Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
    Arthur Juliani
    @awjuliani
    Hi @KarolisRam Could you possibly record a video of the agent? It might have discovered a bug that allows it to skip floors :)
    KarolisRam
    @KarolisRam
    I'll try. Sadly I wasn't setting/saving random seeds, I will try to do a sweep of seeds and figure out what's happening.
    KarolisRam
    @KarolisRam
    hmm even with a set seed the agent starting position seems to be a bit random (the agent is sometimes a bit further forward). Is this intended?
    SP Mohanty
    @spMohanty
    Anouncement : https://discourse.aicrowd.com/t/announcement-upcoming-updates-about-tag-pushes/932
    parilo
    @parilo
    @spMohanty please help with submission https://gitlab.aicrowd.com/parilo/obstacle-tower-challenge/issues/20
    SP Mohanty
    @spMohanty
    I will be offline soon, but managed to catch this one fast enough
    but please do try running your agents locally before submitting
    parilo
    @parilo
    I tried and it worked well
    SP Mohanty
    @spMohanty
    Many of the errors that we see are usually missing dependencies, etc
    I pasted your error logs here : https://gitlab.aicrowd.com/parilo/obstacle-tower-challenge/issues/20
    parilo
    @parilo
    Thank you!
    SP Mohanty
    @spMohanty
    Looks similar to : Unity-Technologies/ml-agents#1967
    parilo
    @parilo
    Maybe I need to rebuild docker image, I use ObstacleEnv 1.2 - is it a problem?
    SP Mohanty
    @spMohanty
    Shouldnt be !! But now that I look closely to your error
    yes, its unusual why it has an issue with finding mlagents
    :/
    well the image is rebuilt on every evaluation
    But you have a successful evaluation from a month ago here : https://gitlab.aicrowd.com/parilo/obstacle-tower-challenge/issues/19
    what changed ?
    SP Mohanty
    @spMohanty
    ===
    In any case, I will have to go offline soon
    but atleast your score of > 5 qualifies you for the second round
    parilo
    @parilo
    But I'm at 51 place
    I thought only first 50 will go to second round
    I will try with debug submission
    SP Mohanty
    @spMohanty
    Okay sure
    parilo
    @parilo
    Thank you!
    SP Mohanty
    @spMohanty
    I will refrain from publicly commenting on the eligibility criteria !!
    But I can still voice my strong position from the beginning to allow everyone to take part in subsequent rounds.
    :angel:
    I am around for another 30 mins, and can quickly try to see why your submission fails
    btw
    running it in debug mode
    will still give you only the same logs
    I pasted on your submission issue
    parilo
    @parilo
    Cool!
    Now trying to rebuild the image and test it locally, so it will be looong way)
    SP Mohanty
    @spMohanty
    Maybe just try to rebuild the image locally
    and get a bash in there and check if it can import the mlagents_env inside a python interp
    and if not why
    should be a simple software runtime spec issue
    parilo
    @parilo
    Ok, will try it
    Anssi
    @Miffyli
    Hmm... Frontpage says "In order to move onto Round 2, an Agent must be able to reach an average score of 5 in Round 1" but rules say "top50 participants" with no mention of five floors
    SP Mohanty
    @spMohanty
    Clarification on the rules can only be provided by sending an email to : OTC@unity3d.com
    Anssi
    @Miffyli
    Figured. Hope they take your comment and let everybody join, the more the merrier :D
    SP Mohanty
    @spMohanty
    Its important to note here though that there are two notions of eligibility :
    • eligibility for Round-1 prizes ( $1100 of GCP credits)
    • eligibility to participate in Round-2
    best to send an email to the email address mentioned above and ask for a clarification, so that we can give an official response on the same
    Anssi
    @Miffyli
    But under Eligibility Criteria it says the "In order to move onto Round 2, an Agent must be able to reach an average score of 5 in Round 1."
    Alright, I'll do that and ask them to share thoughts here / on discussion forum (seems like I am not the only one confused)
    SP Mohanty
    @spMohanty
    Makes sense !
    parilo
    @parilo
    Ok, I got locally
    ModuleNotFoundError: No module named 'mlagents_envs'
    SP Mohanty
    @spMohanty
    Can you fiddle around a bit to see why ?