--- Log opened Fri Mar 10 00:00:58 2017 | ||
-!- wudayoda_ [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Read error: Connection reset by peer] | 00:45 | |
-!- wudayoda [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 00:46 | |
-!- wudayoda [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Remote host closed the connection] | 01:41 | |
-!- wudayoda [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 02:23 | |
-!- wudayoda [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Remote host closed the connection] | 03:06 | |
-!- wudayoda [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 03:47 | |
-!- wudayoda [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 240 seconds] | 03:52 | |
-!- wudayoda [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 04:40 | |
-!- wudayoda [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 240 seconds] | 04:49 | |
-!- wudayoda [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 06:46 | |
-!- wudayoda [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 240 seconds] | 06:50 | |
-!- wudayoda [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 06:59 | |
-!- wudayoda [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 256 seconds] | 07:04 | |
-!- rads23 [8984030c@gateway/web/freenode/ip.137.132.3.12] has joined #shogun | 07:41 | |
-!- rads23 [8984030c@gateway/web/freenode/ip.137.132.3.12] has quit [Ping timeout: 260 seconds] | 07:50 | |
-!- suhas2go [uid201652@gateway/web/irccloud.com/x-rezivwilyytbgaks] has joined #shogun | 07:51 | |
-!- wudayoda [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 09:00 | |
-!- ashu [616375c8@gateway/web/freenode/ip.97.99.117.200] has joined #shogun | 09:56 | |
-!- ashu [616375c8@gateway/web/freenode/ip.97.99.117.200] has quit [Client Quit] | 09:57 | |
-!- rads23 [8984030a@gateway/web/freenode/ip.137.132.3.10] has joined #shogun | 10:14 | |
-!- suhas2go [uid201652@gateway/web/irccloud.com/x-rezivwilyytbgaks] has quit [Quit: Connection closed for inactivity] | 10:16 | |
-!- rads23 [8984030a@gateway/web/freenode/ip.137.132.3.10] has quit [Quit: Page closed] | 10:39 | |
CaBa | wiking: how was the presentation? | 11:06 |
---|---|---|
@wiking | good | 11:06 |
@wiking | i should share the presentation somewhere | 11:10 |
@wiking | ;) | 11:10 |
-!- geektoni [~Mutter@93-34-234-212.ip52.fastwebnet.it] has joined #shogun | 11:32 | |
-!- geektoni [~Mutter@93-34-234-212.ip52.fastwebnet.it] has quit [Client Quit] | 11:35 | |
-!- ShellcatZero [~ShellcatZ@cpe-76-93-136-67.san.res.rr.com] has quit [Read error: Connection reset by peer] | 12:00 | |
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has joined #shogun | 12:02 | |
-!- mode/#shogun [+o HeikoS] by ChanServ | 12:02 | |
-!- ShellcatZero [~ShellcatZ@cpe-76-93-136-67.san.res.rr.com] has joined #shogun | 12:03 | |
@wiking | HeikoS, y0 | 12:03 |
@HeikoS | jojo | 12:03 |
@HeikoS | wiking: how are things? | 12:03 |
@wiking | tired | 12:03 |
@wiking | but good | 12:04 |
@HeikoS | good :D | 12:04 |
@wiking | lemme know what is the expected number of people | 12:04 |
@HeikoS | 30 | 12:04 |
@wiking | look into that stuff | 12:04 |
@wiking | ohohoooo | 12:04 |
@wiking | when? | 12:04 |
@HeikoS | tuesday | 12:04 |
@wiking | kk | 12:04 |
@wiking | no problemo | 12:04 |
@wiking | btw | 12:04 |
@HeikoS | can we test this somehow? | 12:04 |
@wiking | what? | 12:04 |
@HeikoS | would be a bit nervous not doing that | 12:04 |
@wiking | i mean there's that | 12:04 |
@HeikoS | 30 ppl logging on | 12:04 |
@HeikoS | and running stuff | 12:05 |
@wiking | ah well u have 30 github accounts? :D | 12:05 |
@HeikoS | no | 12:05 |
@wiking | k | 12:05 |
@wiking | then no | 12:05 |
@wiking | but i mean | 12:05 |
@wiking | it is simple | 12:05 |
@wiking | just there are some stuff | 12:05 |
@wiking | u should clarify | 12:05 |
@wiking | a) is there any special stuff u need on shogun feature wise? | 12:05 |
@HeikoS | I will finish the notebook on the weekend | 12:05 |
@wiking | b) other then some notebooks do you wanna change the docker container's content | 12:05 |
@HeikoS | what happens if I do need to push something to develop and that needs to be in there? | 12:06 |
@wiking | c) how much resources one would need for running that stuff? | 12:06 |
@HeikoS | ressources are minimal, like a laptop powered thing per person | 12:06 |
@HeikoS | (same computing power) | 12:06 |
@wiking | HeikoS, nono | 12:06 |
@wiking | define | 12:06 |
@wiking | :D | 12:06 |
@wiking | not like some vague shit | 12:06 |
@wiking | define | 12:06 |
@wiking | vpcu | 12:06 |
@wiking | *vcpu | 12:06 |
@wiking | and ram | 12:06 |
@HeikoS | how can I? | 12:06 |
@wiking | https://github.com/shogun-toolbox/docs/pull/5 | 12:06 |
@wiking | this is what happens when u push shit into develop | 12:07 |
@wiking | we have nightly debian packages | 12:07 |
@HeikoS | ah cool will review | 12:07 |
@HeikoS | ok then | 12:07 |
@wiking | of course i can update them faster | 12:07 |
@wiking | than nightly | 12:07 |
@wiking | i mean anytime | 12:07 |
@HeikoS | if I push on Monday, is that in there Tuesday? | 12:07 |
@wiking | on demand :) | 12:07 |
@HeikoS | cool | 12:07 |
@wiking | man | 12:07 |
@wiking | shogun-buildbot dance | 12:07 |
@wiking | shogun-buildbot: dance | 12:07 |
shogun-buildbot | <(^.^<) | 12:07 |
shogun-buildbot | <(^.^)> | 12:07 |
shogun-buildbot | (>^.^)> | 12:07 |
shogun-buildbot | (7^.^)7 | 12:07 |
@HeikoS | good | 12:07 |
shogun-buildbot | (>^.^<) | 12:07 |
@wiking | so yeah he will dance when you want | 12:07 |
@HeikoS | ok so ressources | 12:07 |
@wiking | and generate package when u instruct it | 12:08 |
@wiking | but by default it's nightly | 12:08 |
@HeikoS | I think 4GB ram and 2cpu per person for the afternoon | 12:08 |
@HeikoS | thats way more than is needed but a good guideline | 12:08 |
@HeikoS | is that good? | 12:08 |
-!- mikeling [uid89706@gateway/web/irccloud.com/x-clbrxoaqgcvnoeqc] has joined #shogun | 12:09 | |
@wiking | HeikoS, you know | 12:10 |
@wiking | i dont :) | 12:10 |
@HeikoS | wiking: I mean as in is that specified enough? | 12:11 |
@HeikoS | the ressources will be enough | 12:12 |
@wiking | k | 12:12 |
@HeikoS | wiking: and re the saving stuff | 12:12 |
@wiking | ? | 12:14 |
@HeikoS | so you said that will all go then, i.e. people's notebook will be saved under their "account" | 12:15 |
@HeikoS | ? | 12:15 |
@HeikoS | wiking: another question: what happens if there is a problem (i.e. the service doenst work or so) | 12:15 |
@wiking | what do yo umean? | 12:18 |
@wiking | i mean what happens usually when a service is down? :D | 12:18 |
@HeikoS | wiking: I meant: can I call you? :D | 12:18 |
@wiking | HeikoS, well yeah the saving thing i have a solution... just currenlty it's too slow | 12:18 |
@wiking | that's why it's now as it is | 12:18 |
@HeikoS | ok | 12:18 |
@wiking | but i'll try to fix this over the weekend | 12:19 |
@HeikoS | cool | 12:19 |
@wiking | if the service goes down | 12:19 |
@wiking | u can call me | 12:19 |
@HeikoS | wiking: whats you opinion on dataset, we usually have them on some server and people download them by hand | 12:19 |
@wiking | ? | 12:19 |
@wiking | it's there | 12:19 |
@wiking | now | 12:19 |
@HeikoS | http://www.gatsby.ucl.ac.uk/~gretton/adaptModel/ShortCourseLab.pdf | 12:19 |
@wiking | we can do it the same way | 12:20 |
@wiking | just mount it in | 12:20 |
@wiking | read only | 12:20 |
@wiking | btw | 12:20 |
@HeikoS | okok | 12:20 |
@wiking | we need a different domain | 12:20 |
@wiking | btw | 12:20 |
@wiking | just to be clear | 12:20 |
@wiking | that shogun | 12:20 |
@wiking | has no svmlight | 12:20 |
@HeikoS | yes | 12:21 |
@HeikoS | no worries | 12:21 |
@wiking | you are aware of it right? | 12:21 |
@HeikoS | just need any kernel svm | 12:21 |
@wiking | k | 12:21 |
@HeikoS | and naive bayes should work, but I will check that | 12:21 |
@wiking | kk | 12:21 |
@wiking | i mean go through the thing | 12:21 |
@wiking | if u need 2vcpu + 4 gis of ram | 12:22 |
@wiking | that is | 12:22 |
@wiking | 60 cores and 120 gig ram | 12:22 |
@wiking | right | 12:22 |
@wiking | ? | 12:22 |
@HeikoS | but that is auto allocated right? | 12:22 |
@HeikoS | yes | 12:22 |
@HeikoS | I dont know *exact* number of people, but this will be more that sufficient | 12:22 |
@wiking | what is auto located/ | 12:22 |
@wiking | ? | 12:22 |
@HeikoS | the number of instances | 12:22 |
@HeikoS | running | 12:22 |
@wiking | in what sense? | 12:22 |
@wiking | yeah they should scale | 12:22 |
@HeikoS | like if only 10 people show up | 12:22 |
@wiking | but it takes some time to scale up an instance | 12:23 |
@HeikoS | we dont run 60 cores the while time? | 12:23 |
@wiking | like 2 minutes or so | 12:23 |
@HeikoS | okok | 12:23 |
@wiking | so it's better have some stuff already running there | 12:23 |
@HeikoS | I understand | 12:23 |
@HeikoS | will it turn itself off after a while if not used? | 12:23 |
@wiking | yes it should downscale | 12:23 |
@HeikoS | ok cool | 12:26 |
@HeikoS | this will be pretty awesome | 12:26 |
@HeikoS | see all the mex shit in the pdf? | 12:26 |
@wiking | btw | 12:26 |
@wiking | yes | 12:26 |
@wiking | that is ugly | 12:26 |
@wiking | so | 12:26 |
@wiking | do u think | 12:26 |
@wiking | we should do a benchmark on this: https://aws.amazon.com/ec2/instance-types/x1/ | 12:26 |
@wiking | :) | 12:26 |
@wiking | x1.32xlarge = 1,952 Gb of ram | 12:27 |
@HeikoS | oh wow | 12:27 |
@wiking | it'd be cool to do a test with http://labs.criteo.com/2013/12/download-terabyte-click-logs/ | 12:27 |
@wiking | it's 13 usd/h | 12:27 |
@HeikoS | we have lots of funds still right? :) | 12:27 |
@wiking | yes | 12:28 |
@wiking | we cannot index that amount of data though :) | 12:28 |
@wiking | i have BigInput branch | 12:28 |
@wiking | it's a lot of fucking replace for int32_t -> index_t | 12:28 |
@wiking | but would be cool to have a benchmark | 12:28 |
@HeikoS | agreed | 12:28 |
@wiking | to see not only accuracy | 12:29 |
@wiking | but time | 12:29 |
@wiking | :) | 12:29 |
@wiking | compared to other horizontally scalable stuff | 12:29 |
@HeikoS | wiking: btw | 12:30 |
@HeikoS | what do we need to release ? :D | 12:30 |
@wiking | mmm | 12:30 |
@wiking | fix for cloud | 12:30 |
@wiking | and there are all the read shit | 12:31 |
@wiking | *red shit | 12:31 |
@wiking | in the buildbot | 12:31 |
@wiking | nightlies are failing | 12:31 |
@wiking | PRs are failing | 12:31 |
@wiking | documentations are failing | 12:31 |
@HeikoS | okok | 12:31 |
@HeikoS | will fix some of that | 12:31 |
@HeikoS | what do you think about opencv notebook? | 12:31 |
@HeikoS | needs some installation update | 12:31 |
@HeikoS | of opencv as they split some things out of the lib | 12:31 |
@HeikoS | and cmake newer version for PRE | 12:31 |
@HeikoS | PR | 12:31 |
@wiking | ah | 12:32 |
@wiking | and still have no idea | 12:32 |
@wiking | what to do with octave | 12:32 |
@wiking | and we still hve some stuff going on | 12:32 |
@wiking | with CV | 12:32 |
@HeikoS | yeah | 12:32 |
@HeikoS | tests pass | 12:32 |
@wiking | http://buildbot.shogun-toolbox.org/builders/nightly_default/builds/60/steps/test/logs/stdio | 12:32 |
@wiking | here they dont | 12:33 |
@HeikoS | oh wow didnt realise | 12:33 |
@wiking | and there's omething | 12:33 |
@wiking | with nihgtly_none | 12:33 |
@HeikoS | well ok that needs fi | 12:33 |
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has left #shogun [] | 12:33 | |
@wiking | that rahul was working on | 12:33 |
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has joined #shogun | 12:33 | |
-!- mode/#shogun [+o HeikoS] by ChanServ | 12:33 | |
@HeikoS | wiking: tweet about the cloud when done! :) | 12:37 |
@wiking | yeye | 12:38 |
@wiking | i deliberately didn't | 12:38 |
@wiking | till it's all nice and good | 12:38 |
@wiking | btw | 12:38 |
@wiking | what is the plan | 12:38 |
@wiking | shouldn't we have like | 12:38 |
@wiking | R, java, scala and maybe c++ kernels | 12:38 |
@wiking | for jupyter? | 12:38 |
@wiking | all in one? :) | 12:38 |
@HeikoS | yeah that would be cool | 12:38 |
@HeikoS | so cool in fact | 12:38 |
@HeikoS | though we dont have notebook demos yet | 12:39 |
@HeikoS | but well | 12:39 |
@HeikoS | still cool to run shogun | 12:39 |
@wiking | yes | 12:39 |
@wiking | somebody should test | 12:39 |
@wiking | whether it's possible to have | 12:39 |
@wiking | multiple kernels | 12:39 |
@wiking | i know you can have | 12:39 |
@wiking | python2 and 3 | 12:39 |
@HeikoS | yes that definitely | 12:39 |
@HeikoS | I dont know much aobut jupyter and the other kernels | 12:40 |
@HeikoS | meta notebooks would be so cool | 12:40 |
@wiking | https://github.com/michhar/rpy2_sample_notebooks/blob/master/TestingRpy2.ipynb | 12:41 |
@wiking | but this is not really | 12:41 |
@wiking | i mean i dont want to mix | 12:41 |
@wiking | language in one notebook | 12:42 |
@wiking | just to have the option | 12:42 |
@wiking | of different language | 12:42 |
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has left #shogun [] | 12:42 | |
@wiking | in different notebook | 12:42 |
@wiking | s | 12:42 |
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has joined #shogun | 12:42 | |
-!- mode/#shogun [+o HeikoS] by ChanServ | 12:42 | |
@sukey | Wiki page: Google-Summer-of-Code-2017-Projects edited on shogun-toolbox/shogun by karlnapf | 12:47 |
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has quit [Ping timeout: 240 seconds] | 12:55 | |
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has joined #shogun | 12:58 | |
-!- mode/#shogun [+o HeikoS] by ChanServ | 12:58 | |
@wiking | HeikoS, y0 | 13:59 |
@HeikoS | jo | 13:59 |
@wiking | what do u think we should be more mem heavy | 13:59 |
@wiking | or cpu heavy? | 13:59 |
@wiking | with the instances? | 13:59 |
@HeikoS | mem wont be too large | 13:59 |
@wiking | in general | 13:59 |
@wiking | not about | 13:59 |
@HeikoS | in general | 13:59 |
@HeikoS | phew | 13:59 |
@HeikoS | not sure | 13:59 |
@wiking | the course | 13:59 |
@wiking | currenlty | 14:00 |
@wiking | u get | 14:00 |
@wiking | 1 vpcu | 14:00 |
@HeikoS | it doesnt need to be super fast | 14:00 |
@wiking | and 3 gigs of ram | 14:00 |
@HeikoS | yeah thats fine I think | 14:00 |
@HeikoS | we can adjust later | 14:00 |
@HeikoS | if people complain? | 14:00 |
@wiking | mmm of course | 14:00 |
@wiking | the problem is this | 14:00 |
@wiking | https://aws.amazon.com/ec2/instance-types/ | 14:00 |
@wiking | currenlty i use m4.xlarge | 14:01 |
@wiking | but maybe we should be using c4.xlarge | 14:01 |
@wiking | see what i mean? | 14:02 |
@wiking | because m4.xlarge = 4 people roughly.... | 14:02 |
@HeikoS | yeah I see | 14:02 |
@HeikoS | I think we cannot really decide this now | 14:02 |
@wiking | nono | 14:02 |
@wiking | we can | 14:02 |
@wiking | later | 14:02 |
@wiking | how this is devided up | 14:02 |
@wiking | divided | 14:02 |
@wiking | that's another story | 14:03 |
@wiking | but i have a fealing | 14:03 |
@wiking | *feeling | 14:03 |
@wiking | that we are too heavy per instance | 14:03 |
@wiking | on memory | 14:03 |
@wiking | it's not being used | 14:03 |
@wiking | while cpu is .......well not enough | 14:03 |
@HeikoS | yeah | 14:04 |
@HeikoS | true | 14:04 |
@HeikoS | model selection and stuff | 14:04 |
@HeikoS | people wont do huge datasets in there | 14:04 |
@HeikoS | so agreed | 14:04 |
@wiking | i have an arm64 at hom atm | 14:12 |
@wiking | should try to get shogun running on it :) | 14:12 |
@wiking | i wonder if it works | 14:12 |
@HeikoS | phew :) | 14:21 |
@HeikoS | maybe a good idea | 14:21 |
@HeikoS | though I wonder about priorities :D | 14:21 |
@HeikoS | probably wont work | 14:21 |
@HeikoS | haha | 14:21 |
@HeikoS | but I love more architechtures and OS | 14:21 |
@wiking | HeikoS, why not? | 14:27 |
@wiking | i mean | 14:27 |
@wiking | on fedora | 14:28 |
@wiking | i think we have all that compiled | 14:28 |
@wiking | https://koji.fedoraproject.org/koji/buildinfo?buildID=859779 | 14:29 |
@wiking | there | 14:29 |
@wiking | shogun-5.0.0-0.5.fc26.aarch64.rpm | 14:30 |
@wiking | :) | 14:30 |
@HeikoS | wiking: jo | 14:57 |
@wiking | po | 14:58 |
@HeikoS | wiking: just talked to arthur a bit | 14:58 |
@HeikoS | about getting some $$$ to shogun for setting up the cloud for the course | 14:58 |
@wiking | good :) | 14:58 |
@wiking | i should write the blog post soon | 14:58 |
@HeikoS | if you can estimate number of hours | 14:58 |
@wiking | about the whole story | 14:58 |
@wiking | mmm | 14:58 |
@HeikoS | you did | 14:58 |
@wiking | we can do the accounting | 14:58 |
@HeikoS | and me too | 14:58 |
@wiking | aaah | 14:58 |
@HeikoS | there is two ways | 14:58 |
@wiking | noo it's ok | 14:58 |
@wiking | i can give you the aws accoutning | 14:58 |
@HeikoS | one is: we write an invoice over numfocus (in the future) | 14:58 |
@HeikoS | other option is: we are hired by UCL CS department | 14:59 |
@wiking | mmmm look | 14:59 |
@HeikoS | (thats how I am paid for the course stuff I do) | 14:59 |
@wiking | for me it's only thing is | 14:59 |
@HeikoS | and then can donate to shogun account | 14:59 |
@HeikoS | point is to get something for it | 14:59 |
@wiking | that i would like to have this for future shit for shogun | 14:59 |
@wiking | liek | 14:59 |
@wiking | sonney2k, have u reimbursed | 14:59 |
@HeikoS | yeah I see, just sayintg it is a nice side effect | 14:59 |
@wiking | my fossasia stuff? :D | 14:59 |
@wiking | i guess i should ping him via email/faceboko | 15:00 |
@HeikoS | so we can definitely do that via ucl human ressources and then each of us donates | 15:00 |
@wiking | mmm | 15:00 |
@wiking | if they can hold up with shit | 15:00 |
@HeikoS | and even cooler would be if ucl donates to shogun for that | 15:00 |
@wiking | then lets wait till april | 15:00 |
@HeikoS | yeah agreed | 15:00 |
@wiking | and by then | 15:00 |
@HeikoS | just wanted to share | 15:00 |
@wiking | we know something about numfocus | 15:00 |
@HeikoS | since I just discussed | 15:00 |
@HeikoS | sure sure | 15:00 |
@wiking | kk | 15:00 |
-!- lambday [6a3311c5@gateway/web/freenode/ip.106.51.17.197] has joined #shogun | 15:13 | |
-!- mode/#shogun [+o lambday] by ChanServ | 15:13 | |
@lambday | HeikoS: yo | 15:13 |
@lambday | HeikoS: agreed on reallocation.. removing that... | 15:14 |
@lambday | should fail if not enough memory is there | 15:14 |
@HeikoS | lambday: jojo | 15:15 |
@HeikoS | lambday: just check num rows columns | 15:15 |
@HeikoS | caller responsibility | 15:15 |
@lambday | yeah | 15:15 |
@lambday | this one is way too flexible.. inviting callers to screw things up :D | 15:16 |
@HeikoS | REQUIRE can help | 15:16 |
@lambday | BTW did you notice that this patch reduced the number of memallocs from N to 1? | 15:16 |
@HeikoS | I mean if somebody gives a matrix with less memory allocated than in num_cols they deserve a segfault | 15:16 |
@HeikoS | nope, can you poiunt me to it | 15:17 |
@lambday | the way it worked before... you allocate the total memory once.. then copy the current.... then copy the return values of the get_feature_matrix() of other features.. which themselves did one memalloc extra... | 15:18 |
@lambday | now, since you can pass pre-allocated things, the other features don't alloc those matrices, rather uses the pre-allocated one, with column offset :) | 15:19 |
@lambday | HeikoS: https://github.com/lambday/shogun/blob/221cdcac33b7b08899322f3060e692b248f768c0/src/shogun/features/DenseFeatures.cpp#L1061 | 15:19 |
@HeikoS | lambday: jojo | 15:25 |
@HeikoS | lambday: hey I have a question | 15:25 |
@HeikoS | or more like wiking has | 15:26 |
@lambday | HeikoS: yeah | 15:26 |
@HeikoS | he wants to showcase some of the fast testing speed stuff at a talk next week | 15:26 |
@HeikoS | so we was asking for code to reproduce the plots in the paper | 15:26 |
@HeikoS | so the code for that is kinda messed up / not in a state to share(/reproduce | 15:26 |
@lambday | HeikoS: it's on gatsby's machine | 15:26 |
@HeikoS | but maybe you can point viktor to the script that does the comparisons? | 15:26 |
@lambday | yeah | 15:26 |
@HeikoS | lambday: like especially comparing old shogun vs new shogun | 15:27 |
@lambday | HeikoS: alright.. maybe we can have a separate script for that one.. which takes two libshogun.so as params... | 15:27 |
@lambday | so, the speed-up with number of threads won't help.. he needs something that he can compare on single core.. | 15:28 |
@lambday | otherwise it's not just to the old code | 15:28 |
@HeikoS | lambday: so optimal world: you clean up the scripts so that there is a single clean one that reproduces the plots | 15:30 |
@wiking | mmm | 15:30 |
@HeikoS | lambday: and then wiking can take this script and modify it to generate some cool vis for the presentation | 15:30 |
@wiking | if possible | 15:30 |
@wiking | just a string of data | 15:31 |
@wiking | is much better | 15:31 |
@wiking | no need for the graph | 15:31 |
@wiking | :D | 15:31 |
@wiking | just tell me what's axis x and y | 15:31 |
@wiking | and the data points | 15:31 |
@wiking | :D | 15:31 |
@wiking | after that i can generate the graphs | 15:31 |
@lambday | haha lovely :D | 15:31 |
@wiking | of course | 15:31 |
@wiking | i'm happy to modify the sciprt as well | 15:31 |
@HeikoS | lambday: we need these scripts anyways | 15:31 |
@wiking | whichever is the easiest atm | 15:31 |
@lambday | HeikoS: I agree.. but remember we wanted to do it using pandas | 15:31 |
@HeikoS | lambday: in dougals git | 15:31 |
@wiking | pandas | 15:32 |
@wiking | fuck python | 15:32 |
@HeikoS | lambday: nah, text file stuff is ok | 15:32 |
@wiking | :D | 15:32 |
@HeikoS | lambday: just no piping :) | 15:32 |
@HeikoS | just a script that generates the data for the plot | 15:32 |
@HeikoS | can take a while is OK | 15:32 |
@HeikoS | but just reproduce | 15:32 |
@HeikoS | should be minimal effort no? | 15:32 |
@lambday | gotta see.. do I still have the login? :P | 15:32 |
@HeikoS | dougal also wanted to re-create the plot for camera ready version | 15:32 |
@HeikoS | lambday: you do | 15:32 |
@lambday | HeikoS: cool then.. will clean things up | 15:33 |
@HeikoS | if you could put the stuff on your github somewhere? Then we could copy it | 15:33 |
@HeikoS | cool man thx | 15:33 |
@lambday | HeikoS: yeah sure | 15:33 |
@lambday | HeikoS: u435a was it? | 15:33 |
@HeikoS | lambday: doesnt matter filesystem is the same | 15:33 |
@HeikoS | i am on sperry | 15:33 |
@lambday | no but if I wanted to run.. | 15:34 |
@HeikoS | u435a is the machine you used | 15:34 |
@HeikoS | sperry then | 15:34 |
@lambday | HeikoS: cool let me try loggin in | 15:34 |
-!- suhas2go [uid201652@gateway/web/irccloud.com/x-tptznqwnhoqaakye] has joined #shogun | 15:34 | |
@HeikoS | lambday: let me know if you need help or something | 15:35 |
@lambday | HeikoS: working | 15:35 |
@lambday | HeikoS: sure | 15:35 |
@HeikoS | cool | 15:35 |
@lambday | HeikoS: shall I merge the patch first, once I rename the method and remove the reallocation feature? | 15:36 |
@lambday | HeikoS: also, should we really make this method public? | 15:36 |
@lambday | I have kept it protected | 15:36 |
@HeikoS | lambday: why? | 15:36 |
@HeikoS | ah | 15:36 |
@HeikoS | yeah why not | 15:36 |
@HeikoS | might be useful to copy out the subsetted matrix | 15:36 |
@HeikoS | but | 15:37 |
@HeikoS | maybe only the one that doesnt accept pre-aloocated | 15:37 |
@HeikoS | otherwise messy | 15:37 |
@HeikoS | or hide from SWIG | 15:37 |
@HeikoS | or just leave protected | 15:37 |
@lambday | HeikoS: wonder how the typemap works for SGMatrix.. does it use the same memory that, say, numpy passes or reallocates and copies in SGMatrix | 15:37 |
@HeikoS | nono | 15:37 |
@HeikoS | copy | 15:37 |
@HeikoS | so definitely no SWIG for this one | 15:37 |
@lambday | HeikoS: yeah so it will be messy | 15:37 |
@HeikoS | only for a wrapper that allocates new mem | 15:37 |
@lambday | yeah | 15:38 |
@HeikoS | leave protected | 15:38 |
@lambday | cool | 15:38 |
@lambday | HeikoS: BTW the point I was talking about reducing the number of memalloc : https://github.com/lambday/shogun/blob/221cdcac33b7b08899322f3060e692b248f768c0/src/shogun/features/DenseFeatures.cpp#L1061 | 15:38 |
@HeikoS | I see | 15:39 |
@HeikoS | I mean you can do that | 15:39 |
@HeikoS | but pls make it a sep method | 15:39 |
@HeikoS | sep. of concerns | 15:39 |
@HeikoS | and then call it explicitly in the loop | 15:39 |
@lambday | HeikoS nono this one does not require realloc.. that's the point | 15:39 |
@HeikoS | I see | 15:39 |
@lambday | even if I remove realloc, this method works | 15:39 |
@HeikoS | ok then | 15:39 |
@lambday | cause it is called on the "other" features | 15:40 |
@lambday | which re-uses the pre-allocated matrix | 15:40 |
@lambday | earlier, each of those "other" features was allocating one matrix for itself... | 15:40 |
@lambday | see what I mean/ | 15:40 |
@HeikoS | I get it | 15:42 |
@HeikoS | nice one | 15:43 |
@wiking | ok | 15:44 |
@wiking | people | 15:44 |
@lambday | wiking: we, the people? | 15:44 |
@lambday | ;) | 15:44 |
@wiking | yes | 15:45 |
@wiking | so i'm tired | 15:45 |
@wiking | and crashing | 15:45 |
@wiking | but could you just drop me | 15:45 |
@wiking | an email | 15:45 |
@wiking | about the stuff | 15:46 |
@wiking | ? | 15:46 |
@wiking | would appreciate | 15:46 |
@lambday | wiking: yeah.. sure | 15:46 |
@lambday | brb.. my laptop is fucked up.. | 15:47 |
@HeikoS | good night | 15:47 |
-!- lambday [6a3311c5@gateway/web/freenode/ip.106.51.17.197] has quit [] | 15:47 | |
-!- lambday [6a3311c5@gateway/web/freenode/ip.106.51.17.197] has joined #shogun | 15:53 | |
-!- mode/#shogun [+o lambday] by ChanServ | 15:53 | |
@HeikoS | lambday: you followed the team list? | 15:54 |
@lambday | HeikoS: yo.. there is already a method named copy_feature_matrix.. which copies the matrix arg to "this" one | 15:58 |
@lambday | HeikoS: yeah | 15:58 |
@lambday | HeikoS: umm what do you mean? | 15:58 |
@lambday | HeikoS: how about renaming the existing one to "use_feature_matrix" .... it is asking the class to "use" the matrix that it provides.. | 15:59 |
@lambday | HeikoS: then "copy_feature_matrix" could mean it copies the existing one | 16:00 |
@lambday | https://github.com/shogun-toolbox/shogun/blob/develop/src/shogun/features/DenseFeatures.h#L272 | 16:02 |
@lambday | why is it virtual | 16:03 |
@lambday | :/ | 16:03 |
@lambday | not being overridden anywhere | 16:03 |
@HeikoS | lambday: well .... | 16:05 |
@HeikoS | bad SE | 16:05 |
@HeikoS | we should fix things | 16:05 |
@HeikoS | but maybe in sep patches | 16:05 |
@HeikoS | so issue these things | 16:05 |
@lambday | HeikoS: tell me a good name for this method | 16:05 |
@lambday | can't use copy_feature_matrix anymore | 16:05 |
@HeikoS | I see | 16:05 |
@lambday | :( | 16:05 |
@HeikoS | use_feature_matrix for the other one seems better | 16:06 |
@HeikoS | or even: set_feature_matrix | 16:06 |
@lambday | yeah | 16:06 |
@lambday | yeah set | 16:06 |
@HeikoS | but that needs a separate PR | 16:06 |
@lambday | so natural | 16:06 |
@HeikoS | and then a rebase | 16:06 |
@HeikoS | in your one | 16:06 |
@lambday | err | 16:06 |
@HeikoS | I mean such API changes should be separated from your changes | 16:06 |
@lambday | HeikoS: well, actually it doesn't really "set"... it clones it first | 16:06 |
@HeikoS | see what I mean? | 16:06 |
@HeikoS | ah I see | 16:07 |
@lambday | HeikoS: so, set would convey something different | 16:07 |
@HeikoS | ah such a method doesnt even make sense | 16:07 |
@HeikoS | right? | 16:07 |
@HeikoS | why would one want such function | 16:07 |
@lambday | HeikoS: let me check whether it is being used anywhere | 16:07 |
@HeikoS | caller can copy itself and then set | 16:07 |
@HeikoS | sounds like I added this shit :D | 16:07 |
@HeikoS | but yeah good to check | 16:07 |
@lambday | Kmeans | 16:07 |
@HeikoS | and overloading doesnt do it? | 16:07 |
@lambday | that's the only one place | 16:07 |
@HeikoS | lambday: ok then what about a PR that fixes that | 16:08 |
@lambday | HeikoS: well, with overloading, you lose the default argument value for column_offset = 0 | 16:08 |
@HeikoS | and then you name it copy_* | 16:08 |
@HeikoS | okok | 16:08 |
@HeikoS | then I think it is OK to change | 16:08 |
@lambday | HeikoS: change it to "use_feature_matrix" ? | 16:08 |
@HeikoS | no | 16:08 |
@HeikoS | caller clones matrix and then calls set | 16:09 |
@lambday | makes sense | 16:09 |
@HeikoS | or whatever | 16:09 |
@HeikoS | just a setter | 16:09 |
@lambday | yeah | 16:09 |
@lambday | why clone I don't really understand :D | 16:09 |
@lambday | anyway | 16:09 |
@lambday | maybe it was required | 16:09 |
@lambday | ah multithreaded | 16:10 |
@HeikoS | lambday: I think it is good to check that | 16:10 |
@HeikoS | and then fix :D | 16:10 |
@lambday | yeah... so caller calls clone.. | 16:10 |
@lambday | makes zero sense to clone it within densefeature | 16:10 |
@lambday | yeah just checked | 16:11 |
@lambday | it's used within a openmp for | 16:11 |
@lambday | we shouldn't accept such patches in the first place | 16:11 |
@HeikoS | lambday: there you go :D | 16:11 |
@HeikoS | I agree | 16:11 |
@HeikoS | but things slip through | 16:11 |
@HeikoS | so just fixing | 16:11 |
@HeikoS | just read only saves the day there | 16:12 |
@HeikoS | but the const stuff can be fixed later | 16:12 |
@lambday | HeikoS: read only for which case? | 16:12 |
@HeikoS | to avoid cloning data | 16:12 |
@HeikoS | openmp | 16:12 |
@HeikoS | features rarely are modified | 16:12 |
@HeikoS | so read only views are needed | 16:13 |
@lambday | HeikoS: but the algorithm is updating the matrix over and over.... I wonder why | 16:13 |
@HeikoS | ah it is | 16:13 |
@HeikoS | ok then | 16:13 |
@HeikoS | well leave it | 16:13 |
@HeikoS | as it is | 16:13 |
@HeikoS | just refactor | 16:13 |
@lambday | I mean, do we really need it to be a dense feature | 16:13 |
@lambday | let me check | 16:13 |
@HeikoS | probably not | 16:14 |
@lambday | ah man this code | 16:14 |
@HeikoS | hahaha | 16:16 |
@lambday | only set would suffice.. no clone required | 16:16 |
@HeikoS | yes | 16:16 |
@lambday | as far as I can see | 16:16 |
@HeikoS | no threading issues? | 16:16 |
@lambday | let me change that, this method was added recently.. should be unit-tested properly | 16:17 |
@HeikoS | cool, | 16:17 |
@lambday | HeikoS: nah.. within the openmp multithreaded zone, the features object is not even used.. only the matrix | 16:17 |
@HeikoS | new pr | 16:17 |
@HeikoS | kk | 16:17 |
@lambday | HeikoS: yeah new pr.. | 16:18 |
@lambday | :( | 16:18 |
@HeikoS | cool | 16:19 |
@HeikoS | I like PRs where code gets better | 16:19 |
@HeikoS | rather than more bloated | 16:19 |
@lambday | yeah.. when I was looking at our dense feature code.. I was feeling like changing most of it.. to make it more readable | 16:25 |
@lambday | there was this book which says somewhere "always leave the code in a better state than it was whenever you touch it".. guess it applies nicely if we want things to be manageable .. | 16:26 |
@lambday | have to be strict with our merging policies.. all of the current changes in my pr is happening because you didn't let me merge in the first place :D | 16:27 |
-!- mikeling [uid89706@gateway/web/irccloud.com/x-clbrxoaqgcvnoeqc] has quit [Quit: Connection closed for inactivity] | 16:30 | |
@HeikoS | wiking: I invited the lab | 17:19 |
@HeikoS | but cannot see anything where ppl log on | 17:19 |
@HeikoS | how many etc | 17:19 |
@sukey | Issue #3683 "Memcheck KMeans" opened by lambday - https://github.com/shogun-toolbox/shogun/issues/3683 | 17:23 |
@sukey | Issue #3683 "Memcheck KMeans" lambday added label: "bugfixing" - https://github.com/shogun-toolbox/shogun/issues/3683 | 17:23 |
@sukey | Issue #3683 "Memcheck KMeans" lambday added label: "entrance" - https://github.com/shogun-toolbox/shogun/issues/3683 | 17:23 |
@sukey | Issue #3683 "Memcheck KMeans" lambday added label: "Cleanups" - https://github.com/shogun-toolbox/shogun/issues/3683 | 17:23 |
-!- tj_ [67e3630c@gateway/web/freenode/ip.103.227.99.12] has joined #shogun | 17:28 | |
@HeikoS | lambday: memcheck? | 17:28 |
@HeikoS | errors? :D | 17:28 |
@sukey | Issue #3683 "Memcheck KMeans"- https://github.com/shogun-toolbox/shogun/issues/3683 | 17:29 |
@lambday | HeikoS: first I thought I introduced it.. then checked it with old code.. still got the same error.. | 17:29 |
@HeikoS | cool | 17:30 |
@HeikoS | sigh | 17:30 |
-!- tj_ [67e3630c@gateway/web/freenode/ip.103.227.99.12] has quit [Ping timeout: 260 seconds] | 17:35 | |
@sukey | Pull Request #3684 "Removed copy_feature_matrix method from DenseFeatures" opened by lambday - https://github.com/shogun-toolbox/shogun/pull/3684 | 17:38 |
@sukey | Pull Request #3684 "Removed copy_feature_matrix method from DenseFeatures" - https://github.com/shogun-toolbox/shogun/pull/3684 | 17:43 |
-!- lambday [6a3311c5@gateway/web/freenode/ip.106.51.17.197] has quit [Ping timeout: 260 seconds] | 18:14 | |
-!- tj_ [67e3630c@gateway/web/freenode/ip.103.227.99.12] has joined #shogun | 18:16 | |
-!- deshalb [58e9944a@gateway/web/freenode/ip.88.233.148.74] has joined #shogun | 18:45 | |
-!- deshalb [58e9944a@gateway/web/freenode/ip.88.233.148.74] has left #shogun [] | 18:47 | |
@sukey | Pull Request #3564 "Temp" closed by karlnapf - https://github.com/shogun-toolbox/shogun/pull/3564 | 19:01 |
-!- OXPHOS [92bd305b@gateway/web/freenode/ip.146.189.48.91] has joined #shogun | 19:55 | |
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has quit [Ping timeout: 240 seconds] | 20:26 | |
tj_ | Hi, I am trying to resolve issue https://github.com/shogun-toolbox/shogun/issues/3663. As far as I have understood, InputParser uses the ParseBuffer(which is a ring buffer) to hold the data and does the parsing(with the help of Streaming*File classes) in a separate thread when started. Now the Streaming*Features fetches example objects one by one through a InputParser object. | 20:36 |
tj_ | So there is only one parsing thread(producer) and only one streaming*feature thread(consumer) at all times? Not multiple producer/multiple consumer? | 20:37 |
-!- tj_ [67e3630c@gateway/web/freenode/ip.103.227.99.12] has quit [Quit: Page closed] | 20:54 | |
--- Log closed Sat Mar 11 00:00:59 2017 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!