[10:42:27] hmm who runs the GitHub bots? [10:42:44] I like the reports but I'd rather have in channel reports [12:28:06] Helder, sorry to miss your questions yesterday. [12:28:22] 1. How can you review my 4? I'll drop you some links in a minute. [12:28:39] 2. Is getting "no tasks available" expected behavior? Yes. [12:28:55] Regretfully, we don't have that message translated. [12:30:10] And here's the links to the last 4 diffs. https://etherpad.wikimedia.org/p/ptwiki_last_tasks [12:30:21] * halfak gets back to morning routine. [12:32:51] ^ This is cool, but I wish the username was shorter [12:33:09] o/ [12:33:19] Hey Helder [12:33:34] I'm just about to go AFK again for an hour to bike to the university [12:35:14] afk? [12:35:44] So, looks like those tasks were indeed the last ones I labeled yesterday [12:35:55] This means ptwiki is done :-) [13:55:55] Helder, can you take a look at those diffs and tell me how you would label them? [13:56:49] (when you have a chance) [14:15:34] I actually labeled them throug Wikilabels already [14:15:50] Oh! That shouldn't be possible :S [14:16:06] * halfak will take a look in the DB [14:21:53] halfak, I updated https://etherpad.wikimedia.org/p/ptwiki_last_tasks with the labels I submitted [14:22:27] Thanks [14:23:41] So, in the end we had how many revisions in the dataset? 20000? [14:23:49] That's right [14:23:55] and a few of they were labelled twice? [14:24:03] Yup. I'll check on that right now. [14:26:11] Yeah... it shouldn't be possible that we both labeled it. [14:26:37] It should only be true that you can have at most one auto label and one human label [14:27:20] Seems a few tasks have 3 labels! [14:27:23] * halfak digs more [14:27:39] At least the problem seems to be rare. [14:28:15] and at least we have more labels, instead of less, than necessary [14:28:26] Ha. That's a good point. [14:29:24] Ahh! I think I understand why some of these triple labels exist. [14:29:44] But somehow, the timestamps on them seem to be wrong. [14:29:55] A user is asking me about next steps of the project. Do we have an estimate on when these labels will be used for producing the "damaging" and "goodfaith" models? [14:30:11] https://pt.wikipedia.org/wiki/Usu%C3%A1rio%28a%29_Discuss%C3%A3o:He7d3r#Rotulagem_da_qualidade_das_edi.C3.A7.C3.B5es [14:30:15] Helder, was just considering extracting the data and building models now. [14:30:27] great [14:30:35] It wasn't on my agenda for today, but there's not much standing between us getting that done. [14:31:42] (1) Write a script that hits http://labels.wmflabs.org/campaigns/ptwiki/7/?tasks and converts the labeled tasks into \t pairs [14:32:10] (2) Pull dataset into ores-wikimedia-config and add lines to makefile to do the feature extraction and model building [14:32:26] (3) Run the whole thing on local dev or ores-compute. [14:33:02] (4) Write report on model and if it's good, submit pull request to ores-wikimedia-config that includes the new model [14:33:09] (5) Merge and deploy [14:33:59] Helder, does ptwiki use flagged revisions? [14:34:16] just for a few pages [14:35:31] Gotcha. Same as enwiki then. [14:36:01] \o/ awight merged the language fixes. That's going to make life much easier for our devs. :) [14:37:06] https://pt.wikipedia.org/wiki/Special:StablePages?uselang=en [14:37:56] Woah. Enwiki has *way more* [14:37:59] halfak, how many revs had more than 1 label? [14:38:10] (just in case that infor is readly accessible) [14:38:56] Overall, 1326 [14:39:04] * halfak gets number for ptwiki [14:42:59] does these include both manual and automatic labels? [14:44:36] Yes [14:45:01] OK. SO ptwiki's campaign has 336 tasks (rev_id) with more than one label [14:45:21] thanks! [14:50:48] halfak, are these tickets enough for tracking these problems of duplicated labelling? [14:50:48] https://github.com/wiki-ai/wikilabels/issues/68 [14:50:48] https://phabricator.wikimedia.org/T106861 [14:50:49] or there is more specific things to be reported too? [14:50:49] Helder, yeah. I'll be digging into this today, so I'll fill more specific bits as I get to them. [14:50:49] Thanks :) [14:50:49] ok [14:50:49] Interested in picking up the ptwiki data today and building some models? [14:50:49] I can't [14:50:49] I figured you'd have other stuff. [14:50:49] I'm struggling with debian today [14:50:49] But it was worth a try [14:50:49] Godspeed :D [14:50:49] and still can't access labs [14:50:49] BTW: what is the ssh command again you use for ores? [14:51:03] Depends now. YuviPanda just set us up a new project. [14:51:13] hmm [14:51:13] But for building mores, you probably want ores-compute [14:51:25] So, "ssh ores-compute.eqiad.wmflabs" [15:25:47] halfak Morning. I'll be right there. Just getting my coffee started. [15:25:55] o/ aetilley [15:26:02] Hokay. [15:26:10] * halfak hacks on wikilabels stuff in the meantime [15:30:43] halfak Would you like to call or would irc be better for doing simultaneous dev/setup tasks? [15:31:02] Let's try IRC and do a call if necessary. [15:31:08] Can you remind me where we left off? [15:31:20] I think you had vagrant up working OK, right? [15:32:25] I had just successfully been able to vagrant ssh on both of my machines. But we were going to talk about what exactly this should be used for. [15:33:33] Gotcha. [15:33:58] So. I'd like you to be able to run some commands in ores that will be necessary for constructing models. [15:34:11] * halfak turns his vagrant on. [15:34:27] I need to figure out where the models live inside of vagrant quick [15:34:34] I think they'll be in /srv/ores/config [15:35:26] Hmm... nope [15:35:33] yeah that doens't exist [15:37:10] OK. Run this in your 'vangrant ssh': [15:37:19] sudo systemctl restart celery-dev uwsgi-dev [15:37:44] Do you have a browser on your arch machine? [15:37:51] done [15:38:01] no, sorry [15:38:06] That's OK. [15:38:10] We can work with wget probably [15:38:24] What kind of file? [15:38:50] ? [15:39:28] You wanted to use wget for something. [15:39:31] On arch, not vagrant, run "wget -qO- localhost:8080" [15:39:42] one sec [15:39:44] This will hit the ORES server (hopefully) [15:40:56] done [15:41:13] Did it say "Hi! I'm ORES. :D"? [15:41:26] No. [15:41:45] Maybe you set it to use a different port than 8080? [15:41:49] In your vagrant file? [15:42:13] Ah yes, I think I changed it as the first one. One sec. [15:42:59] I have guest:8080, host: 8081 [15:43:10] try "wget -qO- localhost:8081" [15:43:51] running [15:46:29] Done [15:46:58] No welcome message [15:47:20] Hmm... Try the 8080 command on the machine itself [15:47:44] on vagrant? I was calling it on my machine [15:47:51] sorry.. on vagrant yeah [15:48:00] same wget? [15:48:17] Yup. but it should be port 8080 on the vagrant [15:48:57] Done. No welcome message. [15:49:25] Did this run on vagrant without issue? "sudo systemctl restart celery-dev uwsgi-dev" [15:49:47] I don't remember the exact results. May I run it again? [15:49:53] Yup [15:50:21] Ah yes, now I remember the exact results, which are Nothing. [15:50:40] OK. That's fine [15:51:11] Can you past the output of "netstat -tulpn"? [15:51:44] You might need to sudo that [15:52:08] vagrant@debian:~$ netstat -tulpn [15:52:08] (Not all processes could be identified, non-owned process info [15:52:08] will not be shown, you would have to be root to see it all.) [15:52:08] Active Internet connections (only servers) [15:52:11] Proto Recv-Q Send-Q Local Address Foreign Address State PID/Prog [15:52:14] ram name [15:52:16] tcp 0 0 0.0.0.0:111 0.0.0.0:* LISTEN - [15:52:19] [15:52:25] tcp 0 0 0.0.0.0:22 0.0.0.0:* LISTEN - [15:52:28] [15:52:31] tcp 0 0 127.0.0.1:25 0.0.0.0:* LISTEN - [15:52:31] [15:52:34] tcp 0 0 0.0.0.0:41725 0.0.0.0:* LISTEN - [15:52:38] [15:52:38] aetilley, much better if you paste into a github gist [15:52:41] tcp 0 0 127.0.0.1:6379 0.0.0.0:* LISTEN - [15:52:44] [15:52:46] tcp6 0 0 :::111 :::* LISTEN - [15:52:49] [15:52:53] tcp6 0 0 :::22 :::* LISTEN - [15:52:54] * halfak waits for flood to pass [15:52:56] [15:52:59] tcp6 0 0 ::1:25 :::* LISTEN - [15:53:02] [15:53:05] tcp6 0 0 :::51626 :::* LISTEN - [15:53:10] [15:53:10] udp 0 0 0.0.0.0:977 0.0.0.0:* - [15:53:13] [15:53:16] udp 0 0 127.0.0.1:987 0.0.0.0:* - [15:53:21] [15:53:23] udp 0 0 0.0.0.0:58871 0.0.0.0:* - [15:53:26] [15:53:31] udp 0 0 0.0.0.0:43058 0.0.0.0:* - [15:53:34] [15:53:34] udp 0 0 0.0.0.0:68 0.0.0.0:* - [15:53:37] [15:53:38] e.g. here's mine https://gist.github.com/halfak/baa289032e86bda363c0 [15:53:41] udp 0 0 0.0.0.0:111 0.0.0.0:* - [15:53:44] [15:53:46] udp6 0 0 :::40121 :::* - [15:53:49] [15:53:53] udp6 0 0 :::977 :::* - [15:53:56] [15:53:59] udp6 0 0 :::10797 :::* - [15:54:02] [15:54:06] udp6 0 0 :::111 :::* - [15:54:09] [15:54:12] vagrant@debian:~$ [15:54:16] erg [15:54:19] is that readable on your end? [15:54:26] srry [15:54:32] Um, I'm afraid your going to have to give me a minute to get a gist. [15:55:03] From this I can tell that nothing is running on port 80 though :/ [15:55:35] Ok. [15:56:46] *port 8080 [15:56:53] So yeah. That's weird, but we can move on. [15:58:48] We don't *need* the webserver for this work. [15:59:19] So, next thing I want you to try is -- in vagrant -- run 'source /srv/ores/venv/bin/activate' [15:59:21] ok [15:59:51] This will modify your vagrant terminal to use the version of python (and all it's installed dependencies) that ores uses. [16:00:19] Done [16:00:37] Your prompt should look like "(venv)vagrant@debian:" [16:01:02] Indeed. [16:01:14] OK. Cool. [16:01:27] Now if you run 'revscoring -h' you should get help for our general utilities. [16:02:58] Yes? [16:03:02] aetilley, ^ [16:04:19] Not quite. So disclaimer I will figure out a quick way of writing to Gists on this machine this afternoon. In the meantime I will just email you this error message. [16:04:33] Oh! Sure. [16:05:30] Just wondering, why work on the headless machine? Why not on the macOS? [16:05:33] * awight quietly points to https://phabricator.wikimedia.org/paste/create/ [16:05:41] ah hehe [16:05:55] awight, +1 Know how you can do that from the commandline? [16:06:18] Also, good morning :) [16:06:20] no idea! [16:06:32] Please do tell. [16:06:34] mornoon! [16:06:49] awight^ [16:07:27] fwiw, https://github.com/defunkt/gist [16:07:49] email seems like a more pragmatic solution, though ;) [16:09:09] halfak: I managed to get the Travis builds to turn green... but no luck getting the pretty dots next to commits, yet. [16:09:41] halfak: It might be worth poking at your travis-ci.org profile, perhaps there's a toggle in there [16:09:48] awight, there was a pretty dot next to the language one before you merged it. [16:10:01] awight, I don't grok travis really. I had Yuvi doing that work. [16:11:56] hrm, that dot was not so pretty :p it's still yellow, and links to quantifiedcode [16:12:23] Oh yeah. That quantified code stuff I find useful, but there needs to be a way to flag issues for ignoring. [16:12:51] well, I wouldn't bother you about the travis thing but it's linked to your personal account. I could just relink it to my own if you don't care, then I can toggle away and see what happens. [16:13:53] awight, +1 [16:13:58] Anything I need to do to enable that? [16:13:58] I did manage to turn travis green, at some cold hour of the night-- https://github.com/wiki-ai/revscoring/pull/146 [16:14:13] halfak: naw, you accidentally already gave me permissions [16:14:31] The owner group in wiki-ai? [16:14:52] o/ aetilley still there? [16:15:48] yep. okay I can take it from here [16:16:15] Thanks :) [16:17:33] * halfak resumes hacking on wikilabels [16:17:38] aetilley, ping me when you get back [16:18:04] o/ YuviPanda [16:18:54] Oh I actually saw that this morning but I figured ti [16:19:03] oh good! YuviPanda: I'm beating my head against travis-ci but it's making a rather hollow sound. You wanna jump in? [16:19:07] halfak sorryfor the delay [16:19:13] just sent you an email. [16:19:47] Ahh! Darn that error. [16:19:59] Is this a python 2/3 thing? [16:20:10] Nope. [16:20:12] One sec [16:20:22] We should add to the docs, though: python 3.4 requierd [16:20:34] we should... spell it just like that :( [16:20:41] hehe [16:21:06] aetilley, run 'sudo /srv/ores/venv/bin/pip install -r /vagrant/requirements.txt' [16:21:20] This will just update the installed packages so they reflect what ores/revscoring needs. [16:21:35] It should pull down a new version of requests and call it good. [16:24:01] It's asking me for a root pw. [16:24:39] oh wait [16:24:41] sorry [16:24:43] one sec [16:25:27] Running [16:25:30] Done [16:25:46] Looks like most requirements are already satisfied. [16:26:53] Try running 'revscoring -h' again. [16:28:21] Well that's strange. [16:28:56] Now it can't find revscoring [16:29:11] -bash: revscoring: command not found [16:30:05] What's your prompt look like? [16:30:22] vagrant@debian:/$ [16:30:35] gotta run that source command again. [16:30:43] It looks like we lost our old term session. [16:30:51] source /srv/ores/venv/bin/activate [16:31:10] Oh right. I'm sorry I quit that window and re ssh'ed w/o re sourcing. [16:31:11] OK. Time for me to run to the next meeting. I'll be around if you want to keep working on this. [16:31:21] Ok. Thanks Arron. [16:31:24] I'll just be flow to respond to questions. [16:31:28] *slow [16:31:43] Next thing we need to do is set up revscoring in a VM for you to work with. [16:31:48] That'll be an interesting new problem :) [16:32:04] In the ores vagrant you can only really *use* revscoring [16:32:35] developing it... well that'll be a different beast [16:32:36] Ok resourced and I think I'm getting the same error. [16:32:48] But go do your stuff. I will fiddle. [16:34:03] Woot! Looks like we have our traffic lights. [16:36:21] (venv)vagrant@debian [16:40:25] Yeah I'm getting exactly the same error. Traffic lights? [16:45:57] aetilley: Sorry, that was about continuous integration in Github, e.g. https://github.com/wiki-ai/revscoring/pull/147 [16:46:07] YuviPanda: looks like it's working now! c.f. ^ [17:03:31] * hare blinks [17:09:43] awight|afk: Thanks for the gist tip. [17:09:55] halfak for the record https://gist.github.com/e60c33611052124dcab7 [17:10:29] try 'sudo /srv/ores/venv/bin/pip install requests==2.5.3' [17:10:33] aetilley, ^ [17:12:22] Done [17:12:45] It involved uninstalling requests 2.7.0. [17:12:56] (I guess obviousl) [17:16:04] Same error when running revscoring -h thought. [17:16:06] though* [17:23:51] you should teach aetilley the trick you have shown me :/ [17:24:10] with virtualenv [17:24:16] actually I can teach him this. [17:27:41] brb [18:33:41] ToAruShiroiNeko, unsure about the virtualenv trick you are talking about [19:20:25] YuviPanda, got a couple minutes to chat vagrant and revscores? [19:26:46] halfak: leila has me [19:27:11] I see. And what does she want in trade? [19:27:29] I'll just dump some thoughts in here. Address 'em when you have a chance. [19:28:19] So, ORES vagrant helps us a lot with getting a dev environment set up. But the real problem that necessitates vagrant is revscoring's dependencies. [19:28:40] So what about when someone wants to work on revscoring [19:28:45] Should we also have a vagrant for revscoring? [19:33:07] I'm curious about vagrant adoption... it makes a lot of sense to have within an organization, for standardization and flexibility, but I'm not convinced that third parties are actually installing random packages in dedicated vagrant VMs [19:33:55] It could be that just recommending someone set up an ubuntu or debian VM is the way to go for revscoring development. [19:34:15] well... or just making it easy to install the dependencies [19:34:26] awight, yeah. that's a years long upstream problem [19:34:40] Also our dictionaries are not available on non-linux [19:35:08] A lot of WMF developers have tried mw-vagrant and mostly don't use it, which is saying a lot cos they have so much to gain [19:35:24] right, for non-linux development it might be the only option [19:35:30] Interesting. I thought that adoption of MW Vagrant was wide-spread [19:36:08] I'm interested to hear what Yuvi says, but my halfpenny is that a vagrant for ORES is plenty, as long as it installs revscoring in a way that's convenient for development. [19:36:11] awight: mw-vagrant is fairly widespread, IMO. It's also fairly complex [19:36:23] so putting ORES into mw-vagrant would've been terrible [19:36:35] halfak: I agree. I think we can find a way to make ores vagrant work for revscoring too? [19:37:05] oh, I definitely agree that a standalong vagrantfile is much better than shoehorning into mw-vagrant [19:37:06] YuviPanda, OK. I was tempted to have aetilley just clone revscoring into his vagrant instance earlier [19:37:51] Maybe we can tell vagrant to also mount "../revscoring" if it exists [19:38:02] DDuvall did a great survey of mw-vagrant adoption, https://www.mediawiki.org/wiki/MediaWiki-Vagrant/Surveys/2014Q4 [19:38:05] Assuming that your ores and revscoring clones are in the same dir of your host [19:38:10] halfak: it mounts everything in current dir to /vagrant [19:38:28] halfak: so we can probably just clone revscoring into /vagrant and munge python path [19:38:34] ewww [19:38:37] :P [19:38:43] halfak: err, or a git submodule / subtree [19:38:52] OK. Never done that before. [19:47:28] halfak you set up virtualenv on ubuntu with me to use python 3 [19:47:33] that [19:48:40] Oh! That's not such a trick as it is standard practice for python dev. [19:48:44] But I see what you mean. [19:48:52] Why not just set him up in the local machine? [19:49:01] halfak: revscoring I think, can be ok just happening with a virtualenv [19:49:05] I haven't worked with arch yet, but I suspect we're going to have some new and interesting issues. [19:49:23] halfak: the reason ORES benefits from vagrant is that it has more moving components - uwsgi, celery, redis [19:49:25] revscoring does not [19:49:26] YuviPanda, was hoping to get away from the 2-4 hour onboarding for getting the dependencies all installed. [19:49:32] halfak: right [19:49:32] YuviPanda, +1 [19:49:52] if we make it possible to do dev on revscoring on the ores vagrant as well [19:50:08] Yeah. Since we're already solving all of those problems for the ores vagrant... [19:50:12] yeah [19:50:25] and then people can test it easily too - just hit ORES [19:50:34] Yeah. My thought too. [19:53:23] YuviPanda: what about "python setup.py develop"? I think that sets the python path w/o additional munging [19:53:37] yeah that probably [19:53:40] I've never used it before tho [19:53:56] I haven't either. Interesting though. awight can you point to an example? [19:54:02] Or some blog post about it? [19:54:47] Ooooh. Just finished a wikilabels bit that will add two new types of views. [19:55:00] Once takes a wikitext and formats it in HTML (using an API) [19:55:13] https://pythonhosted.org/setuptools/setuptools.html#develop-deploy-the-project-source-in-development-mode [19:55:18] The other will take a revId and show you formatted HTML [19:55:22] * YuviPanda goes on the way to office [19:55:24] it leaves your source where it is and basically symlinks [19:55:25] The former is useful for snippets of wikitext [19:55:33] o/ YuviPanda [19:55:59] awight: thanks for working on travis! [19:56:24] awight, https://github.com/wiki-ai/revscoring/issues/148 [19:57:03] halfak: we should get madhuvishy to move wikilabels [19:57:15] YuviPanda, to a new project? [19:57:21] YuviPanda: for sure! I think it even works... [19:57:43] halfak: yes, and the central postgres db [19:57:45] and the puppet module [19:58:19] YuviPanda, kk. I just did the config fixes. I'm trying to get a jump on wikilabels progress today so it would be good timing. [19:58:37] halfak: yup [20:01:17] * halfak hopes that madhu-person doesn't mind all the pings, but has still seen them go by. [20:02:35] YuviPanda: halfak sure [20:02:42] ha ha no i don't mind at all [20:06:00] Cool! I have a PR for you to review. [20:06:03] * halfak digs [20:06:19] madhuvishy, https://github.com/wiki-ai/wikilabels/pull/73 [20:06:32] Feedback very welcome if you don't like what you see. [20:07:06] halfak: you should invite leila here [20:07:15] * halfak tried. Will try again [20:07:20] heh :D [20:07:26] ok leaving for realz now [20:08:44] I tried to invite them to the github team too. I can't find leila's github ID and no response from Ellery. [20:09:47] ToAruShiroiNeko: meta and Dutch is there, Italian will be ready soon, what else is needed? [20:10:02] result for meta is interesting [20:10:34] meta lool? [20:12:52] https://gist.github.com/Ladsgroup/cc22515f55ae3d868f47#file-metawiki [20:13:07] Lower scores here [20:13:43] Lots of names [20:13:51] "stephanie" [20:13:56] Friends of gays? [20:14:15] https://meta.wikimedia.org/wiki/Friends_of_gays_should_not_be_allowed_to_edit_articles [20:18:05] :))) [20:29:13] * halfak takes a chance at people not understanding the humor by tweeting that essay. [20:40:24] Nooo. Was just testing that. Does it work? [20:40:28] Amir1, ^ [20:40:48] we tested it long time ago [20:41:16] me and Robin talked about it [20:41:22] for a very long time [20:41:29] Oh! OK. Your last comment was that it didn't work. [20:41:38] btw, AaronSchulz made a good point when we were talking about ORES--he made me think that article importance rating should be an input feature. [20:41:41] yeah [20:42:03] * halfak sees it working in his dev environment. [20:42:06] Well cool : [20:42:08] :) [20:42:21] awight, seems reasonable. [20:42:30] Generalizing that as a feature might be an interesting problem though [20:42:37] \o/ [20:42:39] whyzzat? [20:42:46] Well, not all wikis have it. [20:42:51] ah right [20:42:59] Also, we'd need to define a lot of different template styles [20:43:09] I suppose we could query the category [20:43:10] He was also pointing out that it's a good variable for understanding editor prejudice... [20:43:30] I'm worried about features that have nothing to do with the change that was made though. [20:43:34] Oh rats, it's not in Wikidata yet? too bad [20:43:40] halfak: btw. I built a system to determine best threshold based on a cross validation set (or a test set) [20:43:44] Hmm... might be. [20:43:47] the result is really good [20:43:59] Cool! [20:44:02] Amir1, writeup? [20:44:14] based on f1 score [20:44:21] interesting. I was thinking that article category and importance will very likely be related to * perceived quality and * how people approach potentially reverting [20:44:53] (if you want better precision than recall=less false positives, you can use f0.5 score [20:46:40] Or, you're right that what I'm looking at is something totally different, more like editor modeling, which gets sketchy real fast. [20:50:59] awight, wikiprojects use categories & templates to track article importance and quality. [20:51:07]