[00:10:01] RECOVERY - check load on Ores-Compute is OK: OK - load average: 2.07, 2.99, 4.94 [12:18:50] CUSTOM - check disk on deployment-ores is CRITICAL: CHECK_NRPE: Error - Could not complete SSL handshake. Test [12:19:05] PROBLEM - puppet on deployment-ores is CRITICAL: CHECK_NRPE: Error - Could not complete SSL handshake. [12:19:34] PROBLEM - puppet on deployment-ores is CRITICAL: CHECK_NRPE: Error - Could not complete SSL handshake. [12:19:41] PROBLEM - check disk on deployment-ores is CRITICAL: CHECK_NRPE: Error - Could not complete SSL handshake. [12:19:41] PROBLEM - check load on deployment-ores is CRITICAL: CHECK_NRPE: Error - Could not complete SSL handshake. [12:20:03] PROBLEM - puppet on deployment-ores is CRITICAL: CHECK_NRPE: Error - Could not complete SSL handshake. [12:20:32] PROBLEM - puppet on deployment-ores is CRITICAL: CHECK_NRPE: Error - Could not complete SSL handshake. [12:23:38] CUSTOM - check disk on ores-redis.01 is CRITICAL: NRPE: Command 'check_disk' not defined Test [12:23:58] ACKNOWLEDGEMENT - mem-check on ores-redis.01 is UNKNOWN: NRPE: Unable to read output Ack [12:24:25] PROBLEM - check disk on ores-redis.01 is CRITICAL: NRPE: Command 'check_disk' not defined [12:27:32] PROBLEM - check disk on ores-05 is CRITICAL: NRPE: Command 'check_disk' not defined [12:27:33] PROBLEM - check disk on Ores-Compute is CRITICAL: NRPE: Command 'check_disk' not defined [12:37:55] CUSTOM - Host Ores-worker-05 is UP: PING OK - Packet loss = 0%, RTA = 0.92 ms Test [12:38:07] CUSTOM - Host ores-webhost is UP: PING OK - Packet loss = 0%, RTA = 0.90 ms Test [12:54:32] Those "Could not complete SSL handshake." can be ignored. Needs to be fixed on the server side. [12:58:54] ores-redis-01 has 2.70 GiB left on / [12:59:00] on /srv it has 10.62 GiB left [14:12:52] good morning everyone [14:13:20] halfak: meetings all day like usual:P [14:27:54] halfak: is the frwikisource pr on wikiclass ready to go? [15:00:35] Amir1: are you around ^ [16:37:43] its quiet today [16:40:10] Meetings meetings meetings [16:40:20] Zppix, yeah that PR should be ready to go [16:40:25] okay will merge [16:40:36] I'm just about to be AFK for ~45 minutes while I bike to the University. [16:40:43] o/ [16:40:45] ack [16:57:36] Amir1: I've set up wikiclass and wikilabels with more accurate irc notifications thru gituhb [17:43:45] halfak: instead of random github bot nicks i updated all wiki-ai repos with accurate nicks [17:44:49] Oh! Awesome :) [17:45:09] * halfak finds something to merge [17:45:15] theres plenty :P [17:45:33] How's https://github.com/wiki-ai/wikilabels/pull/185 going? [17:45:52] WIP still its a actual error that needs fixed [17:45:55] its blocked on that [17:46:26] Hmm... anyway we can remove the channel name from the nick? [17:46:36] "#wikimedia-ai-" makes it really long [17:46:47] what channel name from the nick? [17:47:20] thats because its a notice not a actualy text message, it does that to prevent excess noise [17:47:35] thats an IRC thing not a config [17:47:38] Oh I see. [17:47:53] It's what IRC shows when a message is sent without joining. [17:48:07] yep [17:49:00] im also in the process of making ores ci out of precise cause its EOL'd [17:49:20] thats going just as well as wikilabels :P [17:49:37] yeah. I'm getting pretty tired of precise too [17:49:42] Does travis have a new option for us? [17:50:01] We have for linux jessie and trusty jessie is what i want as its latest [17:51:00] Let's use debian jessie. [17:51:05] that's what we use in prod [17:51:08] but when i try ores with jessie it doesnt like the module [17:51:16] halfak: correction prod uses jessie and trusty [17:51:18] "the module" :P [17:51:25] Prod ORES uses Jessie [17:51:31] oh i see [17:51:36] i thought oyu meant ci [17:51:41] you* [17:52:31] can you decipher the build details on my ores pr and se e why its failing? [17:53:58] ImportError: The 'enchant' C library was not found. Please install it via your OS package manager, or use a pre-built binary wheel from PyPI. [17:54:11] apt-get install you an "enchant" [17:55:16] ? [18:13:25] halfak: squash is a wonderful thing :P [18:13:49] Woops. Should have squashed that [18:16:01] oh well it was a good irc bot test :P [18:22:44] :) [18:23:40] all bots for notifications use ssl and an ssl only port halfak fyi [18:56:43] halfak: Earlier, I have told you my intention to add Multi-Layer Perceptron (http://scikit-learn.org/stable/modules/neural_networks_supervised.html) to revscoring. Now, I wonder what you think about it [19:05:46] And, if you're good with it, it would be cool if you could advise whether doing that plus all of what I havent done so far (e.g. training, tuning, testing models) could be finished by the end of this month. [19:27:05] CUSTOM - check disk_space on ores-redis.01 is OK: DISK OK 2gb left on / [19:27:11] o/ [19:27:48] glorian_wd, not sure. Never built a MLP before so I'm not sure how training and parameterizing goes. [19:28:10] Do you know how many parameter combinations you'd want to manage when tuning? [19:29:29] CUSTOM - check disk_space on ores-webhost is OK: DISK OK 8gb left on / [19:29:34] halfak: tuning for MLP? I thought the Scikit learn doc describes the parameters for MLP [19:29:51] glorian_wd, right. Don't make me read docs for you [19:29:54] :P [19:29:58] Give me the cliff notes [19:30:11] Furthermore, I think revscoring already have auto-tune function [19:30:34] So, I guess it able to automatically optimize the parameter tuning? [19:30:44] that's only my guess based on what I know about revscoring so far [19:30:55] You need to set parameter ranges. [19:32:26] halfak: from the doc, "MLP requires tuning a number of hyperparameters such as the number of hidden neurons, layers, and iterations." [19:32:49] I haven't tried MLP before, but I am curious if I can use it in revscoring [19:33:15] Nonetheless, if I don't have time to do it. I can use the models that already existed in revscoring [19:33:23] You technically can! But you should know whether or not you want to ;) [19:33:26] don't have enough time to doit* [19:34:06] halfak: I want to do it :D [19:34:30] if you think it makes sense [19:37:53] If it ends up being a higher fitness model than some of our others, I think it could make sense -- assuming the memory usage isn't crazy. [19:43:37] halfak: given that we haven't reached the experimentation stage, what do you think if I am trying to set it up now? [19:43:52] I think item quality model would be a good test bed for this MLP [19:44:31] if the memory usage is crazy or other problems occur, we can ditch it [19:46:31] I'm OK with giving it a try [19:48:59] im back i miss anything? [19:51:19] Nothing too exciting [19:51:25] I'm working on meta-reviews [19:51:30] and political BS [19:51:33] :) [19:51:43] build the software halfak [19:52:17] I wish I could. But I also with that the social systems around me work. So thus the tension [19:52:39] Or did you mean that we should build a software-halfak [19:52:42] I was joking... [19:52:50] And software-halfak can do reviews and politics ;) [19:52:50] i was ripping off trump's wall [19:53:09] halfak: then, how I can set it up on revscoring? :P [19:53:24] how can I set it up* [19:53:24] anyone good at travis? [19:57:58] still working on https://github.com/wiki-ai/revscoring/pull/298 halfak ? [19:58:19] Zppix, I'm waiting on others to get back to me. [19:58:40] halfak: since jan? [20:00:40] Zppix, yeah. Regretfully [20:00:47] madhuvishy, ^ [20:00:48] :P [20:01:01] (re. tamil badwords) [20:01:47] halfak: let me know when its good to go and ill handle the merge to master when it is [20:02:16] Might be a wait given recent patterns. I had someone from the recent hackathon working on it, but it seems that he's flaked out too. [20:02:46] maybe a notices to tamilwiki for help would be a good idea? [20:04:22] Sure! I'd love it if we could find a local who could work with us and socialize the idea of ORES to tamil wiki. [20:04:36] halfak: then, how I can set it up on revscoring? :P *I think you missed my chat above* [20:04:45] want me to ask in wikitech-l? halfak [20:04:56] Sure Zppix [20:05:03] There's a task for it in the backlog [20:05:16] ok [20:05:27] glorian_wd, revscoring.scorer_models.ScorerModel [20:06:19] which task is it? [20:06:40] https://phabricator.wikimedia.org/T166052 [20:08:02] halfak: do I only need to add a python file (e.g. mlp.py) under https://github.com/wiki-ai/revscoring/tree/master/revscoring/scorer_models? [20:08:17] yeah [20:09:38] halfak: Ok I will try that ;) [20:09:47] in the meantime, I hope you can get to my PR soon :P [20:09:49] thanks :D [20:12:49] halfak: sent [20:14:15] sorry halfak! I've been poking Chandru about it, and will poke again! [20:15:45] halfak: https://lists.wikimedia.org/pipermail/wikitech-l/2017-June/088266.html [20:16:16] Thanks madhuvishy [20:16:23] And thanks Zppix [20:17:30] halfak: hey you have to much to do in one day i like to help wherever i can :D [20:17:44] It's great :D [20:20:25] revscoring needs readme converted to .md but im lazy lol [20:23:02] halfak: let me try trusty for ores ci and see if its still eroring out [20:25:31] * paladox is prepairing for bbc news site to be covered in elections results :) [20:26:26] same error... [20:26:34] Zppix, trustty --> trusty? [20:26:48] good catch [20:31:09] yay [20:31:29] it worked thanks halfak [20:31:44] nice [20:32:01] I'm guessing trusty has enchant installed by default [20:32:37] or enchant doesnt like jessie? [20:33:23] enchant just needs to be installed [20:34:10] trusty and jessie arent to much different afaik [20:43:55] i can setup builds that happen on a timed basis to addition to the normal builds, so we have tests ran to make sure the code isnt broken even after the build when it was first commited [20:44:00] if you want [20:53:04] halfak: [21:26:33] Zppix, fyi if you want to make sure that work gets reported in our newsletter, it's critical that it ends up in the "done" column on our main board. [21:26:47] Re. trusty, did you try installing enchant via travis.yml? [21:27:23] it didnt need it i jsut ran the build [21:27:39] We should be using the same OS as prod though :/ [21:27:41] Hmmm [21:31:05] halfak: i cant get jessie to work and there wont be much of a difference [21:31:31] Hmm... does look like enchant *is* installed. [21:32:03] https://github.com/travis-ci/apt-package-whitelist [21:32:08] No apt whitelist for jessie [21:32:23] Are you sure we eve *should* be able to use jessie? [21:33:27] according to the docs yes [21:34:13] Hmm... Let's stick with what you have. [21:34:24] ok [21:34:33] Also no resolving until things are merged :P [21:34:43] i cant move task to the done column [21:35:10] Why not? [21:35:26] phab doesnt like my device much [21:35:29] when it comes to workboards [21:35:32] Gotcha. [21:36:33] {{done}} [21:36:33] You rule, halfak! [21:36:37] Thanks AsimovBot [21:37:50] ci on wikilabels is still blocked til code fixes are fixed [21:37:54] code issues* [21:40:05] Not a code issue [21:44:28] * halfak goes to confirm [21:47:02] Ok it's not a code change but I guess a PR changed the type of those file from symlink to plain ol text [21:53:00] Zppix, can you rebase and re-push your wikilabels PR? [21:53:13] Extra credit for cleaning up the commits too ;)