Hello World! I'm Nikhil, an undergrad from BITS Pilani, India. I'm interested in working on Docker-based model training for 3rd party ML algorithms. Where can I discuss about this further?
2017-03-14 07328, 2017
saifulbkhan joined the channel
2017-03-14 07318, 2017
Leo_Verto[m]
Right here :P
2017-03-14 07320, 2017
Leo_Verto[m]
Welcome!
2017-03-14 07304, 2017
nikhilweee
Thank you! What are the next steps to be followed? I'm really interested in this issue.
akashn97 find out where the documentation is displayed for the relevant things first
2017-03-14 07317, 2017
samj1912
Thats the first step
2017-03-14 07321, 2017
nupurbaghel_ joined the channel
2017-03-14 07336, 2017
nupurbaghel_ has quit
2017-03-14 07340, 2017
samj1912
Then grep related string to find out the file
2017-03-14 07317, 2017
samj1912
I don't want to just point the file to you, you won't learn how to dive into code otherwise
2017-03-14 07353, 2017
samj1912
This is a fairly easy task
2017-03-14 07306, 2017
samj1912
And you are expected to do this on your own
2017-03-14 07312, 2017
lazka has quit
2017-03-14 07357, 2017
samj1912
Hint : find the correct template for documentation
2017-03-14 07359, 2017
nupurbaghel_ joined the channel
2017-03-14 07319, 2017
nupurbaghel_ has left the channel
2017-03-14 07339, 2017
akashn97
Alright, I got it. Thank you :)
2017-03-14 07342, 2017
nupurbaghel_ joined the channel
2017-03-14 07312, 2017
nupurbaghel_ has quit
2017-03-14 07305, 2017
SothoTalKer
wow, i haven't checked out picard code in a long time XD
2017-03-14 07337, 2017
github joined the channel
2017-03-14 07337, 2017
github
[picard-website] grassknoted opened pull request #65: PW-33 : Documentation changes for PICARD-923 (master...patch-1) https://git.io/vyXQK
2017-03-14 07337, 2017
github has left the channel
2017-03-14 07356, 2017
raymank26 joined the channel
2017-03-14 07329, 2017
akashn97
samj1912: Thank you for your help. :)
2017-03-14 07304, 2017
akashn97 has quit
2017-03-14 07349, 2017
nikhilweee
Leo_Verto[m]: Thanks for your help! I have read the getting started page. The issue I'm interested in (docker-based model training) does not have many details like the other ones, so I thought of discussing more about it here.
2017-03-14 07308, 2017
alastairp
hi nikhilweee
2017-03-14 07314, 2017
alastairp
this is indeed the right place to discuss this
2017-03-14 07356, 2017
alastairp
although I'm quite busy until Thursday afternoon, and so don't have much time to talk before then
2017-03-14 07311, 2017
alastairp
what questions do you have?
2017-03-14 07320, 2017
nikhilweee
Hi alastairp! Thanks for your reply. The issue description says "Researchers can provide us with an image which has a known API". As far as I can comprehend, does this mean developing the API (so that submitters can create images accordingly) and make relevant changes to acousticbrainz-server so that such images can be evaluated easily?
2017-03-14 07327, 2017
nikhilweee
Would this be the whole objective of the to-be GSoC project? Or is there something else we could add? I'm keen to know your suggestions.
2017-03-14 07340, 2017
amanmehta has quit
2017-03-14 07326, 2017
alastairp
nikhilweee: here when we say API I mean a common interface, not a "web api"
2017-03-14 07304, 2017
alastairp
so, to take an example, you could have some software where you have to call "/usr/local/bin/train_algorithm <trainingdata>"
2017-03-14 07330, 2017
alastairp
where trainingdata is a directory containing directories of class names, and inside those directories are examples of each class
2017-03-14 07322, 2017
alastairp
so if someone wanted to make their own algorithm, they could provide a docker image which had their software installed as this `/usr/local/bin/train_algorithm` exectuable
2017-03-14 07326, 2017
alastairp
and the job of the acosuticbrainz server is to take the dataset (from our existing dataset editor), collect the necessary data files, format them and put them in the correct data structure, and then run the docker image which contains this algorithm
2017-03-14 07340, 2017
alastairp
then read and interpret the results
2017-03-14 07324, 2017
saifulbkhan_ joined the channel
2017-03-14 07305, 2017
alastairp
it's also worth mentioning that we had a student who worked on a system very similar to this in SoC last year. It was to run the training process on a remote server, using our existing tools
2017-03-14 07358, 2017
alastairp
so we already have some code which can prepare this data, download it over a web api, process it on a remote server, and then submit the results back to the main acousticbrainz site
2017-03-14 07314, 2017
alastairp
I see this docker project as an extension of last year's project
2017-03-14 07349, 2017
hibiscuskazeneko joined the channel
2017-03-14 07332, 2017
nikhilweee
so the 'API' here more like a docker template, right? Which would expose known methods like <train> and <test>, for example. People can submit their images with these functions exposed, and you can run their algorithms without hassle!
2017-03-14 07344, 2017
alastairp
right
2017-03-14 07300, 2017
alastairp
I've not thought about the project much past this step
2017-03-14 07310, 2017
alastairp
it was more of a "oh, wouldn't it be interesting if we could..."
2017-03-14 07310, 2017
alastairp
the nice thing here is that it means that people are not limited to writing their algorithms in a particular programming language. As long as they implement a train and test entrypoint, we can use it regardless of language
2017-03-14 07356, 2017
nikhilweee
Thank you! I'd be really interested to turn this into a SoC project.
2017-03-14 07331, 2017
alastairp
if you want, you could open a discussion thread in our forums
2017-03-14 07343, 2017
gcilou heads to interview 😬
2017-03-14 07351, 2017
alastairp
gcilou: good luck!
2017-03-14 07358, 2017
gcilou
Thanks :)
2017-03-14 07304, 2017
CallerNo6
I was about to say, I'd really like to see this question/answer in the forums so that other people can benefit from it :-)
nikhilweee: It would be good if you could write down your expectations of what this project involves (based on this discussion)
2017-03-14 07359, 2017
nikhilweee
oh! that would be interesting! I'll move this over to the forums!
2017-03-14 07330, 2017
alastairp
you should also look at how the model training process currently works in acousticbrainz and show how you think this should fit into the existing process
2017-03-14 07352, 2017
alastairp
the idea here is that you should also show us that you understand what currently exists, and how it can be extended with this new system
2017-03-14 07309, 2017
alastairp
yes, that's the project
2017-03-14 07340, 2017
lazka joined the channel
2017-03-14 07357, 2017
deeptibaghel joined the channel
2017-03-14 07318, 2017
suhas2go joined the channel
2017-03-14 07327, 2017
github joined the channel
2017-03-14 07327, 2017
github
[acousticbrainz-server] saifulbkhan opened pull request #232: Provide custom error handlers for API blueprints (master...return-json-error-for-datasets) https://git.io/vy1I8
2017-03-14 07327, 2017
github has left the channel
2017-03-14 07327, 2017
agentsim has quit
2017-03-14 07332, 2017
agentsim joined the channel
2017-03-14 07335, 2017
Quesito
so when/if I find tickets like OTHER-43 can I delete them, to clean up...or what is the protocol?
