[10:41:33] <dcausse>	 errand+lunch
[14:19:02] <inflatador>	 <o/
[14:26:08] <dcausse>	 o/
[14:46:51] <dcausse>	 errand, back in a ~2hours
[14:49:20] <ebernhardson>	 \o
[14:59:46] <inflatador>	 .o/
[15:21:06] <ebernhardson>	 i can't help but be amused, helloworld.java released by a parody band: https://www.youtube.com/watch?v=yup8gIXxWDU
[15:30:10] <inflatador>	 (｡◕‿◕｡)
[16:14:06] <Trey314159>	 That is really committing to the bit! Very entertaining... though I'm not sure my office mate appreciated it as much as I did.
[16:29:51] <ebernhardson>	 the chorus with the semicolons is just...it's a good bit
[16:35:20] <ebernhardson>	 unrelatedtly, setting up the ab test for updated mlr models.  pondering adding a bucket that gets the non-ml scoring
[16:35:34] <ebernhardson>	 mostly curiosity :)
[16:44:08] <dcausse>	 that eclipse loading screen... :)
[16:55:23] <inflatador>	 Workout, back in ~40
[17:37:23] * ebernhardson keeps getting distracted reading old interleaving and user testing code...just to make sure i remember what this does
[17:54:52] <ebernhardson>	 must be better options...i was poking around my old notebooks looking for something to analyze ab tests. I found something from 2018, it starts by reading raw json sequence files out of hdfs so maybe a little too old :)
[18:47:37] <dcausse>	 ebernhardson: still something odd on the mlr setup I think, jawiki was not enabled and the last PS enables only mlr-1024rs-next so the A/B test might not work as expected
[18:48:01] <ebernhardson>	 hmm, looking
[18:48:02] <dcausse>	 should we just ignore it?
[18:48:51] <ebernhardson>	 dcausse: the control will use the default config, so it's whatever is already configured (classic or wsum_inclinks), then the tests will set override that with the new one, i think it should be fine to define models that don't get used. They simply define the profiles iiuc
[18:50:10] <ebernhardson>	 CirrusSearchMLRModel isn't a real cirrus config, part of wmf-config expands that into rescore profiles with names matching the MLRModel key
[18:50:43] <dcausse>	 ah got it, so only for jawiki will test against non-mlr setup
[18:51:49] <ebernhardson>	 dcausse: zh and ko as well, all three have RescoreProfile set to a non-mlr model
[18:52:07] <ebernhardson>	 curiously, ja and ko are on classic and zhwiki is wsum_inclinks, cant imagine why we didn't want inclinks on ja and ko
[18:53:21] <dcausse>	 oh... I should look closer into this, I just trusted the wmgCirrusSearchMLRModel vars to be inline with the rest
[18:53:49] <dcausse>	 I think we don't practice this enough :(
[18:54:06] <ebernhardson>	 indeed, it's been years. I'm sure i have a newer notebook, but the most recent i found yet is 2018
[18:54:58] <dcausse>	 yes... esp the 3 todos: TODO: switch back to 'mlr-1024rs'
[18:55:19] <dcausse>	 :(
[18:55:32] <ebernhardson>	 i've never been the best at remembering to finish things
[18:55:43] <ebernhardson>	 good at working on whats obvious and infront of me :)
[18:56:46] <dcausse>	 I'm culprit of this too...
[19:39:30] <hare>	 I'm terrific at writing down the things I need to finish and then not doing them
[20:28:58] <Inflatador_>	 ryankemper since we are pushing back the deploy, i'll wait till 
[20:29:22] <Inflatador_>	 I get back from the doc. 
[20:34:25] <ryankemper>	 inflatador: yup sounds good