[16:51:07] schana: have you thought about how to ship the bravado dependency, given that it is not packaged for debian? [16:51:27] how much work would it be to remove the dependency? [17:56:31] o/ AbbeyRipstra [17:56:36] can I get an invite for the showcase [17:56:41] To do the IRC Q&A? [18:01:52] o/ ori [18:02:03] hello [18:02:07] Just noticed you asked about getting rid of a python dependency because it's not already packaged [18:02:09] ori: that dependency is currently responsible for validating and parsing input to the api as well as marshaling the responses [18:02:14] lol [18:02:19] schana too :) [18:02:26] So, I think you can pickle that just fun. [18:02:30] *wheel [18:02:36] Sorry not pickle. WHEEEL [18:02:47] This is how we ship python dependencies for ORES. [18:02:53] Security review and ops verified. [18:03:48] the problem is that bravado in turn has its own unpackaged dependencies [18:04:07] ori, that's cool. We create a wheel-set of all dependencies :) [18:04:07] let me try and get a handle on how many packages we'd have to ship [18:04:23] where do you put them? [18:04:28] in a deploy branch / repo? [18:04:30] In a repo [18:04:31] * halfak looks [18:04:47] gerrit is so browse-able :) [18:04:49] this is very useful, by the way -- I appreciate the ping and the advice. [18:05:18] I'm happy to help. I think this wheel business is going to save some time and energy [18:05:19] +1 [18:06:12] do you use official wheels from pypi or build your own? [18:06:14] https://gerrit.wikimedia.org/r/#/admin/projects/research/ores/wheels [18:06:19] we build our own [18:06:27] On a VM with the same config as prod [18:07:20] schana: are you up for seeing how ORES does it and replicating that setup? [18:07:40] sounds good [18:07:58] sweet, thanks [18:08:13] (but probably not tonight) I was only on for the SoS [18:08:23] yeah no worries [18:08:33] schana, feel free to ping me when you want to talk more. :) [18:08:52] halfak: wasn't there a wiki page that you (or Amir) set up with the info? [18:09:02] schana, I don't think we did for the wheeling [18:09:14] https://phabricator.wikimedia.org/diffusion/1880/browse/master/frozen-requirements.txt [18:09:22] https://phabricator.wikimedia.org/diffusion/1880/browse/master/Makefile [18:10:12] thanks halfak, I'll follow up with questions (if any) tomorrow [18:10:15] 'night all [18:10:20] cool [18:10:22] g'night [18:20:23] Hey folks! Research showcase is starting in 10 minutes [18:21:23] I'll be doing questions [18:21:29] helloooo people [18:21:30] Or IRC questions, as it is [18:21:32] hello [18:21:38] the Research showcase will start in 9 min. ;) [18:21:42] :P [18:21:47] Hey halfak did you catch the memo about the new link to join the Hangout? [18:21:48] Here is the youtube link, in case you missed it in the title: https://www.youtube.com/watch?v=fTDkVeqjw80 [18:21:51] Yes [18:21:56] Okay, thank you. [18:22:03] halfak: will you present? [18:22:17] brendan_campbell, I'm guessing that I'm not in the right hangout though. [18:22:25] I just joined the one that was sent to WMFall [18:22:39] leila, no presentation from me. [18:22:42] Just IRC Q&A [18:23:09] halfak: in that case, you should jump in the Hangout, and I'll go out [18:23:10] ;) [18:23:20] Got the right hangout now [18:29:30] is the live stream happening yet? there are two listed (one in email and one here) but neither one is displaying anything yet [18:29:54] Stream should start in a few seconds. [18:29:57] debt, ^ [18:30:17] ok - thanks! I thought it started 15 min ago :) it's working now [18:31:21] debt, ahh. Yeah, we usually use the first 15 minutes to chat with the presenters off the air. [18:31:32] Ping me with your questions! [18:34:33] AbbeyRipstra, please click the button to hide the avatars on the bottom of the hangout [18:35:10] just did it [18:35:14] Thanks :) [18:35:16] :) [18:37:50] I'm a little bit confused about why "news" statements are being detected. [18:38:14] Did someone else catch the *why* of this? [18:39:20] * halfak posts his own questions [18:39:21] Halfak: What is a "temporal discourse" and what differentiates that from other types of statements? [18:39:27] Got it. [18:39:31] Thanks halfak ;) [18:40:14] I like how you resolved this. [18:40:23] to ask the question that is. [18:41:31] Oh! I think I get it. [18:41:58] So we're looking for "news statements" so because we have a corpus of news articles and we'd like to use them to add citations. [18:42:37] More questions! Halfak: Why detect news statements first? Why not just look for relevant news citations for all assertions in Wikipedia? [18:45:35] Halfak: Is the "tree kernel" like a PCFG? What entailment measures were most effective? [18:45:56] I'm going to be the only one asking questions at this rate ;) [18:49:57] Halfak: How is it that you have fewer sections than entities? [18:51:16] Halfak: What's the difference between web and news in the age of peer produced journalism? [18:56:59] Halfak: What makes a citation better than another? [19:05:19] I have that question too Aaron. [19:13:15] o/ GCiampaglia [19:16:19] o/ Besnik [19:16:25] I found https://en.wikipedia.org/wiki/Tree_kernel [19:16:36] I was asking about https://en.wikipedia.org/wiki/Stochastic_context-free_grammar [19:17:07] Looks like the Tree kernel looks at "subset trees" [19:17:27] PCFG looks at common productions which are like a single node-view subset tree [19:18:58] Hi Halfak! [19:19:08] On youtube we lost the focus on the presentation [19:19:11] YT stream has changed focus [19:19:23] AbbeyRipstra, ^ [19:19:52] I am still seeing the presentation.. here [19:19:56] strange [19:20:05] AbbeyRipstra, click on presentation [19:20:19] did that [19:20:22] is it back in focus? [19:20:31] nope [19:20:49] Still no. [19:20:55] Click on Besnik and then back on presentation [19:21:34] It did not work [19:21:34] did it work? [19:21:38] did not [19:21:43] OH! Good now [19:21:48] works! [19:21:50] phew... [19:21:54] YES! [19:21:58] not sure what happend [19:22:07] thanks everyone for letting me know :) [19:23:35] sorry just saw your points Aaaron [19:23:47] this is the paper i was referring to in the Tree Kernels section [19:23:47] http://www.cs.utexas.edu/~ml/papers/dependency-emnlp-08.pdf [19:25:13] Besnik, it's really interesting that PCFGs are not discussed at all. [19:25:33] PCFGs are super old tech [19:25:55] would have to dig on the details regarding their difference w.r.t this paper [19:26:10] in principle CFG are used for constructing the dependency parse trees in NLP [19:26:13] does anyone have questions for Amy? [19:26:27] I do, but I'll ask myself. [19:26:31] k [19:26:57] Looking for questions. [19:27:09] If you're still drafting a question, please ping me to reserve a spot [19:28:15] I have a question for Amy: do you think this system could work for summarizing things other than discussions? For example reviews? [19:28:33] Like Amazon reviews... [19:28:44] On topic of trust, has there been thoughts on how do to address negative summarization? [19:28:51] Boo J-Mo [19:28:56] Taking my job! [19:29:21] just to speed things up :) [19:29:53] * halfak takes jerb back [19:29:53] Thanks J-Mo! [19:30:19] Volker_E, just asked. Ping right away if oyu want a follow-up [19:30:36] halfak: thanks (listening) :) [19:30:36] we are at time - is it OK to go over by a few mins? [19:30:45] We block the room for an extra 15 [19:31:01] thanks! [19:32:58] hangouts crashed! [19:34:21] lol lost my tab again [19:35:41] * halfak claps into his muted mic [19:35:51] great presentations [19:35:54] thanks! [19:36:46] thanks! [19:41:39] * halfak runs to chiropractor.