--- Log opened Tue May 29 00:00:26 2018 | ||
-!- wuwei [wuweilinma@gateway/shell/matrix.org/x-pfqiuxsyziajvgrl] has quit [Ping timeout: 277 seconds] | 04:42 | |
-!- Trixis [~Trixis@unaffiliated/trixis] has quit [Ping timeout: 260 seconds] | 04:45 | |
-!- Netsplit *.net <-> *.split quits: @sukey, tctara_, lisitsyn, shogun-buildbot, durovo1, @ChanServ | 04:47 | |
-!- Netsplit *.net <-> *.split quits: @wiking | 04:47 | |
-!- Netsplit *.net <-> *.split quits: zxtx, rcurtin, sonney2k_, micmn, shogitter | 04:47 | |
-!- Netsplit over, joins: lisitsyn, shogun-buildbot, durovo1, @wiking, @ChanServ, @sukey, micmn, tctara_, rcurtin, zxtx (+2 more) | 04:48 | |
-!- Trixis [~Trixis@unaffiliated/trixis] has joined #shogun | 04:48 | |
-!- Netsplit *.net <-> *.split quits: @sukey, tctara_, durovo1 | 04:50 | |
-!- Netsplit over, joins: durovo1, tctara_ | 04:51 | |
-!- Netsplit over, joins: @sukey | 04:52 | |
-!- sukey2 [~nodebot@ks312251.kimsufi.com] has joined #shogun | 04:52 | |
-!- sukey [~nodebot@ks312251.kimsufi.com] has quit [Remote host closed the connection] | 04:52 | |
-!- durovo1 [~durovo@31.8f.559e.ip4.static.sl-reverse.com] has quit [Write error: Broken pipe] | 04:52 | |
-!- durovo [~durovo@31.8f.559e.ip4.static.sl-reverse.com] has joined #shogun | 04:52 | |
-!- wuwei1 [wuweilinma@gateway/shell/matrix.org/x-lwvcxvbgbdmrbwiy] has joined #shogun | 05:12 | |
-!- durovo2 [~durovo@31.8f.559e.ip4.static.sl-reverse.com] has joined #shogun | 08:27 | |
-!- durovo [~durovo@31.8f.559e.ip4.static.sl-reverse.com] has quit [Remote host closed the connection] | 08:27 | |
@wiking | wuwei1, hey | 09:49 |
---|---|---|
wuwei1 | wiking: hey | 10:21 |
sukey2 | [https://github.com/shogun-toolbox/shogun] Issue https://github.com/shogun-toolbox/shogun/issues/4311 vigsterkr added label: "BUG" | 10:37 |
sukey2 | [https://github.com/shogun-toolbox/shogun] Issue https://github.com/shogun-toolbox/shogun/issues/4311 vigsterkr added label: "performance improvement" | 10:37 |
sukey2 | [https://github.com/shogun-toolbox/shogun] Issue https://github.com/shogun-toolbox/shogun/issues/4311 opened by vigsterkr | 10:37 |
sukey2 | [https://github.com/shogun-toolbox/shogun] Issue https://github.com/shogun-toolbox/shogun/issues/4311 vigsterkr added label: "development tasks" | 10:37 |
@wiking | wuwei1, still here? | 11:11 |
@wiking | micmn, ping? | 11:13 |
@wiking | :) | 11:13 |
wuwei1 | wiking: i'm here | 11:27 |
@wiking | wuwei1, sent an email instead :) | 11:28 |
-!- Farouk [9cc5aeaf@gateway/web/freenode/ip.156.197.174.175] has joined #shogun | 13:06 | |
sukey2 | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/4285 synchronized by vinx13 | 13:11 |
-!- Farouk [9cc5aeaf@gateway/web/freenode/ip.156.197.174.175] has quit [Ping timeout: 260 seconds] | 13:15 | |
-!- Farouk [9cc5aeaf@gateway/web/freenode/ip.156.197.174.175] has joined #shogun | 13:19 | |
wuwei1 | wiking: could u explain a bit composite in pipeline here https://github.com/shogun-toolbox/shogun/wiki/Hackathon-2017-base-api#some-cool-stuff | 13:21 |
sukey2 | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/4285 synchronized by vinx13 | 13:28 |
Farouk | wiking, are you here? Have a quick question on the new API | 13:36 |
sukey2 | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/4312 opened by abhishek-1991 | 14:23 |
@wiking | hello hello | 14:29 |
@wiking | Farouk, sorry i had some offline things to do | 14:29 |
@wiking | but now i'm here | 14:29 |
@wiking | wuwei1, oh yeah | 14:29 |
Farouk | Hey wiking. No worries. I think I found a similar problem with RegressionLabels while changing some examples to the new API | 14:34 |
Farouk | If I use the old API with RegressionLabels it works fine | 14:34 |
Farouk | but if I switch to the new API I get: | 14:34 |
@wiking | wuwei1, i guess your question is what Composite stands for? | 14:34 |
Farouk | [ERROR] In file /tmp/src/shogun/src/shogun/labels/Labels.h line 143: Sorry, not yet implemented . | 14:35 |
shogitter | (vinx13) wiking: yeah | 14:35 |
@wiking | Farouk, do you have part of this pushed into a pr? | 14:35 |
Farouk | No its the examples I've been porting since yesterday. Wait let me push it and I will let you know | 14:36 |
@wiking | wuwei1, ok so that is about somehow be able to compose multiple machines | 14:36 |
@wiking | Farouk, that'd be good thanks | 14:36 |
shogitter | (vinx13) wiking: i'm not very sure about how things work, if not linear but DAG | 14:36 |
@wiking | wuwei1, because say we have Bagging you could do a Bagging(LibSVM, NN) | 14:36 |
@wiking | and then you have an 'averaging' in Bagging | 14:37 |
@wiking | of the result | 14:37 |
shogitter | (vinx13) btw getting some problem with irc client so i'm using gitter now | 14:37 |
shogitter | (vinx13) yeah, that's equivalent to: bagging = new Bagging(); bagging.add(..) right? | 14:39 |
@wiking | vinx13 sure no problem at all | 14:41 |
@wiking | vinx13 exactly | 14:41 |
@wiking | i mean first we can go with the builder pattern | 14:41 |
@wiking | for linear pipelines | 14:41 |
shogitter | (vinx13) yeah | 14:42 |
@wiking | and then we can add later the composite thing | 14:42 |
@wiking | but yeah as you can see | 14:42 |
@wiking | i'm not so sure if we need to support this kind of api | 14:43 |
@wiking | with(trans, IS_FITTED). | 14:43 |
@wiking | i.e. the boolean for fit or not fit | 14:43 |
@wiking | one should assume that it's not fit | 14:43 |
shogitter | (vinx13) we should ignore those IS_FITTED when calling train_machine? | 14:44 |
@wiking | nono | 14:46 |
@wiking | i dont think that api actually is necessary | 14:46 |
@wiking | we just have | 14:46 |
@wiking | with(trans) | 14:46 |
@wiking | and that's all | 14:46 |
@wiking | lets see later if we wanna support that | 14:46 |
@wiking | but i'm really not sure if we need an api like that | 14:47 |
shogitter | (vinx13) sure | 14:47 |
shogitter | (vinx13) btw the composite should only work with a few machines like Bagging ? | 14:47 |
shogitter | (vinx13) ah.. have checked BagingMachine, it doesn't support multiple kinds of machines, maybe we should use combination rule for composite ? | 14:50 |
sukey2 | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/4309 synchronized by FaroukY | 15:02 |
Farouk | wiking, check the synchronized pr in the file rain_forest_regression.sg | 15:03 |
Farouk | RegressionLabels ==> works ok, Labels, get the error | 15:04 |
sukey2 | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/4309 synchronized by FaroukY | 15:19 |
sukey2 | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/4310 synchronized by shubham808 | 15:33 |
sukey2 | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/4285 synchronized by vinx13 | 15:42 |
@wiking | Farouk, ok here sorry | 15:49 |
@wiking | had some meetings | 15:49 |
@wiking | vinx13 yeah that's why i sad for the time being... :) | 15:49 |
@wiking | vinx13 we'll refactor a bit the Bagging Machine | 15:49 |
@wiking | but yeah for the time being just think about pipeline as a linear thing | 15:49 |
@wiking | Farouk, ok i'll check now the pr | 15:49 |
@wiking | Farouk, it's this right: https://github.com/shogun-toolbox/shogun/pull/4309 | 15:50 |
Farouk | wiking, no worries, im here all day~ | 15:50 |
@wiking | Farouk, https://github.com/shogun-toolbox/shogun/pull/4309/files#diff-a89b037ce72a965cb4fdf35083d4897fR9 | 15:50 |
@wiking | sorry | 15:50 |
@wiking | i dont know why this doesn't wanna take a line | 15:50 |
@wiking | ok so Labels labels_train = labels(f_labels_train) | 15:51 |
@wiking | -RegressionLabels labels_test(f_labels_test) | 15:51 |
@wiking | +Labels labels_test = labels(f_labels_test) | 15:51 |
@wiking | is the thing that gives you the erro? | 15:51 |
@wiking | kernel_ridge_regression_nystrom.sg | 15:51 |
@wiking | ? | 15:51 |
sukey2 | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/4312 merged by vigsterkr | 15:53 |
sukey2 | [https://github.com/shogun-toolbox/shogun] New commit https://github.com/shogun-toolbox/shogun/commit/7e9699e5a1c7df75271e8dd26fc71e615e8138b4 by vigsterkr | 15:53 |
@wiking | wuwei1, ok lets see what the CIs are saying for your transformer thingy and then i'll merge | 15:54 |
Farouk | I have no idea what just happened ! | 15:54 |
Farouk | I just tested it again and somehow it works | 15:54 |
Farouk | when it was the new API it just told me not implemented | 15:54 |
shogitter | (vinx13) wiking: sure | 15:55 |
@wiking | Farouk, ok lemme know when you know where it actually happens | 15:55 |
@wiking | and which example can i test | 15:55 |
@wiking | vinx13: https://github.com/shogun-toolbox/shogun/pull/4285/files#r191432677 | 15:55 |
@wiking | :P | 15:55 |
Farouk | no it just worked again! I've been running the test since morning, and its been failing, then i just pushed to github, then recompiled and retested and the tests passed! | 15:56 |
Farouk | wait make meta_examples generates the tests right? | 15:56 |
@wiking | should | 15:57 |
@wiking | lets see what the CIs react | 15:57 |
@wiking | :) | 15:57 |
shogitter | (vinx13) wiking: i will rename apply -> transform now :) | 15:57 |
@wiking | vinx13 :) thnx | 15:57 |
@wiking | vinx13 have you had a chance to think about the ngraph story? :) | 15:58 |
shogitter | (vinx13) had a quick look, the apis look good | 15:59 |
shogitter | (vinx13) i'm thinking about cpu/gpu things | 15:59 |
@wiking | yeah it's much smoother/clearer | 15:59 |
@wiking | vinx13 i mean they kind of like solved that :) | 15:59 |
@wiking | because the way they create tensors (matrix sgvector etc) | 16:00 |
@wiking | that actually you call the backend | 16:00 |
@wiking | to do that | 16:00 |
@wiking | so they have Backend::create_tensor(Type, Shape) | 16:00 |
@wiking | so that assures you that it's on the right device | 16:00 |
@wiking | because if its a GPUBackend | 16:00 |
@wiking | you'll get a tensor on the GPU | 16:00 |
@wiking | etc | 16:00 |
@wiking | it's a nice and clean design imo | 16:00 |
shogitter | (vinx13) yeah, i'm thinking how we deal with those operations like index on matrix | 16:01 |
@wiking | well | 16:01 |
@wiking | you know | 16:01 |
@wiking | that's the trick here | 16:01 |
@wiking | you actually dont do direct operations on the matrix :) | 16:01 |
@wiking | everything is a Function | 16:01 |
shogitter | (vinx13) of course | 16:01 |
shogitter | (vinx13) but unfortunately algos need that, for now | 16:02 |
@wiking | yeah but we can start with a simple algo | 16:03 |
shogitter | (vinx13) i agree that operations should implemented as objects | 16:03 |
@wiking | and try to have a design for that | 16:03 |
@wiking | because i mean i hope you've seen | 16:03 |
@wiking | that in ngraph | 16:03 |
@wiking | actually the operations are nothing else | 16:03 |
@wiking | but writing C++ code :D | 16:03 |
@wiking | literally | 16:03 |
@wiking | op = streamstream << "Eigen...." << ..... << std::endll; | 16:04 |
@wiking | and then these string streams are compiled with an JIT :) | 16:04 |
@wiking | so it's a bit crazy | 16:04 |
-!- HeikoS [~heiko@host81-153-166-11.range81-153.btcentralplus.com] has joined #shogun | 16:04 | |
-!- mode/#shogun [+o HeikoS] by ChanServ | 16:04 | |
@wiking | but this way it's highly optimized | 16:04 |
@wiking | we dont need this atm | 16:04 |
@wiking | it'd be better already to get rid of those macro generated | 16:04 |
@wiking | virtual methods | 16:04 |
@wiking | and not to have that interface an ever extending thing | 16:05 |
@wiking | meaning even if we have a container that is typeless | 16:05 |
@wiking | the linalg api is bound to be ever extending if we dont support thing like Function :) | 16:05 |
@wiking | as then we'll add things to it ad-hoc | 16:05 |
@wiking | (various decomposers etc.) | 16:05 |
@wiking | Farouk, ok the cis are done | 16:06 |
@wiking | lemme check it | 16:06 |
@wiking | or at least travis is done | 16:06 |
Farouk | Is there a problem with Java and C# ? | 16:07 |
@wiking | Farouk, https://travis-ci.org/shogun-toolbox/shogun/jobs/385195492#L3253 | 16:07 |
@wiking | https://travis-ci.org/shogun-toolbox/shogun/jobs/385195493#L2888 | 16:07 |
@wiking | differnet places it fails | 16:07 |
@wiking | in case of java | 16:07 |
@wiking | it's about EvaluationResult casting | 16:07 |
@wiking | in case of c# it's | 16:08 |
@wiking | ): error CS0266: Cannot implicitly convert type `Kernel' to `CombinedKernel'. An explicit conversion exists (are you missing a cast?) | 16:08 |
@wiking | https://github.com/shogun-toolbox/shogun/pull/4309/files#diff-850db6d5082e9cf6bbf0ddb8e5ad2ab3R60 | 16:08 |
@wiking | it should be | 16:09 |
@wiking | Kernel k = kernel(machine_mkl.get("kernel")) | 16:09 |
@wiking | instead of CombinedKernel k = kernel(machine_mkl.get("kernel")) | 16:09 |
@wiking | RealVector w = k.get_subkernel_weights() | 16:09 |
@wiking | no? | 16:09 |
Farouk | Ohh yea! Sorry about that! | 16:10 |
Farouk | for CrossValidationResult result = cross.evaluate(), do I need an obtain from generic? | 16:14 |
@wiking | just submitted a mini review | 16:15 |
@wiking | but yeah | 16:15 |
sukey2 | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/4309 synchronized by FaroukY | 16:29 |
-!- sukey2 [~nodebot@ks312251.kimsufi.com] has quit [Remote host closed the connection] | 16:35 | |
-!- sukey [~nodebot@ks312251.kimsufi.com] has joined #shogun | 16:35 | |
-!- mode/#shogun [+o sukey] by ChanServ | 16:35 | |
@wiking | sukey, flip | 16:35 |
@sukey | (/?? ? ?)/? ~ ??? | 16:35 |
-!- shubham808 [0e8bf0fb@gateway/web/cgi-irc/kiwiirc.com/ip.14.139.240.251] has joined #shogun | 17:16 | |
shubham808 | HeikoS hey | 17:16 |
@HeikoS | shubham808: hi! | 17:16 |
shubham808 | so lets start with the premature test stuff | 17:17 |
@HeikoS | shubham808: so tomorrow 15 utc is good | 17:17 |
@HeikoS | lets meet here | 17:17 |
@HeikoS | alternatively I also have some time now | 17:17 |
@HeikoS | ok | 17:17 |
shubham808 | now is good | 17:17 |
shubham808 | for me too | 17:17 |
shubham808 | so | 17:18 |
shubham808 | there is a pr to drop jinja2 from serialization tests | 17:18 |
shubham808 | we could add our own test there | 17:20 |
shubham808 | like: https://pastebin.com/wS90K7Gx | 17:21 |
shubham808 | u see line 46 ? | 17:22 |
shubham808 | *43 | 17:22 |
@HeikoS | checking | 17:25 |
@HeikoS | yes thats a good idea | 17:26 |
-!- travis-ci [~travis-ci@ec2-184-73-117-198.compute-1.amazonaws.com] has joined #shogun | 17:26 | |
travis-ci | it's Abhishek Kumar Srivastava's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: https://travis-ci.org/shogun-toolbox/shogun/builds/385211230 | 17:26 |
-!- travis-ci [~travis-ci@ec2-184-73-117-198.compute-1.amazonaws.com] has left #shogun [] | 17:26 | |
@HeikoS | shubham808: thats a good thing to work towards to, need to merge that PR soon then I guess? | 17:26 |
shubham808 | yeah | 17:26 |
shubham808 | but there are still a few problems | 17:27 |
shubham808 | first is the continue/break thing in the macro | 17:28 |
shubham808 | next is the ASSERT_TRUE(i) | 17:28 |
shubham808 | this is to make sure we are stopped by the callback | 17:31 |
shubham808 | but we will not trigger this if the algorithm terminates in a single iteration | 17:34 |
shubham808 | any ideas for a work around ? | 17:34 |
@HeikoS | shubham808: sorry something came up | 17:53 |
@HeikoS | I think algorithms that terminate within a single iteration should not be tested | 17:53 |
@HeikoS | as it doesnt make sense | 17:53 |
shubham808 | it depends on the dataset | 17:54 |
shubham808 | i think | 17:54 |
shubham808 | yeah we could just not test those | 17:55 |
@HeikoS | do you have an example? | 17:57 |
@HeikoS | like an algorithm that terminates immediately? | 17:57 |
@HeikoS | I have no feel;ing for what happens in that case | 17:57 |
@HeikoS | shubham808: but I think for now, we could just pass/skip the test | 17:57 |
@HeikoS | we can deal with this problem later | 17:57 |
shubham808 | alright | 17:58 |
shubham808 | now about the algorithms we will be testing | 18:00 |
@HeikoS | yep | 18:00 |
@HeikoS | shubham808: so? | 18:01 |
shubham808 | let me find the line link | 18:01 |
shubham808 | just a sex | 18:01 |
shubham808 | *sec | 18:01 |
shubham808 | https://github.com/micmn/shogun/blob/11b1a342c96a8c1b51c5c8514fd0cb94c09cac88/tests/unit/base/trained_model_serialization_unittest.cc.py#L130 | 18:03 |
shubham808 | i have added another type list next to it | 18:03 |
shubham808 | we will be limited by the number of algorithms the serialization test works on | 18:04 |
shubham808 | as a start i made the new type list identical to Machine list | 18:05 |
-!- travis-ci [~travis-ci@ec2-54-81-11-238.compute-1.amazonaws.com] has joined #shogun | 18:06 | |
travis-ci | it's Abhishek Kumar Srivastava's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: https://travis-ci.org/shogun-toolbox/shogun/builds/385211230 | 18:06 |
-!- travis-ci [~travis-ci@ec2-54-81-11-238.compute-1.amazonaws.com] has left #shogun [] | 18:06 | |
@HeikoS | ok sure | 18:07 |
@HeikoS | seems ok | 18:07 |
@HeikoS | (for now) | 18:07 |
shubham808 | yeah | 18:07 |
shubham808 | for now | 18:07 |
shubham808 | alright this should be enough for a wip pr | 18:07 |
shubham808 | we could rebase it later when we merge the drop jinja2 pr | 18:08 |
@HeikoS | yep exactly | 18:08 |
@HeikoS | Im keen on getting that one in | 18:08 |
@HeikoS | probably tomorrow or so | 18:08 |
shubham808 | okay cool | 18:08 |
@HeikoS | \ | 18:09 |
@HeikoS | shubham808: I will have to run soon, shall we still have our meeting tomorrow? | 18:10 |
@HeikoS | To discuss next steps and stuff? | 18:10 |
shubham808 | yes sure | 18:10 |
shubham808 | bye | 18:11 |
@HeikoS | shubham808: thx! | 18:11 |
@HeikoS | Ill have more time tomorrow! | 18:11 |
shubham808 | great :) | 18:12 |
@HeikoS | just catching up with all those emails >( | 18:12 |
@HeikoS | :) | 18:12 |
-!- shubham808 [0e8bf0fb@gateway/web/cgi-irc/kiwiirc.com/ip.14.139.240.251] has quit [Quit: http://www.kiwiirc.com/ - A hand crafted IRC client] | 18:17 | |
@sukey | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/4310 synchronized by shubham808 | 18:22 |
@sukey | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/4313 opened by karlnapf | 18:42 |
@sukey | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/4285 synchronized by vinx13 | 19:41 |
-!- HeikoS [~heiko@host81-153-166-11.range81-153.btcentralplus.com] has quit [Ping timeout: 240 seconds] | 20:33 | |
-!- HeikoS [~heiko@host81-153-166-11.range81-153.btcentralplus.com] has joined #shogun | 20:41 | |
-!- mode/#shogun [+o HeikoS] by ChanServ | 20:41 | |
-!- HeikoS [~heiko@host81-153-166-11.range81-153.btcentralplus.com] has quit [Ping timeout: 256 seconds] | 20:49 | |
-!- Farouk [9cc5aeaf@gateway/web/freenode/ip.156.197.174.175] has quit [Ping timeout: 260 seconds] | 21:40 | |
--- Log closed Wed May 30 00:00:28 2018 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!