[12:36:33] https://github.com/wiki-ai/wikiclass/pull/16 [14:32:23] o/ Amir1 [14:32:30] didn't get as far as I hoped last night. [14:32:36] These old SVC models are really a burden. [14:32:39] o/ [14:32:51] I swear they didn't take so long to train in the past. [14:33:02] :D [14:33:04] I've had a model training for 8 hours and it's still not done. [14:33:14] halfak: https://github.com/wiki-ai/wikiclass/pull/16 [14:33:21] can you check this? [14:33:23] Amir1, already merged [14:33:29] I'm dropping mw itself [14:33:40] oh, amazing. I didn't get email [14:33:55] thanks [14:34:21] WRT SVC. It's working for itself, so I wouldn't worry too much :) [14:34:34] do you want to check i18n PRs? [14:34:59] I need to work on wheels since wikiclass is merged [14:37:19] "It's working for itself,"? [14:39:08] You told that it's training [14:39:25] so it doesn't need human interaction [14:39:35] or, I got it wrong :D [14:40:20] Oh! Yeah. That's fair, but I've been running tests -- trying to get it trained in a way that reflects what we have deployed. [14:40:37] I need to iterate, so it needs to work faster. [14:41:36] hmm [14:41:38] 9.5 hours this has been training :\ [14:42:03] it's unusally too long [14:43:13] strange [14:43:52] keep me posted [14:44:04] and tell me if I can do anything, tests, etc. [14:44:46] halfak: is it for a particular wiki or model? or you couldn't get even one model done [14:45:00] Waiting on the enwiki damaging model [14:45:02] Just one [14:45:30] I think something might be wrong with this particular implementation of SVM. [14:45:32] what about giving a try for another model, maybe this one is corrupted somehow [14:45:48] The only thing that could be corrupt is the dataset. [14:45:56] We start the model from scratch each time. [14:46:29] * halfak restarts the process [14:47:19] hmm, I tried SVC tuning reports of revscoring on some data I had (not related to revscoring at all) and it was pretty fast [14:47:32] several weeks ago [14:48:39] once we are done with i18n PRs I make a patch and discard form_builder route, as you said [14:50:01] Amir1, interesting. I actually removed the SVC lines from our tuner because they took too long. [14:50:27] I de-commented them in the yaml file [14:50:51] What machine were you running on? [14:50:58] I know it takes longer than, let's say, RF but it was considerably fast [14:55:34] Just started up a different SVC build [14:58:32] I'm going to be using a lot of resources on ores-compute-01 for a little while. [14:59:47] totally ok, nothing to do with it right now [15:38:02] wiki-ai/revscoring#573 (model_utility_tests - 059211f : halfak): The build passed. https://travis-ci.org/wiki-ai/revscoring/builds/115087093 [15:39:06] wiki-ai/revscoring#576 (model_utility_tests - a9464bb : halfak): The build passed. https://travis-ci.org/wiki-ai/revscoring/builds/115087546 [15:39:40] Amir1, thinking about merging the i18n for forms. [15:39:50] But if we do that, then the form builder is broken. [15:39:56] So what do you think we should do? [15:39:57] \o/ [15:40:06] for builder is okay [15:40:16] the biggest problem is the CDN thing [15:40:24] ? [15:40:24] but right after the merge of i18n [15:40:35] I don't know what you are talking about CDN thing. [15:40:37] I make a PR discarding form_builder [15:40:47] The form builder does not reflect the new structure of form configurations. [15:40:54] yeah it does [15:40:56] We still need the form builder. [15:41:06] But not FormBuilder.js [15:41:07] I made some changes in the form_builder [15:41:10] We never actually used that. [15:41:14] Oh OK. [15:41:16] * halfak checks that [15:41:56] I don't see a change for the form builder. [15:42:25] let me find it [15:42:48] which PR you are looking halfak ? [15:42:54] https://github.com/wiki-ai/wikilabels/pull/96 [15:43:52] * halfak just uploaded revscoring 1.0.2 to PyPI [15:44:00] oh, halfak. This one doesn't affect form_builder at all. [15:44:06] https://github.com/wiki-ai/wikilabels/pull/97 [15:44:20] this one changes form_builder [15:44:42] https://github.com/wiki-ai/wikilabels/pull/97/files#diff-09ad181be69d94256ccddafe411b7bd7R18 [15:46:02] in PR 96, we don't need to touch form_builder because the server.py is refactored in a way that external behavior won't change [15:46:50] I tested form_builder on PR 96 and it was the same with i18n included [15:47:51] Amir1, the form builder doesn't work with your new yaml files. [15:48:09] Because it expects i18n in the yaml [15:48:55] I tested it [15:49:00] maybe I did something wrong [15:49:09] let me re-test [15:51:27] halfak: if you check form_builder.html in templates, the yaml config is there, it doesn't load them from yaml files [15:52:14] Amir1, indeed. But we are now unable to *use* the form_builder to produce the yaml files we need for building forms. So the form_builder is incompatible with the new scheme. [15:53:32] so basically we should delete the i18n part from the html? [15:54:12] we should write a manual explaining that people should make the json files for i18n [15:54:31] what do you think halfak? [15:55:05] Amir1, yeah. That'd be a good stop-gap. Eventually, I think we'll want to update the form_builder so that it can handle this i18n stuff. [15:55:30] ok, I update the htm file [15:55:35] then I need to go [15:55:53] Looks like some JS changes will be in order [15:56:16] the other PR? [15:56:47] I tried to sort them alphabetically anywhere I see them [15:57:00] halfak: ^ [15:57:02] I need to go [15:57:10] be back in two or three hours [15:57:15] kk. o/ [15:57:44] JS changes will need to be made for the form_builder to work without i18n in the yaml. [15:57:52] But we can look at that later. [15:58:00] * when you're back [15:58:03] Amir1, ^ [18:24:50] back [18:25:02] o/ halfak [18:26:26] o/ Amir1 [18:26:36] Just about to run away to get lunch and take care of dog. [18:27:02] ok, Is there anything I can do? [18:27:37] swagger? [18:27:50] hmmm [18:27:53] ok. [18:27:55] Not a blocker, but something we need in prod [18:27:59] Oh! The i18n stuff [18:28:08] ANd setting up the form_builder [18:28:21] I fixed [18:28:40] halfak: I pushed my changes before I went [18:35:26] OK. Will have to check that when I get back [18:37:27] sure [19:20:56] o/ Amir1 [19:21:01] just got back. Looking at PR [19:54:38] Form builder is still broken! [19:54:46] I'll just fix it. [20:39:43] i18n is now merged.