Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Repo info
Activity
    Jakub
    @jakubczakon
    are enough to test if it should run ?
    SP Mohanty
    @spMohanty
    Unless you have a lot of disk IOs in your code, I dont see how device-write-bps and device-read-bps would be a bottleneck.
    And yes, memory of 8gb and cpu=1 is a good starting point.
    I am happy to increase the memory avaialbility to upto 16gb if that helps.
    Jakub
    @jakubczakon
    from my local test it really doesn't matter. It goes through that dataset under 1cpu na 8g easily in under 5h
    SP Mohanty
    @spMohanty
    @jakubczakon : How many images do you have in your local test
    ?
    Jakub
    @jakubczakon
    60k
    from what I gather there is 30k in the stage2 test set
    ok overoptimistic here. it goes through in 1h so over 5h for 60k but still
    SP Mohanty
    @spMohanty
    Can you point me to the issue where you got the timeout
    ?
    Jakub
    @jakubczakon
    sure
    SP Mohanty
    @spMohanty
    the first one didnt timeout. It seems to be some other issue.
    the second one did. It was progressing really slowly !
    If you make another submission like the second one, I can keep an eye on whats hapenning inside
    Jakub
    @jakubczakon
    Great, thanks! I will submit that in a moment
    Janick Rohrbach
    @janickrohrbach_gitlab
    @spMohanty Seems like your server is clogged again. Multiple submissions from different people are pending at the moment. I assume they are all queued for evaluation. Could you please take a look at this? https://www.crowdai.org/challenges/mapping-challenge/submissions
    SP Mohanty
    @spMohanty
    @janickrohrbach_gitlab : Our for lunch at the moment. Will get to it right after lunch. Sorry about that.
    The evaluation system still needs some polishing.
    SP Mohanty
    @spMohanty
    @janickrohrbach_gitlab : Okay figurred out whats happenning.
    My hunch was it was the last few days of the VizDoom challenge and hence there was a lot of activity there.
    But turns out there was an issue with the timeout, and 3 of your slow submissions were running for around 20 hours
    I just manually flushed all three of them
    Janick Rohrbach
    @janickrohrbach_gitlab
    Thanks for looking into it. I hope the submissions go through tonight.
    Jakub
    @jakubczakon
    Hi there. I had a lot of trouble getting second level model to work under competition constraints. What worked for me was to port it from lgbm to pytorch and heavily cut the feature extraction on the second level. I hope this helps.
    Jakub
    @jakubczakon
    @spMohanty I have a question about the time limit. It reads in the repo that
    Note Your evaluation will have a total timeout of 5 hours.
    I am wondering is it hard or soft limit? I had some evaluations that ran for longer than that. Finally yesterday I managed to have it fully under 5 hours from "evaluation initiated" till "evaluation successfull" but I am wondering if a longer evaluation would go through as well.
    I've put a lot of work in having it work in under 5 hours so it would be a bummer if it was a soft limit.
    SP Mohanty
    @spMohanty
    @jakubczakon : By design, the 5 hours timeout are from the beginning of the pod execution to the end of the pod execution.
    In many cases, the job is queued in the setup for a long period of time before being actually assigned resources, thats why some of the jobs ran for more than 5 hours.
    But the effective time when the container was running will always be less than 5hours.
    Jakub
    @jakubczakon
    I see that's cool. So pretty much if it goes through to "evaluation successfull" it means that it was under 5hours of container running.
    SP Mohanty
    @spMohanty
    @jakubczakon : Correctt.
    vulpetYTH
    @vulpetYTH
    @spMohanty Regarding to the late opening of stage 2, could you extend the deadline to 8.31 ?
    Participants could use the extending time to prepare their training model and submit format.
    And I think this change will attract more participants to submit their solution.
    SP Mohanty
    @spMohanty
    @vulpetYTH : We will unfortunately have to stick to the current deadline of 8.20, because we will also have to do the travel arrangements for the top parrticipants to the conference in Italy.
    and a generarl announcement, that the evaluator went crazy last evening, and hence all the queued jobs had to be removed.
    If your submission is still stuck at "waiting in queue for evaluation", please consider resubmitting.
    Also, please be considerate that many participants would be trying to submit, and hence please do not submit more than one submissions.
    Or not submit at all, if you think you have the best score.
    Sorry for the inconvenience.
    Annie
    @annieyan
    I am wondering if this challenge dataset can be used for research purposes? I will have to miss the challenge deadline...
    Manjunath Prasad
    @mprasad10_gitlab
    @spMohanty As per your general announcement, I'm trying to submit one last time, but it seems all my submission is getting stuck at "waiting in queue for evaluation". Could you let me know, can I re-submit one last submission?
    SP Mohanty
    @spMohanty
    @mprasad10_gitlab : Unfortunately not anymore as the challenge has ended.
    @Annie : Yes you can. We will release all the citation requirements soon.
    SP Mohanty
    @spMohanty
    Some of the top participants should be at IEEE DSAA, I believe, it would be great to meet up :D
    I arrived an hour ago during lunch !
    Manjunath Prasad
    @mprasad10
    Hi, I'm at the conference.
    SP Mohanty
    @spMohanty
    Heading to the coffee area in a min. I should be easy to spot :D long hair and a leather jacket :D
    Manjunath Prasad
    @mprasad10
    Will catch you up here
    Janick Rohrbach
    @janickrohrbach_gitlab
    Hi, I just arrived in Turin and will go to the conference now
    saket kunwar
    @saketkunwar2005_gitlab
    arrived in turin but very late.. will come 2maro Oct 2 before lunch or earlier.
    daveluo
    @daveluo
    hi, I should be at conf by lunch or earlier as well
    saket kunwar
    @saketkunwar2005_gitlab
    I am at the conference.. it's at a different place than that of yesterday
    SP Mohanty
    @spMohanty
    This is the place we will be meeting for beers : Open Baladin Torino
    Piazzale Valdo, Fusi, 10123 Torino TO, Italy
    +39 011 835863
    https://goo.gl/maps/hDTiv2XMrRx
    BigDudeSuperstar
    @mr.heineken_gitlab
    Hi all, is anyone still around here? I'm yearning to reproduce the results, yet I cannot register anymore at CrowdAI, could anyone share the data please?
    BigDudeSuperstar
    @mr.heineken_gitlab
    @spMohanty anyone?
    Alex
    @A1exRobson
    Hi everyone, I am trying to access the dataset for this challenge but can't since crowdAI is closing and they no longer allow new account creation. I was wondering if anyone might supply an alternative location I can attain the data (all the git repos I can find, the data is in .gitignore). Honestly at this point even a scrappy zip would be a godsend. Does anyone still have access that could help?
    Jakub
    @jakubczakon
    Hey Alex, crowdAI is closing but AIcrowd seems alive and kicking! Maybe try it there https://www.aicrowd.com/challenges/mapping-challenge