hmm. I would say that the process that #88 fixes works, it’s just slow
2015-08-11 22337, 2015
alastairp
so let’s not worry about that for now
2015-08-11 22301, 2015
alastairp
for extra copy of the dataset, I think that was so that someone could continue editing even if it was evaluating
2015-08-11 22326, 2015
alastairp
I think it’s better if we just lock the dataset if it’s in a queue or being evaluated. Can you edit the document and implement that?
2015-08-11 22350, 2015
Gentlecat
sure
2015-08-11 22323, 2015
Gentlecat
but then I don't understand how it would work with extra copies
2015-08-11 22332, 2015
alastairp
no, neither do I
2015-08-11 22337, 2015
Gentlecat
separate is going to be created?
2015-08-11 22342, 2015
alastairp
I guess you’d make a temporary copy, do the evaluation on that
2015-08-11 22348, 2015
alastairp
and then delete it when the model is made
2015-08-11 22355, 2015
Gentlecat
that seems very inconvenient
2015-08-11 22345, 2015
alastairp
so, let’s not do it :(
2015-08-11 22348, 2015
alastairp
* :)
2015-08-11 22343, 2015
Gentlecat
ok
2015-08-11 22301, 2015
Gentlecat
do we still need to lock dataset during evaluation?
2015-08-11 22315, 2015
alastairp
hmm
2015-08-11 22321, 2015
alastairp
actually, I don’t know
2015-08-11 22326, 2015
alastairp
what are the situations:
2015-08-11 22337, 2015
alastairp
1) if it’s in the queue, and you edit, the edited one will be evaluated
2015-08-11 22312, 2015
alastairp
2) if it’s started, the evaluation will be happening on extracted yaml files, and so changes won’t affect it
2015-08-11 22334, 2015
alastairp
do we have implemented something which says you can’t re-submit if it’s in process?
2015-08-11 22335, 2015
Gentlecat
data is copied before evaluation starts, not when it's added into the queue
2015-08-11 22305, 2015
Gentlecat
what do you mean re-submit?
2015-08-11 22311, 2015
Gentlecat
re-submit for evaluation?
2015-08-11 22350, 2015
Gentlecat
you can't create evaluation job for a dataset if there's already one running/in the queue
2015-08-11 22333, 2015
alastairp
right, that’s what I mean
2015-08-11 22359, 2015
alastairp
and if you open a dataset that is in the queue, does it show a message telling you this?
2015-08-11 22322, 2015
Gentlecat
no
2015-08-11 22306, 2015
alastairp
let’s do that, and then leave it
2015-08-11 22326, 2015
Gentlecat
evaluator just picks current version of the dataset
2015-08-11 22332, 2015
alastairp
fine
2015-08-11 22341, 2015
Gentlecat
there are no snapshots or anything like that
2015-08-11 22348, 2015
alastairp
“this dataset is in the queue, if you edit it, the updated version will be used"
2015-08-11 22312, 2015
Gentlecat
"...if you edit it before evaluation starts..."
2015-08-11 22320, 2015
alastairp
“this dataset is being evaluated. you can edit it, but the changes won’t take effect (and you won’t be able to submit it) until the current evaluation has finished"
2015-08-11 22343, 2015
alastairp
yeah, the thing I don’t like here is that there’s a possible small race condition
2015-08-11 22354, 2015
alastairp
not quite a race condition, but if you think it hasn’t started
2015-08-11 22356, 2015
alastairp
and you make a change
2015-08-11 22306, 2015
alastairp
but it starts before you save
2015-08-11 22312, 2015
alastairp
though, I’m not too worried about that. I’d rather make some tickets for future work and move on to #3 or #4
2015-08-11 22320, 2015
Gentlecat
I'm probably not going to have enough time to implement #3 and #4
2015-08-11 22342, 2015
alastairp
yeah - just 1 is fine
2015-08-11 22357, 2015
Gentlecat
*or
2015-08-11 22304, 2015
Gentlecat
maybe .history file generation
2015-08-11 22305, 2015
alastairp
#3 would give us complete end-to-end dataset, model generation and evaluation
2015-08-11 22317, 2015
alastairp
hmm
2015-08-11 22325, 2015
Gentlecat
and I still have a couple of bugs to fix
2015-08-11 22326, 2015
alastairp
you don’t think you have enough time even to do one?
2015-08-11 22349, 2015
alastairp
(one sec, I’m just heading to the office. I’ll be back in 20)
2015-08-11 22339, 2015
reosarevok joined the channel
2015-08-11 22346, 2015
Gentlecat
in #3 it's still unclear after we have the model
2015-08-11 22318, 2015
Gentlecat
everything after and including "To evaluate at a larger scale..."
my thoughts are that after this step, someone looks at their "83% accuracy" an says "cool, I want to use this model to actually generate high-level data"