#musicbrainz-devel

so we should perform this one step on line53, and then generate highlevel data for this model (#88)

and we will then have two pieces of information: the ground truth in the dataset editor; and the high-level estimates from running our model

I have code which will give an overall accuracy of the model based on these two bits of information

if you got it to this stage, it'd be a full dataset/evaluation cycle

alastairp: do we need to generate history file during every evaluation?

and where should we store it?

yeah, I'm not sure at what stage we should generate it

if we wait, then the evaluation / parameter selection will happen faster

since there is one less step

so we could not do it during every evaluation; only when someone is happy with their accuracy

you could just store it in a data directory

data/models/uuid/date/file.history

ok

uuid of evaluation job?

ah, is there uuid for dataset and for evaluation?

(I was thinking of dataset)

but if there is one for evaluation, use that. you don't need a date then (since you can get that data from the database)

uhm. am I doing something wrong? release editor doesn't go through. any change and clicking "enter edit" and it spins for a while and then goes akc ot normal but no submitting is done? tired both beta and regular

alastairp: do you know how long this step is?

I think having that running separately makes things unnecessarily more complicated

we probably need another queue for generating history files

evaluation script will need significant changes if we are going to reuse it for this task

It's basically the same length as a parameter permutation

So, with 720 permutations, we'd run 721 cyclea

Cycles

alastairp: hm, wtf https://pastee.org/ctub5

Ah! I got that a few days ago

I thought it was a Mac iasue

I'll ask around to see if we can find a solution

that's strange

I don't think I even changed anything. it's the same VM, same dataset

hmm

did you update gaia?

because I made a change recently to the python bindings

I don't think so

I'll try to set it up again

I'll try a few demos - we should just be able to load up the python interface in a shell to see where the problem is

I'll revert my change and see if that was the problem

ahhhh

Gentlecat: double-check that your class names are strings, not uuid objects

in fact, strings. not anything else

you should be able to reproduce it in a python shell with p = gaia2.Point(); p[9] = "foo"

whereas p["string"] = "foo" will give a more normal error

let me check

I think it's a unicode string

this raises the same exception: p[unicode("hello")] = "test"

ah!

there we go

1) open ticket in gaia

2) convert to str

alastairp: wait, it's different function

these examples reference Point_setValue function

I just changed your create_groundtruth method

line 91

str(cls["name"])

I mean Point_setLabel

maybe it's better to use .encode("UTF-8")?

yeah, that should work just as well

do we have any restrictions on label names?

perhaps we should actually limit them to [a-zA-Z0-9]+

no, not really

yeah, that's one way to do it, but seems hacky

I don't mind if you str() or .encode

oh - I bet it worked for you before you set the default encoding of the database connection, right? :)

right!

that's what it was :)

OK, cool. I was trying to fix this issue before I gave you some more feedback on the website interface

now that I've succesfully done a model, I'll do this for you too

feedback \o/

how should a query to the web service be written with something like 7" vinyl ? for example /ws/2/release?fmt=json&query=format:(7" vinyl)&offset=0 does not work because of the quotation mark

can you urlencode it? " is %22

Gentlecat: did you open a ticket in gaia for non-string values?

not yet, it's on the list

ok, cool

alastairp: tried it but the same error comes

{"error":"Search server could not complete query: Bad request"}

alastairp: btw, any plans to merge https://github.com/alastair/python-musicbrainzn... and release a new version? :)

ruchiranga: https://github.com/scholarslab/SolrSearch/issue... "The " needs to be escaped for Solr as \" -- this is different from html encoding"

Gentlecat: so many plans

we'll need that for CB https://bitbucket.org/metabrainz/critiquebrainz...

yeah, since it was rebased I've been meaning to merge it

but I wanted to hold off on a release until we added some of the missing parts from the last 3 (*3*!) schema changes :(

alastairp: oh thanks :)

alastairp: https://github.com/MTG/gaia/issues/24 looks good?

yeah, that’s fine. I also told dmitry, so he knows what the problem is

it's safe to assume that gaia doesn't work with python 3 at all, right?

I guess my efforts to make server compatible were kind of pointless

would've been neat to have that though

well, we could run the server in python3, and the workers in python2 :)

Project musicbrainz-server_master build #174: SUCCESS in 18 min: http://ci.musicbrainz.org/job/musicbrainz-serve...

* Michael Wiencek: Ensure en is always bundled, so jed-data is always existent

* Michael Wiencek: MBS-8514: relationship credits can't be edited

New post: blog: Server update, 2015-08-10 <http://blog.musicbrainz.org/2015/08/11/server-u...;

Project musicbrainz-server_master build #175: SUCCESS in 19 min: http://ci.musicbrainz.org/job/musicbrainz-serve...

* Michael Wiencek: Make the gulpfile PWD-independent

Project musicbrainz-server_beta build #618: SUCCESS in 19 min: http://ci.musicbrainz.org/job/musicbrainz-serve...

* Michael Wiencek: Don't return "result" of _warn_about_invalid_ids

* Michael Wiencek: Ensure en is always bundled, so jed-data is always existent

* Michael Wiencek: MBS-8514: relationship credits can't be edited

* Michael Wiencek: Make the gulpfile PWD-independent