muddo [~muddo@gateway/tor-sasl/muddo] has joined #shogun | 10:27 | |
blackburn | n4nd0: you mean it trained successfully? | 10:27 |
n4nd0 | blackburn: yes | 10:27 |
blackburn | nicer | 10:28 |
blackburn | nice* | 10:28 |
n4nd0 | blackburn: https://github.com/iglesias/shogun/commit/61c4eeb055ac17bbbb3c4bed6f41e863064d94e1 | 10:28 |
n4nd0 | blackburn: that program breks because apply is not implemented yet, what is a good signal this far :) | 10:28 |
n4nd0 | brb | 10:28 |
blackburn | ok | 10:28 |
blackburn | n4nd0: I'd suggest you to add method add_subset to kernel | 10:33 |
blackburn | n4nd0: for custom kernel it just delegates to set_row and set_col subsets | 10:34 |
blackburn | for others just checks lhs==rhs and sets subsets for lhs and rhs | 10:34 |
n4nd0 | do you mean add_subset in KernelMachine or just in Kernel? | 10:38 |
blackburn | in kernel would be enough I think | 10:41 |
blackburn | thisway you may simply delegate it in add_machine_subset and remove_machine_subset | 10:41 |
n4nd0 | aha I see what you mean | 10:42 |
n4nd0 | and the method in kernel should be virtual so it can be specialized in CustomKernel | 10:43 |
blackburn | yeah | 10:44 |
blackburn | LUNCH TIME | 10:44 |
blackburn | :D | 10:44 |
n4nd0 | enjoy your meal | 10:45 |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking] | 10:59 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 10:59 | |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Ping timeout: 264 seconds] | 11:04 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 11:10 | |
blackburn | re | 11:22 |
-!- Jiuding [d2198538@gateway/web/freenode/ip.210.25.133.56] has quit [Ping timeout: 245 seconds] | 11:23 | |
-!- blackburn_ [5bdfb203@gateway/web/freenode/ip.91.223.178.3] has joined #shogun | 12:30 | |
-!- blackburn [5bdfb203@gateway/web/freenode/ip.91.223.178.3] has quit [Ping timeout: 245 seconds] | 12:32 | |
-!- l0nr4n [~l0nr4n@hugogascon.ml.tu-berlin.de] has joined #shogun | 12:45 | |
-!- l0nr4n [~l0nr4n@hugogascon.ml.tu-berlin.de] has quit [Changing host] | 12:45 | |
-!- l0nr4n [~l0nr4n@unaffiliated/l0nr4n] has joined #shogun | 12:45 | |
-!- vikram360 [~vikram360@117.192.163.50] has joined #shogun | 13:04 | |
-!- harshit_ [~harshit@182.68.142.173] has joined #shogun | 13:09 | |
-!- vikram360 [~vikram360@117.192.163.50] has quit [Ping timeout: 246 seconds] | 13:10 | |
-!- vikram360 [~vikram360@117.192.164.215] has joined #shogun | 13:10 | |
-!- harshit_ [~harshit@182.68.142.173] has quit [Ping timeout: 252 seconds] | 13:46 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: Lost terminal] | 13:54 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 13:55 | |
-!- gsomix [~gsomix@85.26.165.233] has joined #shogun | 13:55 | |
gsomix | hi | 13:56 |
gsomix | philosophy is too deep for me :) | 13:56 |
-!- gsomix [~gsomix@85.26.165.233] has quit [Client Quit] | 14:00 | |
-!- gsomix [~gsomix@85.26.165.233] has joined #shogun | 14:00 | |
-!- harshit_ [~harshit@182.68.142.173] has joined #shogun | 14:18 | |
n4nd0 | blackburn_: how important is to do the m_parameters->add(...) for each of the members of the class in the constructor? | 14:22 |
n4nd0 | blackburn_: does it give problems sometimes? | 14:22 |
blackburn_ | n4nd0: it is required for proper serialization | 14:22 |
n4nd0 | wihout clear reason it has started to fail for a boolean member | 14:22 |
blackburn_ | n4nd0: fail to compile? | 14:23 |
blackburn_ | what is the error/member? | 14:24 |
n4nd0 | seg fault | 14:24 |
n4nd0 | is a float sorry not a boolean | 14:24 |
n4nd0 | float64_t | 14:24 |
blackburn_ | wait, segfault related to float64_t?? | 14:24 |
n4nd0 | yes | 14:24 |
blackburn_ | show me the code | 14:24 |
blackburn_ | i.e. it fails with m_parameters->add and float64_t?? | 14:25 |
blackburn_ | that shouldn't be possible | 14:25 |
n4nd0 | http://snipt.org/ugDh0 | 14:26 |
n4nd0 | the line between the rpints and seg faults | 14:26 |
n4nd0 | info from gdb | 14:26 |
n4nd0 | Program received signal SIGSEGV, Segmentation fault. | 14:26 |
n4nd0 | 0xb4dd2170 in shogun::Parameter::add_type (this=0x917ec10, type=0xbfffecec, param=0x917d50c, name=0xb4e732bf "m_tolerance", description=0xb4e732ad "Tolerance member.") at base/Parameter.cpp:2369 | 14:26 |
n4nd0 | 2369if (strcmp(m_params.get_element(i)->m_name, name) == 0) | 14:26 |
blackburn_ | whoaa | 14:27 |
n4nd0 | ?? | 14:27 |
blackburn_ | no idea how can it be possible | 14:28 |
n4nd0 | but it started to happen when I added this m_slog member | 14:29 |
n4nd0 | a SGVector< int32_t > | 14:29 |
n4nd0 | before that it was working good | 14:29 |
n4nd0 | I cannot see any connection though | 14:29 |
blackburn_ | try to comment it out then | 14:32 |
n4nd0 | shit, now it seg faults in the next m_parameters->add | 14:34 |
n4nd0 | something must have got messed while adding a new member | 14:35 |
n4nd0 | I don't see why though | 14:35 |
sonne|work | n4nd0: stack overflow? | 14:35 |
blackburn_ | sonne|work: how can it cause it | 14:35 |
n4nd0 | sonne|work: do you mean because there may be too many attributes? | 14:36 |
sonne|work | maybe too much allocated on stack... | 14:37 |
sonne|work | what does valgrind say? | 14:37 |
n4nd0 | in the moment that method is executed almost nothing has been allocated | 14:37 |
blackburn_ | sonne|work: how can I reach you to get answer? :D | 14:37 |
sonne|work | I am just reading your email | 14:37 |
blackburn_ | I asked you in mail yesterday | 14:38 |
sonne|work | I think what you want to do is a lot of work | 14:38 |
blackburn_ | in irc | 14:38 |
blackburn_ | and in mail again! | 14:38 |
blackburn_ | no, I already did a half of it | 14:38 |
sonne|work | and I don't think that one should do this | 14:38 |
-!- menonnik [~chatzilla@59.178.164.96] has joined #shogun | 14:38 | |
sonne|work | just for the purpose of getting it published | 14:38 |
blackburn_ | I like this idea pretty much | 14:38 |
blackburn_ | no, I like idea itself | 14:38 |
sonne|work | cheng seemed fine without this so there is no need | 14:38 |
blackburn_ | sonne|work: it would be better in this way | 14:39 |
sonne|work | well of course you can do whatever you want but you need e.g. kernels/distances some interface to some other language or cmdline interface, some minimal build system | 14:39 |
blackburn_ | no! | 14:39 |
sonne|work | that is >1 week of work | 14:39 |
menonnik | sir, i wanted some clarifications regarding implementation if C5.0. Are we training the data to form a tree or a rule set? | 14:40 |
blackburn_ | sonne|work: it can even improve further development | 14:40 |
n4nd0 | sonne|work: it executed with valrgind without seg fault | 14:40 |
blackburn_ | I liked callbacks of OCAS | 14:40 |
sonne|work | but ocas has all this extra - extra matlab interface, extra implementation of dot features for sparse & dense, extra parallelizaed code, extra implementation for double and float... | 14:41 |
blackburn_ | sonne|work: I don't want to create my own lib! | 14:41 |
blackburn_ | I just want to extract code to some lib everyone can borrow as easy as libsvm | 14:42 |
blackburn_ | or liblinear | 14:42 |
blackburn_ | my point is just extract deep internals to libedrt.cpp | 14:42 |
blackburn_ | with callback interface | 14:42 |
blackburn_ | everything else should stay | 14:42 |
sonne|work | liblinear also has a cmdline interface & file format reader/writer | 14:43 |
blackburn_ | well if they won't like it again I'll create some simple interface | 14:44 |
blackburn_ | but this time I just want to extract its guts to make it borrowable | 14:44 |
blackburn_ | I hope 3rd submission is possible? | 14:44 |
sonne|work | blackburn_: to publish you need to be able to show that your system works somehow - so you need to provide some example code - some interface | 14:44 |
sonne|work | they will need an interface | 14:44 |
blackburn_ | it works with shogun.. | 14:45 |
blackburn_ | you suggest the same | 14:45 |
sonne|work | you should aim at getting it in this time for sure | 14:45 |
blackburn_ | I do not understand your point :( | 14:45 |
sonne|work | and I don't see how a separation helps here | 14:45 |
blackburn_ | because it would be hard to re-use that code anywhere | 14:46 |
sonne|work | what I mean is reviewers were rather positive. this and an improved selling of EDRT alone should suffice to get it accepted | 14:46 |
blackburn_ | but with libedrt.cpp and callback it would be reusable | 14:46 |
blackburn_ | all the converter stuff will just delegate its apply to libedrt | 14:47 |
blackburn_ | sonne|work: so do you prefer to let code stay as it is? | 14:47 |
sonne|work | you didn't describe exactly what you want to do. to me it seems one can just use libshogun and do all the re-use one wants | 14:47 |
sonne|work | blackburn_: well yes but maybe you need to specify what you want to do before I can give a definitive answer | 14:48 |
blackburn_ | sonne|work: I suggest to extract LLE/LTSA/all other stuff internals | 14:48 |
blackburn_ | with callback distances/feature vector access/kernels | 14:48 |
n4nd0 | sonne|work, blackburn_ : how can it be that it executes with valgrind but not alone? | 14:49 |
blackburn_ | n4nd0: valgrind does some sandboxing.. | 14:49 |
sonne|work | blackburn_: print functions, exceptions too - right? | 14:50 |
blackburn_ | sonne|work: in that case one could borrow libedrt code and provide custom distances | 14:50 |
blackburn_ | with no need to extend shogun somehow | 14:50 |
blackburn_ | it is more custom than shogun | 14:50 |
blackburn_ | sonne|work: why would I need any exceptions? | 14:51 |
blackburn_ | but probably yes | 14:51 |
blackburn_ | I don't think there should be any printing though | 14:51 |
sonne|work | error messages / progress | 14:51 |
blackburn_ | yeah, you are right, it would be needed | 14:52 |
blackburn_ | sonne|work: so shogun's LLE would stay but it would delegate m_distance->distance(i,j) as callback | 14:52 |
sonne|work | and then you need implementations for every callback | 14:53 |
sonne|work | as illustration at least | 14:53 |
blackburn_ | I need only distances, feature vector access and kernels | 14:53 |
blackburn_ | nothing more | 14:53 |
sonne|work | (which is a lot already) | 14:53 |
blackburn_ | &m_distance->distance | 14:53 |
sonne|work | so say gaussian kernel, euclidian dist, and dense realvalued features | 14:53 |
blackburn_ | &m_kernel->kernel | 14:54 |
blackburn_ | sonne|work: why should I write this? | 14:54 |
sonne|work | to publish a separate libedrt - you would need to show that it is sth useful so you need the functionality + some illustration how it can be used in the most simple case | 14:55 |
blackburn_ | got it | 14:56 |
blackburn_ | well not that much work | 14:56 |
blackburn_ | sonne|work: again, would I need to do separate libedrt?? | 14:56 |
-!- gsomix [~gsomix@85.26.165.233] has quit [Quit: ????? ? ?? ??? (xchat 2.4.5 ??? ??????)] | 14:56 | |
blackburn_ | I see no difference between not 'touching it at all' and 'extract some code' | 14:57 |
blackburn_ | in both ways it is dependent on shogun | 14:57 |
blackburn_ | my intention is not about getting published | 14:59 |
blackburn_ | but about getting it more useful | 14:59 |
sonne|work | blackburn_: in which way is it more useful? what can you do what you couldn't? | 15:00 |
blackburn_ | sonne|work: one can borrow the code and use it w/o much changing it | 15:01 |
blackburn_ | w/o changing at alll | 15:01 |
blackburn_ | as for shogun - it makes me able to provide pretty nice custom options | 15:01 |
blackburn_ | like computing full distance matrix or not | 15:01 |
blackburn_ | same for kernels | 15:02 |
sonne|work | what do you mean by borrow the code? | 15:02 |
blackburn_ | sonne|work: e.g. waffles can integrate this code in a day | 15:02 |
blackburn_ | it is really easy to provide euclidean distance and feature vector access | 15:04 |
blackburn_ | nothing more is required | 15:04 |
blackburn_ | sonne|work: i.e. just like ocas but developed inside shogun | 15:05 |
sonne|work | so what you mean is: one only needs to integrate libedrt not whole libshogun? | 15:05 |
blackburn_ | yeah | 15:05 |
blackburn_ | libedrt is shogun independent in that means | 15:05 |
blackburn_ | sonne|work: looks like you don't much like it? | 15:06 |
-!- harshit_ [~harshit@182.68.142.173] has quit [Ping timeout: 260 seconds] | 15:07 | |
blackburn_ | sonne|work: point is while libedrt provides such custom interfaces | 15:10 |
blackburn_ | i.e. distance/kernel callbacks | 15:10 |
sonne|work | not so much no... to use libedrt in shogun one then needs to create (like for libocas) all kinds of wrapping functions | 15:10 |
blackburn_ | but in shogun it is possible to use it everywhere | 15:10 |
blackburn_ | and use file features/etc | 15:11 |
blackburn_ | sonne|work: all wrapping functions are already done in features, distances and kernels | 15:11 |
sonne|work | no not all - you need one more layer of wrapping functions, like compute_output in CSVMOcas | 15:11 |
blackburn_ | sonne|work: ?? | 15:12 |
sonne|work | that calls dense_dot_range | 15:12 |
sonne|work | so you would have e.g. KernelPCA and need a mean to get k(i,j) | 15:12 |
blackburn_ | no, I'm not going to extract kernel pca | 15:12 |
sonne|work | so you need some kernel_function() | 15:12 |
blackburn_ | it is a different kind of preprocessor | 15:12 |
blackburn_ | only converters | 15:13 |
sonne|work | which underneath is calling kernel->kernel(i,j) | 15:13 |
blackburn_ | sonne|work: yes, my interface prototype requires | 15:13 |
blackburn_ | double (*kernel)(int, int) | 15:13 |
blackburn_ | &m_kernel->kernel | 15:13 |
blackburn_ | that's all | 15:13 |
blackburn_ | same with distance | 15:14 |
blackburn_ | and double* (*feature_vector)(int) | 15:14 |
sonne|work | no that won't work | 15:14 |
blackburn_ | why? | 15:14 |
sonne|work | because kernel is not a static method | 15:15 |
sonne|work | so you need a wrapper function ' kernel_function(i,j)' within your class that knows about a kernel object | 15:15 |
blackburn_ | ah I see | 15:15 |
blackburn_ | it is not a problem at all | 15:15 |
sonne|work | within that you can then do kernel->kernel(i,j) | 15:15 |
blackburn_ | I can just provide pointer there | 15:15 |
blackburn_ | it is not a problem still | 15:15 |
blackburn_ | sonne|work: I have to go now, thank you | 15:17 |
blackburn_ | I hope you got my idea | 15:17 |
sonne|work | yeah I got your idea but I still don't like it :) | 15:17 |
blackburn_ | sonne|work: damn | 15:18 |
blackburn_ | sonne|work: I'll show code next week and then we decide | 15:18 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: Lost terminal] | 15:18 | |
blackburn_ | I still think it would be useful even inside shogun (I can change things easier) | 15:18 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 15:19 | |
blackburn_ | see you | 15:19 |
-!- blackburn_ [5bdfb203@gateway/web/freenode/ip.91.223.178.3] has quit [Quit: Page closed] | 15:19 | |
-!- menonnik [~chatzilla@59.178.164.96] has quit [Read error: Connection timed out] | 15:20 | |
-!- menonnik [~chatzilla@59.178.164.96] has joined #shogun | 15:21 | |
-!- harshit_ [~harshit@182.68.142.173] has joined #shogun | 15:25 | |
n4nd0 | sonne|work: so it seems that getting rid of a couple of class members in QDA that were not used any more solved the problem | 15:32 |
n4nd0 | sonne|work: but I still don't understand why it did happen | 15:32 |
sonne|work | n4nd0: did you by chance allocate some variables on stack? | 15:33 |
sonne|work | i mean like | 15:33 |
sonne|work | double bla[10000] | 15:33 |
sonne|work | ? | 15:33 |
sonne|work | that would explain it | 15:33 |
sonne|work | (stack overflow) | 15:33 |
n4nd0 | no | 15:33 |
sonne|work | SGVector<float64_t> x(1000) doesn't create on stack btw | 15:33 |
sonne|work | I mean the object yes but not the vector | 15:34 |
n4nd0 | there was an SGVector, a float64_t, a boolean and some pointers | 15:34 |
sonne|work | hmmhh | 15:34 |
n4nd0 | one of the members was SGVector | 15:34 |
n4nd0 | maybe I should have used SGVector* even if I jus wanted one vector? | 15:35 |
n4nd0 | just* | 15:35 |
sonne|work | no SGVector is merely a pointer and a boolean underneath | 15:36 |
n4nd0 | then I don't understand why | 15:36 |
sonne|work | yeah me neither | 15:38 |
-!- menonnik [~chatzilla@59.178.164.96] has quit [Quit: ChatZilla 0.9.88 [Firefox 10.0.2/20120215223356]] | 15:41 | |
n4nd0 | sonne|work: about valgrind, I get a huge quantity of errors when I run my python examples | 15:49 |
n4nd0 | I am using a suprresions file, but still | 15:49 |
n4nd0 | I have read sth about compiling python, do you know if it improves the results doing it that way? | 15:50 |
sonne|work | n4nd0: yeah one needs more suppressions for python... | 15:52 |
n4nd0 | sonne|work: I will check how to do that then ... | 15:53 |
sonne|work | recompiling with certain settings would help - but these errors are there on purpose (faster ...) | 15:53 |
n4nd0 | the problem is that I find it quite difficult to understand whether the code is buggy or not with all those messages | 15:54 |
sonne|work | what you can do is debug either just libshogun based c++ example or from octave modular or grep for shogun:: | 15:57 |
sonne|work | it is tough for me too btw... | 15:57 |
n4nd0 | libshogun based c++ example looks like the best option from my point of view | 16:02 |
n4nd0 | it also increases the development time though | 16:02 |
sonne|work | n4nd0: well octave modular is also a good option | 16:03 |
sonne|work | (octave doesn't have valgrind errors) | 16:03 |
n4nd0 | I will take a look with octave | 16:04 |
n4nd0 | valgrind + octave right? | 16:05 |
sonne|work | yes | 16:06 |
sonne|work | just valgrind octave :) | 16:06 |
n4nd0 | thank you | 16:06 |
-!- harshit_ [~harshit@182.68.142.173] has quit [Ping timeout: 244 seconds] | 16:10 | |
n4nd0 | sonne|work: do you happen to know why some methods seem to be not accesible from octave? | 16:21 |
n4nd0 | for example apply is not found from the example classifier_lda_modular.m | 16:21 |
n4nd0 | for the class LDA | 16:21 |
n4nd0 | does this have sth to do with swig? | 16:22 |
-!- pythonroar [d2198538@gateway/web/freenode/ip.210.25.133.56] has joined #shogun | 16:24 | |
pythonroar | how to hand up my student application? | 16:25 |
-!- muddo [~muddo@gateway/tor-sasl/muddo] has quit [Quit: Leaving] | 16:27 | |
n4nd0 | pythonroar: the period is still not open, wait for next week | 16:27 |
n4nd0 | pythonroar: check more info at http://news.gmane.org/gmane.comp.ai.machine-learning.shogun | 16:27 |
-!- muddo [~muddo@gateway/tor-sasl/muddo] has joined #shogun | 16:36 | |
-!- pythonroar [d2198538@gateway/web/freenode/ip.210.25.133.56] has quit [Quit: Page closed] | 16:39 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 16:43 | |
-!- vikram360 [~vikram360@117.192.164.215] has quit [Ping timeout: 246 seconds] | 16:54 | |
-!- vikram360 [~vikram360@117.192.188.225] has joined #shogun | 16:55 | |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Read error: No route to host] | 17:02 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 17:02 | |
-!- muddo [~muddo@gateway/tor-sasl/muddo] has quit [Ping timeout: 276 seconds] | 17:13 | |
-!- muddo [~muddo@gateway/tor-sasl/muddo] has joined #shogun | 17:13 | |
-!- muddo [~muddo@gateway/tor-sasl/muddo] has quit [Remote host closed the connection] | 17:22 | |
-!- PhilTillet [~Philippe@npasserelle10.minet.net] has joined #shogun | 17:27 | |
PhilTillet | hey hey hey! | 17:27 |
PhilTillet | Hello -ChanServ- | 17:28 |
@sonney2k | PhilTillet, hi | 17:41 |
PhilTillet | sonney2k, hi ! | 17:41 |
@sonney2k | PhilTillet, did you find some time to continue the benchmarking? | 17:42 |
PhilTillet | there's some progress :) I also have a conference to prepare :p I have done the gaussian kernel for what we talked about yesterday | 17:43 |
PhilTillet | based on the code of a matrix-matrix product | 17:43 |
@sonney2k | well you cannot really assume that the whole matrix is in memory but yeah sometimes it is | 17:44 |
PhilTillet | i thought it worked to split the training set ? | 17:44 |
@sonney2k | ahh you mean the set of x_i ? | 17:46 |
@sonney2k | yes that one is in there | 17:46 |
@sonney2k | but not the set of x (might be computed on demand) | 17:46 |
PhilTillet | but didn't you link a paper from ntu where it was explained? | 17:48 |
PhilTillet | how to train when data cannot fit in memory | 17:48 |
-!- muddo [~muddo@gateway/tor-sasl/muddo] has joined #shogun | 17:50 | |
@sonney2k | PhilTillet, wait this is a totally different problem | 17:50 |
@sonney2k | one thing is to apply a trained classifier (this is what you are parallelizing) the other is about training | 17:52 |
PhilTillet | I see | 17:53 |
PhilTillet | So what do you do anyway on a CPU when training data cannot fit memory? | 17:54 |
@sonney2k | stream the data from file and use anonline algorithm or sub-sample or ... | 17:55 |
@sonney2k | (have to leave train) | 17:59 |
@sonney2k | cu | 17:59 |
-!- blackburn [~qdrgsm@83.234.54.68] has joined #shogun | 18:09 | |
-!- gsomix [~gsomix@188.168.2.59] has joined #shogun | 18:36 | |
-!- harshit_ [~harshit@182.68.142.173] has joined #shogun | 19:19 | |
-!- blackburn [~qdrgsm@83.234.54.68] has quit [Ping timeout: 265 seconds] | 19:40 | |
-!- blackburn [5bde8018@gateway/web/freenode/ip.91.222.128.24] has joined #shogun | 19:40 | |
-!- PhilTillet [~Philippe@npasserelle10.minet.net] has quit [Ping timeout: 252 seconds] | 19:53 | |
-!- l0nr4n [~l0nr4n@unaffiliated/l0nr4n] has quit [Quit: This computer (or maybe me) has gone to sleep] | 20:05 | |
harshit_ | hello | 20:41 |
-!- blackburn_ [5bde8018@gateway/web/freenode/ip.91.222.128.24] has joined #shogun | 20:46 | |
-!- blackburn [5bde8018@gateway/web/freenode/ip.91.222.128.24] has quit [Ping timeout: 245 seconds] | 20:49 | |
gsomix | harshit_, hi :) | 20:55 |
harshit_ | Hey gsomix, hav a few questions, do you hav time ? | 20:56 |
gsomix | harshit_, unfortunately I do not have time. before I had prepared for laboratory work in physics and is now going to bed. | 21:02 |
gsomix | however, I wish you good night :) | 21:03 |
gsomix | bb | 21:03 |
harshit_ | no worries .. hav a nice sleep | 21:03 |
blackburn_ | harshit_: | 21:04 |
-!- romi_ [~mizobe@187.57.1.50] has quit [Ping timeout: 246 seconds] | 21:05 | |
harshit_ | hey blackburn_ : do you hav time ? | 21:05 |
blackburn_ | yes | 21:05 |
harshit_ | i am porting newton svm to shogun for a while and now its almost done | 21:05 |
harshit_ | Do i need to port the non linear version of newton svm too ? | 21:06 |
blackburn_ | it is up to you | 21:06 |
harshit_ | okay, and i am making use of gmres function in lapack , | 21:07 |
blackburn_ | looks like this gsoc idea should be removed once you finish it | 21:08 |
harshit_ | so can i make its wrapper in mathematics/lapack.h | 21:08 |
shogun-buildbot | build #186 of nightly_none is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_none/builds/186 | 21:08 |
harshit_ | i mean, am i allowed to push updates to mathematics/lapack.h | 21:10 |
harshit_ | Also wanted to discuss my deep learning idea . | 21:10 |
harshit_ | btw are you in the team of shogun ? | 21:11 |
-!- blackburn_ [5bde8018@gateway/web/freenode/ip.91.222.128.24] has quit [Ping timeout: 245 seconds] | 21:13 | |
-!- PhilTillet [~Philippe@tillet-p42154.maisel.int-evry.fr] has joined #shogun | 21:16 | |
-!- romi_ [~mizobe@187.101.162.14] has joined #shogun | 21:18 | |
shogun-buildbot | build #185 of nightly_all is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_all/builds/185 | 21:26 |
-!- blackburn [5bde8018@gateway/web/freenode/ip.91.222.128.24] has joined #shogun | 21:33 | |
-!- blackburn [5bde8018@gateway/web/freenode/ip.91.222.128.24] has quit [Ping timeout: 245 seconds] | 21:52 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 22:03 | |
@sonney2k | n4nd0, I meant the octave_modular interface | 22:05 |
@sonney2k | not the octave_static one | 22:05 |
n4nd0 | sonney2k: I tried the example in that interface | 22:05 |
n4nd0 | sonney2k: maybe I wrote static, sorry, but I tried in modular interface | 22:06 |
@sonney2k | n4nd0, what did you try? | 22:06 |
n4nd0 | sonney2k: octave_modular/classifier_lda_modular.m | 22:07 |
@sonney2k | and what happened? | 22:07 |
n4nd0 | error: member not found | 22:07 |
n4nd0 | error: called from: | 22:07 |
n4nd0 | error: classifier_lda_modular.m at line 25, column 1 | 22:07 |
-!- l0nr4n [~l0nr4n@unaffiliated/l0nr4n] has joined #shogun | 22:09 | |
n4nd0 | sonney2k: any clue about that? | 22:11 |
@sonney2k | I am compiling for octave_modular... | 22:12 |
n4nd0 | ok | 22:12 |
n4nd0 | what happens is that apply is not found in lda but I've no idea why | 22:12 |
@sonney2k | IIRC it worked yesterday | 22:13 |
n4nd0 | maybe there is something wrong with my environment | 22:13 |
@sonney2k | n4nd0, it just works here | 22:15 |
@sonney2k | which version of swig and octave? | 22:15 |
n4nd0 | GNU Octave, version 3.2.4 | 22:16 |
@sonney2k | and are you sure that you did make install as root? | 22:16 |
@sonney2k | lapack/atlas got also found? | 22:16 |
n4nd0 | SWIG Version 2.0.4 | 22:16 |
@sonney2k | same here | 22:16 |
@sonney2k | n4nd0, does classifier_libsvm_modular work? | 22:17 |
n4nd0 | aham | 22:17 |
n4nd0 | look here | 22:17 |
n4nd0 | Checking for Atlas support ... no | 22:17 |
@sonney2k | maybe that is the culprit | 22:17 |
n4nd0 | facepalm for me ... | 22:17 |
n4nd0 | I also note this | 22:17 |
n4nd0 | Checking for Atlas support ... no | 22:17 |
n4nd0 | ups sorry | 22:18 |
n4nd0 | this one I meant | 22:18 |
@sonney2k | install libatlas-dev / libatlas-base-dev | 22:18 |
n4nd0 | Checking for octave-config ... no | 22:18 |
n4nd0 | is that an issue? | 22:18 |
@sonney2k | n4nd0, octave-config is in octave3.2-headers pkg here | 22:19 |
@sonney2k | I use this to determine link flags | 22:19 |
n4nd0 | Atlas is still not detected by configure, I have installed the packages you told me | 22:23 |
n4nd0 | maybe with libatlas-cpp-0.6-dev? | 22:23 |
-!- harshit_ [~harshit@182.68.142.173] has quit [Quit: Leaving] | 22:26 | |
@sonney2k | n4nd0, hmmhh check configure.log | 22:29 |
@sonney2k | I have libatlas3gf-base installed too | 22:29 |
n4nd0 | sonney2k: oh, didn't know about this configure.log, thanks | 22:31 |
n4nd0 | sonney2k: yeah I had that package too, must be another thing, I have to check it | 22:31 |
-!- romi_ [~mizobe@187.101.162.14] has quit [Ping timeout: 246 seconds] | 22:31 | |
@sonney2k | though that shouldn't be the issue - if you have lapack/blas it should work nevertheless | 22:32 |
n4nd0 | from the log I am reading clapack.h: No such file or directory | 22:34 |
n4nd0 | didn't expect that since I have been using blas from C++ | 22:34 |
@sonney2k | n4nd0, ohh looks like clapack.h moved from /usr/include to /usr/include/atlas | 22:43 |
@sonney2k | so try ./configure --includes=/usr/includes/atlas | 22:44 |
-!- romi_ [~mizobe@187.57.11.40] has joined #shogun | 22:44 | |
n4nd0 | sonney2k: Atlat support ... yes :) | 22:46 |
n4nd0 | let's see if this fixes the problem in octave | 22:46 |
@sonney2k | n4nd0, don't forget to do make install as root (necessary for octave...) | 22:53 |
@sonney2k | anyways I am off to bed. cu | 22:53 |
n4nd0 | sonney2k: it is working now :) | 22:53 |
@sonney2k | great | 22:54 |
n4nd0 | sonney2k: thank you I'll check tom valgrind with octave then to check qda | 22:54 |
n4nd0 | sonney2k: it is time to sleep for me too | 22:54 |
n4nd0 | good night! | 22:54 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 22:57 | |
CIA-64 | shogun: Soeren Sonnenburg master * r1336edf / src/configure : add /usr/include/atlas detection upon configure - http://git.io/CVEklA | 22:59 |
shogun-buildbot | build #552 of r_static is complete: Failure [failed test_1] Build details are at http://www.shogun-toolbox.org/buildbot/builders/r_static/builds/552 blamelist: sonne@debian.org | 23:09 |
shogun-buildbot | build #574 of cmdline_static is complete: Failure [failed test_1] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cmdline_static/builds/574 blamelist: sonne@debian.org | 23:12 |
shogun-buildbot | build #553 of octave_static is complete: Failure [failed test_1] Build details are at http://www.shogun-toolbox.org/buildbot/builders/octave_static/builds/553 blamelist: sonne@debian.org | 23:16 |
-!- muddo [~muddo@gateway/tor-sasl/muddo] has quit [Ping timeout: 276 seconds] | 23:17 | |
shogun-buildbot | build #541 of python_static is complete: Failure [failed test_1] Build details are at http://www.shogun-toolbox.org/buildbot/builders/python_static/builds/541 blamelist: sonne@debian.org | 23:20 |
PhilTillet | So many bots !!! | 23:30 |
PhilTillet | :o | 23:30 |
-!- muddo [~muddo@gateway/tor-sasl/muddo] has joined #shogun | 23:30 | |
-!- romi_ [~mizobe@187.57.11.40] has quit [Ping timeout: 246 seconds] | 23:42 | |
--- Log closed Fri Mar 23 00:00:19 2012 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!