10:24 AM
alastairp
right. I'm just explaining what I thought you were talking about
2016-06-23 17549, 2016
10:24 AM
Gentlecat
it's just a way to get more data about models that we are using
2016-06-23 17553, 2016
10:24 AM
alastairp
this is why we weren't understanding each other
2016-06-23 17531, 2016
10:25 AM
alastairp
but now I understand that you're talking about the point once we decide a model is good and promote it "live" on the website
2016-06-23 17539, 2016
10:25 AM
Gentlecat
yeah, I meant to use cross-dataset validation just for challenges
2016-06-23 17552, 2016
10:25 AM
Gentlecat
otherwise it gets a bit too complicated for what I'm doing
2016-06-23 17502, 2016
10:26 AM
Gentlecat
correct
2016-06-23 17508, 2016
10:26 AM
alastairp
OK. so let's try again
2016-06-23 17523, 2016
10:26 AM
alastairp
you want to record feedback on the /mbid page (probably)
2016-06-23 17527, 2016
10:26 AM
alastairp
and for that you need...
2016-06-23 17546, 2016
10:26 AM
alastairp
"datasets" representing the metadata in our existing models
2016-06-23 17503, 2016
10:27 AM
Gentlecat
to know what values this model can output
2016-06-23 17516, 2016
10:27 AM
alastairp
the names of the classes, you mean?
2016-06-23 17519, 2016
10:27 AM
Gentlecat
so that user who submits feedback can choose which one might be right
2016-06-23 17521, 2016
10:27 AM
Gentlecat
yes
2016-06-23 17529, 2016
10:27 AM
alastairp
2016-06-23 17518, 2016
10:28 AM
alastairp
though, I think that's just the metadata, not the class each one is in
2016-06-23 17541, 2016
10:28 AM
Gentlecat
I don't even need recordings
2016-06-23 17543, 2016
10:28 AM
alastairp
you know you can get a class list from the probablilities in any ll file?
2016-06-23 17547, 2016
10:28 AM
Gentlecat
just the structure
2016-06-23 17504, 2016
10:30 AM
Gentlecat
hmm, that might be another way to do it then
2016-06-23 17540, 2016
10:30 AM
Gentlecat
2016-06-23 17553, 2016
10:30 AM
Gentlecat
values in `all`?
2016-06-23 17551, 2016
10:31 AM
Gentlecat
wondering how that output will look like with models that we generate
2016-06-23 17517, 2016
10:32 AM
alastairp
yes, all
2016-06-23 17531, 2016
10:32 AM
alastairp
all will always contain... all of the class values
2016-06-23 17556, 2016
10:32 AM
Gentlecat
if keys there can consist of just a limited set of characters
2016-06-23 17529, 2016
10:33 AM
alastairp
it's whatever is in the groundtruth file we send to gaia
2016-06-23 17542, 2016
10:33 AM
Gentlecat
don't remember how we are handling class names
2016-06-23 17554, 2016
10:33 AM
alastairp
I don't know if there are limitations
2016-06-23 17504, 2016
10:34 AM
alastairp
there is an open ticket about giving better labels to class names
2016-06-23 17508, 2016
10:34 AM
alastairp
I wonder if we need a mapping
2016-06-23 17529, 2016
10:34 AM
alastairp
but that mapping would probably be related to a dataset (or snapshot), so I guess we're back to needing a dataset to base it off
2016-06-23 17501, 2016
10:35 AM
alastairp
e.g., dataset_calss could have a "readable name" and a "internal name", which is a slug of readable name
2016-06-23 17509, 2016
10:35 AM
Gentlecat
I that it's a good idea to have them anyway
2016-06-23 17515, 2016
10:35 AM
Gentlecat
right
2016-06-23 17521, 2016
10:35 AM
Leo_Verto has quit
2016-06-23 17540, 2016
10:35 AM
alastairp
the existing datasets have no readable name
2016-06-23 17547, 2016
10:35 AM
Gentlecat
also assuming that this is the "workflow" that we require to get a model running
2016-06-23 17550, 2016
10:35 AM
alastairp
if you go this direction, you should create one
2016-06-23 17505, 2016
10:36 AM
Gentlecat
dataset -> snapshot -> model
2016-06-23 17526, 2016
10:36 AM
alastairp
yeah, it seems reasonable
2016-06-23 17531, 2016
10:36 AM
Gentlecat
and currently used models just have the last part
2016-06-23 17539, 2016
10:36 AM
Gentlecat
oh, speaking of models
2016-06-23 17547, 2016
10:36 AM
alastairp
this is close to what I wanted to do with my 2 train/test snapshots
2016-06-23 17505, 2016
10:37 AM
Gentlecat
can we actually "enable" them currently in hl calculation script?
2016-06-23 17514, 2016
10:37 AM
alastairp
right - eventually it'd be good to have dataset for each of our existingmodels. We can do this with daniele's project
2016-06-23 17515, 2016
10:37 AM
Gentlecat
or are they hard-coded?
2016-06-23 17526, 2016
10:37 AM
alastairp
there's a status field on the model table
2016-06-23 17512, 2016
10:38 AM
alastairp
however, I don't think the hl calc script knows what to do if we have 3 million files with models 1-14 created, and we add model 15
2016-06-23 17526, 2016
10:38 AM
alastairp
going forward, it'll calculate 15 models for new submissions
2016-06-23 17535, 2016
10:38 AM
Gentlecat
probably just starts calculating them for new ones, yes
2016-06-23 17539, 2016
10:38 AM
alastairp
but we need to go back and fill in data for all the existing ones
2016-06-23 17541, 2016
10:38 AM
Gentlecat
I think that is fine
2016-06-23 17511, 2016
10:39 AM
Gentlecat
maybe we can have some other task runner that makes sure that backlog is covered by new models too
2016-06-23 17512, 2016
10:39 AM
alastairp
the tables are split up such that we add a new highlevel_model row
2016-06-23 17522, 2016
10:39 AM
alastairp
that's actually not a bad idea
2016-06-23 17533, 2016
10:39 AM
alastairp
instead of complicating 1 script
2016-06-23 17553, 2016
10:42 AM
alastairp
2016-06-23 17529, 2016
10:43 AM
alastairp
and I'd like to come to a decision on deleting snapshots, too
2016-06-23 17544, 2016
10:43 AM
alastairp
my feeling is that if a snapshot is not being used anywhere, we can remove it
2016-06-23 17552, 2016
10:44 AM
alastairp
so once an eval job has been submitted to a challenge, it probably can't be removed, but any time before then, if the eval job is deleted, the shapshot should be too
2016-06-23 17516, 2016
10:46 AM
Gentlecat
sounds fine to me
2016-06-23 17527, 2016
10:46 AM
alastairp
cool
2016-06-23 17539, 2016
10:46 AM
alastairp
do you want me to add that to 188, or will you?
2016-06-23 17506, 2016
10:47 AM
Gentlecat
create a pull request instead
2016-06-23 17521, 2016
10:47 AM
Gentlecat
it's easier to work like that
2016-06-23 17530, 2016
10:47 AM
alastairp
but we already have a pull request
2016-06-23 17535, 2016
10:47 AM
Gentlecat
pull request into my branch
2016-06-23 17553, 2016
10:47 AM
Gentlecat
just so if I need to update something I can still do it
2016-06-23 17503, 2016
10:48 AM
Gentlecat
because with 182 I can't
2016-06-23 17515, 2016
10:48 AM
ruaok
alastairp: did you and chirlu come to a conclusion about index types for the listen table?
2016-06-23 17532, 2016
10:48 AM
Gentlecat
and it's harder to see what changes are mine, and what are yours
2016-06-23 17545, 2016
10:48 AM
Gentlecat
that's why I was suggesting not to work on the same branch
2016-06-23 17512, 2016
10:49 AM
Gentlecat
I also tend to rebase and amend often
2016-06-23 17510, 2016
10:50 AM
alastairp
ruaok: brin are useful if the the data is strictly (or mostly strictly) in order
2016-06-23 17555, 2016
10:50 AM
alastairp
with our bulk imports, I don't think we're going to get as much out of them unless we reorder the tables from time to time
2016-06-23 17500, 2016
10:51 AM
ruaok
and use CLUSTER periodically?
2016-06-23 17521, 2016
10:51 AM
ruaok
ok, then I really need to work on the JSON separation.
2016-06-23 17530, 2016
10:51 AM
ruaok
any chance you could look at the PR I opened yesterday.
2016-06-23 17535, 2016
10:51 AM
alastairp
Gentlecat: I'm sorry, I really don't agree with this. It's adding much more complexity for something which I don't think is needed
2016-06-23 17540, 2016
10:51 AM
ruaok
I'd rather not create more branches at this point
2016-06-23 17548, 2016
10:51 AM
alastairp
if you're rebasing and amending, mark your PR as WIP and I won't touch it
2016-06-23 17513, 2016
10:52 AM
alastairp
but if you've submitted a PR ready for review and merge, others should be able to contribute to the branch
2016-06-23 17533, 2016
10:52 AM
Gentlecat
well, I need to make changes during a review
2016-06-23 17547, 2016
10:52 AM
Gentlecat
I'll just work from a fork like most people do anyway
2016-06-23 17509, 2016
10:53 AM
Gentlecat
that's what we are doing with musicbrainz
2016-06-23 17506, 2016
10:54 AM
alastairp
the thing with this workflow is I feel it makes the PR submitter a bottleneck
2016-06-23 17508, 2016
10:54 AM
Gentlecat
(it would be fine if we just coordinated that, but you didn't ask)
2016-06-23 17520, 2016
10:54 AM
alastairp
for fuck's sake
2016-06-23 17522, 2016
10:54 AM
alastairp
we have git
2016-06-23 17528, 2016
10:54 AM
alastairp
it knows how to fucking merge commits
2016-06-23 17549, 2016
10:56 AM
ruaok
woo. Solar Impulse is in Spain. :)
2016-06-23 17536, 2016
11:01 AM
mihaitish joined the channel
2016-06-23 17543, 2016
11:04 AM
Leo_Verto joined the channel
2016-06-23 17502, 2016
11:12 AM
D4RK-PH0ENiX has quit
2016-06-23 17538, 2016
11:20 AM
armalcolite
ruaok: is the API working fine?
2016-06-23 17514, 2016
11:21 AM
ruaok
on the docker-without-kafka branch?
2016-06-23 17518, 2016
11:21 AM
ruaok
submission works for sure.
2016-06-23 17537, 2016
11:21 AM
armalcolite
yeah,
2016-06-23 17500, 2016
11:22 AM
armalcolite
2016-06-23 17526, 2016
11:22 AM
armalcolite
i merged and created a branch on my fork, as i cant create a branch on parent repo
2016-06-23 17550, 2016
11:22 AM
ruaok
ah, I hadn't seen it yet.
2016-06-23 17554, 2016
11:22 AM
ruaok
let me try in a minute.
2016-06-23 17528, 2016
11:25 AM
ruaok is soo stoked to have docker
2016-06-23 17532, 2016
11:25 AM
ruaok
and no java based BS.
2016-06-23 17543, 2016
11:28 AM
D4RK-PH0ENiX joined the channel
2016-06-23 17532, 2016
11:37 AM
ruaok
2016-06-23 17535, 2016
11:37 AM
ruaok
armalcolite: ^^
2016-06-23 17548, 2016
11:37 AM
ruaok
it keeps repeating that same submission over and over.
2016-06-23 17501, 2016
11:38 AM
ruaok
and most of the data in the submission is missing.
2016-06-23 17503, 2016
11:40 AM
ZarkBit has quit
2016-06-23 17529, 2016
11:48 AM
ZarkBit joined the channel
2016-06-23 17519, 2016
12:01 PM
Freso|koume has quit
2016-06-23 17516, 2016
13:00 PM
ruaok
2016-06-23 17519, 2016
13:00 PM
ruaok
meh.
2016-06-23 17523, 2016
13:00 PM
ruaok
not that link. :)
2016-06-23 17549, 2016
13:00 PM
ruaok
tweetdeck fail.
2016-06-23 17550, 2016
13:00 PM
ruaok
2016-06-23 17551, 2016
13:00 PM
Gentlecat
bad customers are bad? :D
2016-06-23 17527, 2016
13:01 PM
ruaok
can you please look at that Gentlecat?
2016-06-23 17537, 2016
13:01 PM
Gentlecat
there's not much to look at
2016-06-23 17547, 2016
13:01 PM
Freso|koume joined the channel
2016-06-23 17550, 2016
13:01 PM
Gentlecat
we need a redirect on login page (at least)
2016-06-23 17502, 2016
13:02 PM
Gentlecat
a fix is to go to https page
2016-06-23 17520, 2016
13:04 PM
ruaok
ok, can you please make this happen?
2016-06-23 17522, 2016
13:04 PM
ruaok
2016-06-23 17535, 2016
13:04 PM
ruaok
ok, I need to stop using tweetdeck. fail. fail fail.
2016-06-23 17546, 2016
13:04 PM
ruaok
2016-06-23 17529, 2016
13:05 PM
Gentlecat
there's not much I can do on the server side apart from adding a flag for forcing HTTPS on login page, which is ugly
2016-06-23 17534, 2016
13:05 PM
Gentlecat
it's a setup issue
2016-06-23 17554, 2016
13:05 PM
ruaok
setup issue?
2016-06-23 17500, 2016
13:06 PM
Gentlecat
I don't remember what we decided, but I'm pretty sure at least that page was supposed to be redirected
2016-06-23 17513, 2016
13:06 PM
Gentlecat
don't know if nginx config was updated wherever it is
2016-06-23 17513, 2016
13:06 PM
reosarevok
ruaok: they do but I'm running around and it's a national holiday so please tell him yes yourself :)
2016-06-23 17525, 2016
13:06 PM
ruaok
reosarevok: ok.
2016-06-23 17534, 2016
13:06 PM
Gentlecat
setup/deployment
2016-06-23 17506, 2016
13:07 PM
Gentlecat
do requests to AB go through the gateway?
2016-06-23 17522, 2016
13:07 PM
ruaok
AB is not at DWNI.
2016-06-23 17527, 2016
13:07 PM
Gentlecat
ok
2016-06-23 17504, 2016
13:11 PM
Gentlecat
2016-06-23 17529, 2016
13:11 PM
ruaok
as long as there are no args to the page, it might work.
2016-06-23 17558, 2016
13:11 PM
Gentlecat
2016-06-23 17504, 2016
13:12 PM
Gentlecat
"full original request URI (with arguments)"
2016-06-23 17550, 2016
13:13 PM
ruaok
I *love* being page slapped with documentation pages. really, I do.
2016-06-23 17522, 2016
13:14 PM
Gentlecat
is that sarcasm?
2016-06-23 17536, 2016
13:14 PM
ruaok
yes.
2016-06-23 17516, 2016
13:15 PM
Gentlecat
ok
2016-06-23 17528, 2016
13:15 PM
Gentlecat makes not to not ever correct you
2016-06-23 17511, 2016
13:16 PM
ruaok
I dont mind being corrected.
