[10:43:56] <wikibugs>	 06Revision-Scoring-As-A-Service, 10Wikilabels: assets url return empty string - https://phabricator.wikimedia.org/T154865#2926687 (10Ladsgroup)
[11:09:59] <wikibugs>	 06Revision-Scoring-As-A-Service, 10Wikilabels: assets url return empty string - https://phabricator.wikimedia.org/T154865#2926704 (10Ladsgroup) I've filed https://github.com/miracle2k/flask-assets/issues/121 so might get a clue
[11:28:31] <wikibugs>	 06Revision-Scoring-As-A-Service, 10Wikilabels: assets url return empty string - https://phabricator.wikimedia.org/T154865#2926707 (10Ladsgroup) https://github.com/wiki-ai/wikilabels/pull/148
[11:28:47] <wikibugs>	 06Revision-Scoring-As-A-Service, 10Wikilabels, 15User-Ladsgroup: assets url return empty string - https://phabricator.wikimedia.org/T154865#2926710 (10Ladsgroup) a:03Ladsgroup
[17:03:11] <wikibugs>	 06Revision-Scoring-As-A-Service, 10ORES, 07Documentation: Draw rough data flow diagrams for ORES - https://phabricator.wikimedia.org/T154441#2926842 (10awight) a:05Halfak>03awight
[17:04:09] <awight>	 What are the primary metrics we use for evaluating model health using test set results?  ROC AUC?
[17:14:35] <awight>	 Also, I've added a step I'm unsure about--what comes between raw wiki labels results and the test and training sets?  There must be some averaging of responses?
[17:14:37] <Amir1>	 awight: yes, it's ROC AUC you can see them in tuning reports in ediaquality repo
[17:16:20] <awight>	 Amir1: hi, thanks!
[17:19:25] <awight>	 I feel a bit silly about doing these diagrams as a learning excercise, but then emitting mostly my imagination of what should be happening rather than getting into the implementation details of what's actually happening...
[17:19:43] <awight>	 Hopefully I can bridge that gap
[17:22:14] <Amir1>	 Awesome, thank you
[17:27:08] <awight>	 I'm pretty happy with the DFDs now, https://github.com/adamwight/ores-diagrams
[17:27:48] <awight>	 Probably good to move on to refining the use cases.  Those got weird too, though, cos I'm putting all the low-level actions in there rather than high-level business goal stuff
[17:40:55] <awight>	 o/
[20:38:17] <wikibugs>	 06Revision-Scoring-As-A-Service, 10Research Ideas, 10Wikimedia-Developer-Summit (2017): What should an AI do you for you? Building an AI Wishlist. - https://phabricator.wikimedia.org/T147710#2927013 (10Basvb) Is something already known about the stream? Would love to participate from a distance.