| --- Log opened Sun May 18 00:00:09 2014 | ||
| --- Day changed Sun May 18 2014 | ||
| -!- iglesiasg [~iglesiasg@524AE0A7.cm-4-3d.dynamic.ziggo.nl] has quit [Quit: Leaving] | 00:00 | |
| -!- thoralf [~thoralf@p4FECB69B.dip0.t-ipconnect.de] has quit [Quit: Konversation terminated!] | 00:05 | |
| -!- khalednasr [~k.nasr92@41.69.174.211] has quit [Quit: Leaving] | 00:25 | |
| -!- travis-ci [~travis-ci@ec2-54-197-12-165.compute-1.amazonaws.com] has joined #shogun | 00:27 | |
| travis-ci | [travis-ci] it's tklein23's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/25422295 | 00:27 |
|---|---|---|
| -!- travis-ci [~travis-ci@ec2-54-197-12-165.compute-1.amazonaws.com] has left #shogun [] | 00:27 | |
| -!- chakri [srichakrad@gateway/shell/bnc4free/x-jdlqswkslirqbjti] has quit [Ping timeout: 265 seconds] | 00:49 | |
| -!- besser82 [quassel@fedora/besser82] has quit [Remote host closed the connection] | 00:50 | |
| -!- travis-ci [~travis-ci@ec2-54-87-68-13.compute-1.amazonaws.com] has joined #shogun | 01:01 | |
| travis-ci | [travis-ci] it's tklein23's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/25423062 | 01:01 |
| -!- travis-ci [~travis-ci@ec2-54-87-68-13.compute-1.amazonaws.com] has left #shogun [] | 01:01 | |
| -!- besser82 [quassel@fedora/besser82] has joined #shogun | 01:14 | |
| -!- mode/#shogun [+o besser82] by ChanServ | 01:14 | |
| -!- chakri [srichakrad@gateway/shell/bnc4free/x-knvzsgtyqwvaosym] has joined #shogun | 01:23 | |
| -!- HeikoS [~heiko@c83-250-213-187.bredband.comhem.se] has joined #shogun | 01:54 | |
| -!- mode/#shogun [+o HeikoS] by ChanServ | 01:54 | |
| shogun-buildbot | build #302 of debian wheezy - memcheck is complete: Failure [failed memory check] Build details are at http://buildbot.shogun-toolbox.org/builders/debian%20wheezy%20-%20memcheck/builds/302 blamelist: tklein23 <tklein23@users.noreply.github.com>, Thoralf Klein <thoralf@fischlustig.de> | 02:21 |
| -!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] | 02:52 | |
| shogun-buildbot | build #7 of nightly_fedora is complete: Failure [failed git] Build details are at http://buildbot.shogun-toolbox.org/builders/nightly_fedora/builds/7 blamelist: tklein23 <tklein23@users.noreply.github.com>, Thoralf Klein <thoralf@fischlustig.de> | 03:00 |
| shogun-buildbot | build #303 of debian wheezy - memcheck is complete: Failure [failed memory check] Build details are at http://buildbot.shogun-toolbox.org/builders/debian%20wheezy%20-%20memcheck/builds/303 blamelist: tklein23 <tklein23@users.noreply.github.com>, Thoralf Klein <thoralf@fischlustig.de> | 04:35 |
| -!- HeikoS [~heiko@c83-250-213-187.bredband.comhem.se] has quit [Ping timeout: 255 seconds] | 05:10 | |
| -!- HeikoS [~heiko@c83-250-213-187.bredband.comhem.se] has joined #shogun | 05:12 | |
| -!- mode/#shogun [+o HeikoS] by ChanServ | 05:12 | |
| -!- PirosB3 [~pirosb3@cpc4-newt31-2-0-cust631.newt.cable.virginm.net] has joined #shogun | 05:50 | |
| -!- parijat [75c7dcfd@gateway/web/freenode/ip.117.199.220.253] has joined #shogun | 05:58 | |
| -!- iglesiasg [~iglesiasg@524AE0A7.cm-4-3d.dynamic.ziggo.nl] has joined #shogun | 05:59 | |
| -!- mode/#shogun [+o iglesiasg] by ChanServ | 05:59 | |
| parijat | iglesiasg: hi! good morning. ;-) | 06:01 |
| @iglesiasg | parijat, Good morning! | 06:01 |
| parijat | iglesiasg: I saw your comments. I will update them together with other comments that you may have | 06:02 |
| @iglesiasg | parijat, I am going to continue reviewing now | 06:03 |
| parijat | parijat: great! I am going to be available in irc if you need to clarify something | 06:03 |
| parijat | iglesiasg: for you :-P | 06:04 |
| -!- Saurabh7 [~Saurabh7@117.222.8.168] has joined #shogun | 06:04 | |
| -!- PirosB3 [~pirosb3@cpc4-newt31-2-0-cust631.newt.cable.virginm.net] has quit [Quit: PirosB3] | 06:06 | |
| @iglesiasg | parijat, Great, thanks! | 06:06 |
| parijat | afk. brb in 2 mins | 06:06 |
| parijat | I'm back | 06:17 |
| shogun-buildbot | build #805 of nightly_default is complete: Success [build successful] Build details are at http://buildbot.shogun-toolbox.org/builders/nightly_default/builds/805 | 06:21 |
| @iglesiasg | parijat, in the prune tree routine | 06:27 |
| @iglesiasg | parijat, when data.attribute_id is -1, why can we be sure that there will be always two children? | 06:27 |
| parijat | iglesiasg: data.attribute_id is not -1 | 06:28 |
| @iglesiasg | parijat, ah sorry. Yeah, when it is not -1 but the attribute is not nomial | 06:28 |
| @iglesiasg | nominal* | 06:28 |
| -!- parijat_ [75dc29d5@gateway/web/freenode/ip.117.220.41.213] has joined #shogun | 06:31 | |
| parijat_ | iglesiasg: connection problem! I was saying that the if else is for whether the attribute is nominal or not | 06:31 |
| @iglesiasg | parijat_, yep | 06:32 |
| parijat_ | iglesiasg: the not equality to -1 is to ensure array can be indexed | 06:32 |
| -!- parijat [75c7dcfd@gateway/web/freenode/ip.117.199.220.253] has quit [Ping timeout: 240 seconds] | 06:32 | |
| -!- parijat_ is now known as parijat | 06:33 | |
| @iglesiasg | parijat_, yeah, but if attribute_id can actually be -1, I think the check should be before than indexing the array | 06:33 |
| @iglesiasg | parijat, see what I mean? | 06:33 |
| parijat | iglesiasg: yes I could create a larger if encompassing this entire block, but this works just as fine I checked. So I left it this way | 06:35 |
| @iglesiasg | parijat, hmm | 06:35 |
| @iglesiasg | parijat, I meant this | 06:35 |
| @iglesiasg | instead of | 06:35 |
| @iglesiasg | if (m_nominal[current->data.attribute_id] && (current->data.attribute_id!=-1)) | 06:35 |
| @iglesiasg | doing | 06:35 |
| @iglesiasg | if ((current->data.attribute_id!=-1) && m_nominal[current->data.attribute_id]) | 06:36 |
| @iglesiasg | because if current->data.attribute_id == -1 the m_nominal[current->data.attribute_id] is going to break | 06:36 |
| @iglesiasg | in any case, you know better when data.attribute_id will be -1 | 06:37 |
| @iglesiasg | it just looks weird to me doing | 06:37 |
| @iglesiasg | if (array[i] && i >= 0) | 06:37 |
| parijat | iglesiasg: okay ya! I can change that.. | 06:37 |
| @iglesiasg | parijat, explain me anyway when can it be that data.attribute_id is -1, I sort of recall checking that when you pushed id3 | 06:38 |
| parijat | iglesiasg: but I think C++ compiler does something intelligent here.. because that thing is -1 everytime we reach a leaf node and I didnt get any runtime errors yet | 06:38 |
| parijat | iglesiasg: its -1 for leaf nodes. -1 is the default value and in leaf nodes we need not set it to any other value because we need not choose any more attribute to classify data further | 06:40 |
| @iglesiasg | parijat, all right | 06:40 |
| parijat | iglesiasg: so basically data.attribute_id is -1 is a check for leaf nodes | 06:40 |
| @iglesiasg | parijat, got it | 06:41 |
| parijat | iglesiasg: :-) | 06:41 |
| @iglesiasg | parijat, I am debugging in my head the program when attribute_id is -1 :) | 06:42 |
| @iglesiasg | parijat, so, if I understand it correctly, when prune_tree_machine is called with current being a leaf node, predicted_unpruned and predicted_pruned are the same. Is that right? | 06:45 |
| parijat | iglesiasg: ya you are right. | 06:45 |
| parijat | iglesiasg: there is nothing to prune in a leaf node | 06:46 |
| @iglesiasg | parijat, which means that prune_tree_machine is not going to do anything for leaf nodes | 06:46 |
| @iglesiasg | parijat, exactly | 06:46 |
| @iglesiasg | parijat, what about doing like this at the very top of prune_tree_machine | 06:46 |
| @iglesiasg | / if current node is a leaf node, pruning the current node makes no difference in the tree | 06:47 |
| @iglesiasg | if (current->data.attribute.id == -1) | 06:47 |
| @iglesiasg | return | 06:47 |
| @iglesiasg | or something like that (the comment doesn't need to be the same :P) | 06:47 |
| parijat | iglesiasg: ya thats a better way to do it. | 06:47 |
| @iglesiasg | parijat, cool, then we can get rid of the check in the if | 06:48 |
| parijat | iglesiasg: plus unnecessary code execution | 06:48 |
| @iglesiasg | indeed | 06:48 |
| @iglesiasg | parijat, now, tell me about the second branch of the if. For non-nominal nodes, there will always be two children, right? | 06:49 |
| parijat | iglesiasg: yes always.. | 06:49 |
| parijat | iglesiasg: we split across a single threshold remember? | 06:49 |
| @iglesiasg | parijat, ah yeah, true | 06:50 |
| @iglesiasg | let's just put a sanity check with a require | 06:50 |
| @iglesiasg | in case someone ever updates that with several thresholds or whatever, the code in prune doesn't break or behaves unexpectedly | 06:51 |
| parijat | iglesiasg: alright sure! | 06:51 |
| parijat | iglesiasg: could you please put 2-3 word comments for these things that we discussed in the PR. It would help me remember while I make the update | 06:52 |
| @iglesiasg | parijat, already done | 06:52 |
| parijat | iglesiasg: thanks! | 06:53 |
| parijat | brb in a miin | 06:53 |
| parijat | i'm back | 06:58 |
| @iglesiasg | parijat, maybe we can discuss a moment about the comment I just wrote on line 484 | 06:59 |
| @iglesiasg | brb | 06:59 |
| parijat | iglesiasg: threshold cant be larger than or less than all elements, but the validation data chosen can be such that all data points from current node move to only left child or only right child. In that case the mentioned special cases can happen | 07:02 |
| parijat | iglesiasg: to back up.. this case arises in the unittest which I designed | 07:10 |
| @iglesiasg | parijat, all right! Can you make a comment with that? | 07:13 |
| @iglesiasg | parijat, I think it is more clear then | 07:13 |
| parijat | iglesiasg: sure! | 07:13 |
| @iglesiasg | parijat, also, for the next time please keep separate pull requests | 07:13 |
| @iglesiasg | parijat, one for the pruning, another one for handling missing values, etc | 07:14 |
| @iglesiasg | parijat, the review process is much faster then.. | 07:14 |
| @iglesiasg | parijat, do you know you can issue several pull requests at the same time?Just use different branches | 07:14 |
| parijat | iglesiasg: ok. actually I added handling missing values first... then later I added the pruning thing | 07:15 |
| @iglesiasg | parijat, a new one for the pruning would have done it then ;) | 07:15 |
| @iglesiasg | as long as it is not dependent on the previous one, you could have done a couple of separate PRs | 07:16 |
| parijat | iglesiasg: ok. atleast in this case, these are not related. so I could have actually sent in separate PRs | 07:17 |
| -!- pickle27 [~pickle27@192-0-136-118.cpe.teksavvy.com] has quit [Remote host closed the connection] | 07:18 | |
| parijat | iglesiasg: about NOT_A_NUMBER.. I tried using it at first, but is there a way to check equality for it? | 07:20 |
| @iglesiasg | parijat, I see in CMath that there is fequals that calls CMath::is_nan | 07:21 |
| @iglesiasg | parijat, did you try that? | 07:21 |
| parijat | iglesiasg: nope.. I was just foolishly using simple == stuff | 07:22 |
| parijat | iglesiasg: I will change back to NOT_A_NUMBER and try fequals. | 07:24 |
| @iglesiasg | parijat, great | 07:25 |
| -!- parijat_ [75c7d894@gateway/web/freenode/ip.117.199.216.148] has joined #shogun | 07:32 | |
| -!- parijat [75dc29d5@gateway/web/freenode/ip.117.220.41.213] has quit [Ping timeout: 240 seconds] | 07:33 | |
| -!- parijat_ is now known as parijat | 07:35 | |
| -!- iglesiasg is now known as iglesiasg|afk | 07:36 | |
| @iglesiasg|afk | see you later! | 07:44 |
| -!- iglesiasg|afk [~iglesiasg@524AE0A7.cm-4-3d.dynamic.ziggo.nl] has quit [Quit: Leaving] | 07:44 | |
| -!- parijat [75c7d894@gateway/web/freenode/ip.117.199.216.148] has quit [Quit: Page closed] | 07:47 | |
| -!- kislay [~Abhijeet@103.240.206.190] has quit [Ping timeout: 264 seconds] | 07:56 | |
| -!- Saurabh7 [~Saurabh7@117.222.8.168] has quit [Quit: Leaving] | 08:08 | |
| -!- kislay [~Abhijeet@223.186.141.192] has joined #shogun | 08:12 | |
| -!- Saurabh7 [~Saurabh7@117.222.8.168] has joined #shogun | 08:13 | |
| -!- Saurabh7 [~Saurabh7@117.222.8.168] has quit [Quit: Leaving] | 09:50 | |
| -!- jiaolong [5aa3e8bc@gateway/web/freenode/ip.90.163.232.188] has joined #shogun | 10:49 | |
| -!- Saurabh7 [~Saurabh7@117.222.8.168] has joined #shogun | 11:25 | |
| -!- PirosB3 [~pirosb3@cpc4-newt31-2-0-cust631.newt.cable.virginm.net] has joined #shogun | 11:38 | |
| -!- Saurabh7 [~Saurabh7@117.222.8.168] has quit [Quit: Leaving] | 12:45 | |
| -!- shogun-notifier- [~irker@7nn.de] has joined #shogun | 13:39 | |
| shogun-notifier- | shogun: Wu Lin :develop * 45567e6 / / (3 files): https://github.com/shogun-toolbox/shogun/commit/45567e6c0cb3201615f33e03176b78f015fe6258 | 13:39 |
| shogun-notifier- | shogun: update the integration class | 13:39 |
| shogun-notifier- | shogun: Wu Lin :develop * 10e4c1d / src/shogun/mathematics/Integration.cpp,src/shogun/mathematics/Integration.h: https://github.com/shogun-toolbox/shogun/commit/10e4c1d6160abe6748c127b68d28aa5f3515b54b | 13:39 |
| shogun-notifier- | shogun: change to BSD | 13:39 |
| shogun-notifier- | shogun: Heiko Strathmann :develop * c705bd1 / / (3 files): https://github.com/shogun-toolbox/shogun/commit/c705bd15824b9306bc2d343f41702f3452c8ca65 | 13:39 |
| shogun-notifier- | shogun: Merge pull request #2216 from yorkerlin/yorker-integration | 13:39 |
| shogun-notifier- | shogun: | 13:39 |
| shogun-notifier- | shogun: update the integration class | 13:39 |
| shogun-notifier- | shogun: Parijat Mazumdar :develop * 09f808d / / (3 files): https://github.com/shogun-toolbox/shogun/commit/09f808d2e7f9752052215f208ff64b626379fabf | 13:53 |
| shogun-notifier- | shogun: Features added for handling missing values and tree pruning | 13:53 |
| shogun-notifier- | shogun: Fernando Iglesias :develop * 1e50351 / / (3 files): https://github.com/shogun-toolbox/shogun/commit/1e50351bc7dc0342fc52d606a6ec71f18c7f734a | 13:53 |
| shogun-notifier- | shogun: Merge pull request #2226 from mazumdarparijat/C45 | 13:53 |
| shogun-notifier- | shogun: | 13:53 |
| shogun-notifier- | shogun: Add feature for tree pruning and handling missing values in C4.5 | 13:53 |
| shogun-buildbot | build #2290 of bsd1 - libshogun is complete: Failure [failed compile test] Build details are at http://buildbot.shogun-toolbox.org/builders/bsd1%20-%20libshogun/builds/2290 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Wu Lin <yorker.lin@gmail.com> | 13:53 |
| -!- HeikoS [~heiko@c83-250-213-187.bredband.comhem.se] has quit [Ping timeout: 265 seconds] | 13:59 | |
| -!- HeikoS [~heiko@c83-250-213-187.bredband.comhem.se] has joined #shogun | 14:05 | |
| -!- mode/#shogun [+o HeikoS] by ChanServ | 14:05 | |
| -!- lambday [67157f4f@gateway/web/freenode/ip.103.21.127.79] has joined #shogun | 14:10 | |
| -!- HeikoS [~heiko@c83-250-213-187.bredband.comhem.se] has quit [Ping timeout: 240 seconds] | 14:16 | |
| -!- Saurabh7 [~Saurabh7@117.222.8.168] has joined #shogun | 14:23 | |
| -!- jiaolong [5aa3e8bc@gateway/web/freenode/ip.90.163.232.188] has quit [Quit: Page closed] | 14:37 | |
| shogun-buildbot | build #2291 of bsd1 - libshogun is complete: Success [build successful] Build details are at http://buildbot.shogun-toolbox.org/builders/bsd1%20-%20libshogun/builds/2291 | 14:46 |
| shogun-buildbot | build #1727 of rpm1 - libshogun is complete: Failure [failed test] Build details are at http://buildbot.shogun-toolbox.org/builders/rpm1%20-%20libshogun/builds/1727 blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com>, Parijat Mazumdar <mazumdarparijat@gmail.com> | 14:56 |
| -!- [1]Saurabh7 [~Saurabh7@117.200.182.214] has joined #shogun | 15:24 | |
| -!- Saurabh7 [~Saurabh7@117.222.8.168] has quit [Ping timeout: 255 seconds] | 15:26 | |
| -!- [1]Saurabh7 is now known as Saurabh7 | 15:26 | |
| -!- [1]Saurabh7 [~Saurabh7@59.88.37.225] has joined #shogun | 15:38 | |
| -!- [2]Saurabh7 [~Saurabh7@117.212.152.19] has joined #shogun | 15:40 | |
| -!- Saurabh7 [~Saurabh7@117.200.182.214] has quit [Ping timeout: 258 seconds] | 15:40 | |
| -!- [2]Saurabh7 is now known as Saurabh7 | 15:40 | |
| -!- [1]Saurabh7 [~Saurabh7@59.88.37.225] has quit [Ping timeout: 252 seconds] | 15:42 | |
| -!- [1]Saurabh7 [~Saurabh7@117.222.9.66] has joined #shogun | 15:49 | |
| -!- Saurabh7 [~Saurabh7@117.212.152.19] has quit [Read error: Connection reset by peer] | 15:49 | |
| -!- [1]Saurabh7 is now known as Saurabh7 | 15:49 | |
| -!- kislay [~Abhijeet@223.186.141.192] has quit [Ping timeout: 258 seconds] | 16:00 | |
| -!- lisitsyn [~qdrgsm@80.252.20.67] has quit [Ping timeout: 255 seconds] | 16:30 | |
| shogun-buildbot | build #304 of debian wheezy - memcheck is complete: Failure [failed memory check] Build details are at http://buildbot.shogun-toolbox.org/builders/debian%20wheezy%20-%20memcheck/builds/304 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Wu Lin <yorker.lin@gmail.com> | 16:32 |
| -!- lisitsyn [~qdrgsm@80.252.20.67] has joined #shogun | 16:38 | |
| -!- kislay [~Abhijeet@110.226.208.114] has joined #shogun | 16:42 | |
| @besser82 | lambday, HeikoS said sth about some cmake stuff you may need help with?!? | 16:51 |
| -!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] | 16:53 | |
| -!- lambday [67157f4f@gateway/web/freenode/ip.103.21.127.79] has quit [Ping timeout: 240 seconds] | 16:54 | |
| -!- khalednasr [~k.nasr92@41.69.144.44] has joined #shogun | 17:09 | |
| -!- besser82 [quassel@fedora/besser82] has quit [Read error: Connection reset by peer] | 17:14 | |
| -!- chakri [srichakrad@gateway/shell/bnc4free/x-knvzsgtyqwvaosym] has quit [Remote host closed the connection] | 17:14 | |
| -!- besser82 [quassel@gateway/shell/bnc4free/x-lnpfvhdudmkuqmma] has joined #shogun | 17:36 | |
| -!- besser82 [quassel@gateway/shell/bnc4free/x-lnpfvhdudmkuqmma] has quit [Changing host] | 17:36 | |
| -!- besser82 [quassel@fedora/besser82] has joined #shogun | 17:36 | |
| -!- mode/#shogun [+o besser82] by ChanServ | 17:36 | |
| -!- chakri [srichakrad@gateway/shell/bnc4free/x-gbvhltybbqawmhze] has joined #shogun | 17:47 | |
| shogun-buildbot | build #305 of debian wheezy - memcheck is complete: Failure [failed memory check] Build details are at http://buildbot.shogun-toolbox.org/builders/debian%20wheezy%20-%20memcheck/builds/305 blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com>, Parijat Mazumdar <mazumdarparijat@gmail.com> | 18:55 |
| -!- pickle27 [~pickle27@192-0-136-118.cpe.teksavvy.com] has joined #shogun | 19:16 | |
| kislay | hey pickle27 | 19:21 |
| pickle27 | hey dude | 19:29 |
| pickle27 | kislay: its coming along | 19:32 |
| pickle27 | couple of things | 19:32 |
| pickle27 | I think we probably want the interface to use class methods that take the cv mat as an arg | 19:32 |
| pickle27 | rather than creating the obj with the constructor and calling getMatrix | 19:33 |
| pickle27 | we'll then make the constructor private so you can't make an instance of the factory | 19:33 |
| pickle27 | also | 19:33 |
| pickle27 | I'd rather that it was one factory class and you pass a flag to indicate which conversion method to use | 19:34 |
| pickle27 | then we'll benchmark from there and leave only the fastest implementation | 19:34 |
| pickle27 | I think you'll find it easier to write the benchmark with that design / architecture | 19:34 |
| kislay | okay. yeah that will be better too. | 19:35 |
| kislay | so instead of applying the cv Mat through constructor, We need to use it class method | 19:36 |
| pickle27 | yeah! | 19:36 |
| pickle27 | then when you use it it will look something like | 19:36 |
| pickle27 | SGMatrix = SGMatrixFactory::GetMatrix(cv::Mat) | 19:36 |
| pickle27 | you can probably pick better names | 19:37 |
| kislay | yeah. :) | 19:37 |
| kislay | hmm.. I am getting what you said. | 19:38 |
| kislay | so, the way we use it should be in a single line . | 19:38 |
| kislay | here It takes 3 lines to get the work done | 19:38 |
| -!- Saurabh7 [~Saurabh7@117.222.9.66] has quit [Quit: I love my HydraIRC -> http://www.hydrairc.com <-] | 19:50 | |
| pickle27 | exactly | 19:50 |
| pickle27 | it'll feel more like a function than a class really | 19:50 |
| kislay | yeah | 19:51 |
| pickle27 | but having it as a class lets us hide more stuff if we need to | 19:51 |
| kislay | yup. Thats right. I will update the gist. | 19:51 |
| kislay | And yeah, I have been updating the doc. Will add more to it. Just see it. Ithink I added few things there. | 19:52 |
| pickle27 | sounds good! | 19:59 |
| -!- lisitsyn [~qdrgsm@80.252.20.67] has quit [Read error: Connection reset by peer] | 20:06 | |
| -!- PirosB3 [~pirosb3@cpc4-newt31-2-0-cust631.newt.cable.virginm.net] has quit [Quit: PirosB3] | 20:15 | |
| -!- PirosB3 [~pirosb3@cpc4-newt31-2-0-cust631.newt.cable.virginm.net] has joined #shogun | 20:15 | |
| -!- PirosB3 [~pirosb3@cpc4-newt31-2-0-cust631.newt.cable.virginm.net] has quit [Client Quit] | 20:16 | |
| -!- pickle27 [~pickle27@192-0-136-118.cpe.teksavvy.com] has quit [Remote host closed the connection] | 21:09 | |
| -!- soumyaC [uid15286@gateway/web/irccloud.com/x-hzeaclwgxxmqfqoe] has joined #shogun | 21:29 | |
| -!- thoralf [~thoralf@46.114.130.127] has joined #shogun | 21:38 | |
| thoralf | Hello. | 21:39 |
| -!- abinash_panda [75ef5e70@gateway/web/freenode/ip.117.239.94.112] has joined #shogun | 22:23 | |
| thoralf | Hey Abinash | 22:27 |
| abinash_panda | thoralf: Hi ! | 22:28 |
| abinash_panda | thoralf: Is there any implementation of evaluation metrics for multi-label classification in shogun | 22:30 |
| thoralf | abinash_panda: No, I don't think so. | 22:36 |
| abinash_panda | thoralf: So, for comparing the accuracy, I would be implementing CMultilabelAccuracy in week 2. | 22:39 |
| abinash_panda | I am thinking of implementing an example-based metric as specified in http://cse.seu.edu.cn/people/zhangml/files/TKDE'13LV.pdf | 22:39 |
| abinash_panda | any suggestions | 22:39 |
| thoralf | abinash_panda: Yeah, just something like accuracy. | 22:42 |
| thoralf | abinash_panda: Don't spend too much time on this. If we're good in time, we can do something more general. | 22:43 |
| abinash_panda | thoralf: Ok. So, I would be implementing it also in the week 2. | 22:43 |
| thoralf | abinash_panda: Alright. | 22:44 |
| abinash_panda | thoralf: Yeah. I think it can be implemented very easily, so I would not be spending too much time on this one. | 22:44 |
| thoralf | abinash_panda: Exact. Especially when we're doing the example based metrics. | 22:45 |
| thoralf | We can simply add/average numbers. | 22:45 |
| thoralf | Label-based may be annoying, because shogun works more-less example-based. | 22:46 |
| abinash_panda | Ok. | 22:47 |
| thoralf | Maybe the connection will abort soon. | 22:51 |
| * thoralf is on the train. | 22:51 | |
| abinash_panda | thoralf: I have updated the GSoC schedule. If you could have a look and suggest any changes to be made. | 22:54 |
| abinash_panda | The link is https://docs.google.com/a/itbhu.ac.in/document/d/1PnlKr_VRDS6QOxquZILP5BpPH9FgCtekxOKSz6Rq0l0/edit | 22:54 |
| thoralf | Okay. | 22:54 |
| thoralf | 2 minutes. | 22:54 |
| thoralf | abinash_panda: Please check ./examples/undocumented/python_modular/classifier_multilabeloutputliblinear_modular.py | 22:55 |
| thoralf | Okay, weeks 1-2 are ambitious. I think the schedule is too optimistic, but in week 3 we have some buffer to compensate. | 22:59 |
| thoralf | In week 4 you're planning to reuse Hashed*Features. First of all, I thing we shouldn't use DENSE features at all. Only concentrate on sparse features. | 22:59 |
| thoralf | Second, I think we cannot really re-use HashedSparseFeatures, because it's missing an option to use different seeds for hashing. | 23:01 |
| thoralf | Again, I think week 4 is too optimistic, but weeks 5-6 are easier. So I guess you will need to carry some work from week 4 to the following weeks. Are you aware of that? | 23:02 |
| abinash_panda | thoralf: Ok. | 23:03 |
| abinash_panda | Any suggestions on modification | 23:03 |
| thoralf | I think in week 7 we should directly head for CLR and skip the pairwise comparison. | 23:05 |
| thoralf | The schedule is already very tight. | 23:05 |
| abinash_panda | Ok. | 23:06 |
| thoralf | For week 7 you can schedule a ipython notebook to present what you did until the mid-term-evaluation. | 23:06 |
| thoralf | And preparing the CLR stuff. | 23:06 |
| thoralf | For CLR you'll need another decoding (argmax method). Everything else should stay the same (if we did it right). | 23:06 |
| -!- thoralf [~thoralf@46.114.130.127] has quit [Remote host closed the connection] | 23:07 | |
| -!- thoralf [~thoralf@46.114.130.127] has joined #shogun | 23:08 | |
| thoralf | Hey again. | 23:08 |
| thoralf | [22:59] <thoralf> Second, I think we cannot really re-use HashedSparseFeatures, because it's missing an option to use different seeds for hashing. | 23:08 |
| thoralf | [23:01] <thoralf> Again, I think week 4 is too optimistic, but weeks 5-6 are easier. So I guess you will need to carry some work from week 4 to the following weeks. Are you aware of that? | 23:08 |
| thoralf | [23:03] <thoralf> I think in week 7 we should directly head for CLR and skip the pairwise comparison. | 23:08 |
| thoralf | [23:03] <thoralf> The schedule is already very tight. | 23:08 |
| thoralf | [23:04] <thoralf> For week 7 you can schedule a ipython notebook to present what you did until the mid-term-evaluation. | 23:08 |
| thoralf | [23:04] <thoralf> And preparing the CLR stuff. | 23:08 |
| thoralf | [23:05] <thoralf> For CLR you'll need another decoding (argmax method). Everything else should stay the same (if we did it right). | 23:09 |
| abinash_panda | thoralf: Yeah. I got it. | 23:10 |
| thoralf | Btw., hierachical prediction isn't much more complicated than CLR. | 23:11 |
| thoralf | So we could do this as well. | 23:11 |
| thoralf | Decoding will be a bit more fun, because it enables some neat optimizations. | 23:11 |
| thoralf | But the rest is straight-forward. | 23:11 |
| abinash_panda | Ok. So I can move the implemenation of CLR to week 7 and implementation of hierarchical predictions to week 8. Is it too optimistic | 23:13 |
| thoralf | abinash_panda: Maybe you can answer the question on your own: How familiar are you with C++? How much time can you spend (and how much additional effort if you run out of time)? | 23:14 |
| thoralf | abinash_panda: It's basically up to you. | 23:14 |
| thoralf | abinash_panda: But we can try... and I'm happy to guide you. ;) | 23:15 |
| abinash_panda | thoralf: I would like to give it a try then :) | 23:15 |
| thoralf | Alright. :D | 23:15 |
| thoralf | The schedule looks good so far. But I think there's some risk breaking deadlines, because if so many new things. | 23:18 |
| abinash_panda | Ok. I am ready to put additional effort if I ran out of time ;) | 23:19 |
| thoralf | It's not problem if you run out of time, but please report every few days about your progress so I can create a emergency plan. | 23:19 |
| thoralf | (just in case I/you see problems) | 23:19 |
| abinash_panda | Yeah, sure | 23:20 |
| thoralf | What about the power outage in your city? Everything back to normal? | 23:21 |
| abinash_panda | Yeah. Now everything is back to normal. | 23:21 |
| abinash_panda | So, I have a very silly doubt (sorry for asking dumb questions). As GSoC starts from tomorrow, I would be committing the codes to my repository. So, should I send the PR after everything is implemented or something else? | 23:23 |
| thoralf | I don't remember the last power outage here... so may I ask how often this happens in India? | 23:23 |
| abinash_panda | In India, it varies from city to city. During summers, in most part of northern India, there is regular power outage. | 23:24 |
| thoralf | abinash_panda: Okay, the thing is that we're trying to submit only "complete" features. If a feature takes more than a few days, try to split it to smaller packages and PR them one by one. | 23:25 |
| thoralf | abinash_panda: For example, if you plan to implement different MultilabelModels, then PR them separately. | 23:26 |
| thoralf | abinash_panda: MultilabelEvaluation would be a separate PR as well. | 23:26 |
| abinash_panda | Ok. I got it | 23:27 |
| thoralf | abinash_panda: An example using the Multilabel classes could be a PR. | 23:27 |
| thoralf | Rule of thumb: If the code is broken or incomplete, we won't merge it. But, if you like to, we can do a feature branch. | 23:27 |
| thoralf | (Or you do a feature branch in your repo an PR as soon you achieved the next working version) | 23:28 |
| thoralf | abinash_panda: Did you check examples/undocumented/python_modular/classifier_multilabeloutputliblinear_modular.py ? | 23:29 |
| thoralf | abinash_panda: It's similar to what we do. Just as an inspiration. | 23:29 |
| abinash_panda | Yeah. I checked the example. | 23:30 |
| abinash_panda | I am currently going through the implementation of apply_multilabel_output() method. | 23:35 |
| abinash_panda | thoralf: It's getting late in night over here. Got to go. I would start working from tomorrow :) | 23:40 |
| thoralf | Oh, funny. MulticlassMachine supports both multiclass and multilabel | 23:40 |
| thoralf | (Well, multilabel is only multiple-outputs-multi-class, but still - it supports both.) | 23:41 |
| thoralf | Okay, sleep well. So tomorrow you're starting your own MultilabelModel? | 23:42 |
| abinash_panda | Yeah. I would start implementing MultilabelModel :) | 23:43 |
| thoralf | I should be on IRC whole day tomorrow, just in case. ;) | 23:44 |
| thoralf | I'm already very excited about this project. | 23:44 |
| thoralf | Because it's starting something that has been missing in shogun since years. :D | 23:45 |
| thoralf | On one hand, there's some easy tasks and easy wins. | 23:45 |
| thoralf | On the other hand, we'll see how to integrate this "new" approach into the whole system. | 23:45 |
| thoralf | And probably a few difficulties. | 23:46 |
| thoralf | But for me it looks possible. | 23:46 |
| abinash_panda | I'm also very excited to work on this project. I would ping you in case of any doubts or difficulties ;) | 23:47 |
| thoralf | Good luck and sleep well. ;) | 23:47 |
| abinash_panda | Good night :) | 23:48 |
| -!- abinash_panda [75ef5e70@gateway/web/freenode/ip.117.239.94.112] has quit [] | 23:48 | |
| -!- thoralf [~thoralf@46.114.130.127] has quit [Quit: Konversation terminated!] | 23:49 | |
| --- Log closed Mon May 19 00:00:54 2014 | ||
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!