search

Unity-Technologies / obstacle-tower-env

Obstacle Tower Environment
Apache License 2.0
542 stars 125 forks source link

GCP - Deployment fail - Tensorflow resource warnings #56

Open psarras opened 5 years ago

psarras commented 5 years ago

Hi,

I am trying to follow the tutorial for deploying on GCP. However I am getting:

image

Is this because my account is free? Is there a way to run this challenge with the free account or do we need to use a loca machine?

Thanks

awjuliani commented 5 years ago

Hi @psarras

Does this page help with your issue: https://stackoverflow.com/questions/45227064/how-to-request-gpu-quota-increase-in-google-cloud ?

psarras commented 5 years ago

Hi @awjuliani

Yes I think it does. But I guess this means that free accounts cannot access any GPU resources.

I tried on Friday to install everything locally and do the training on my machine but I am on windows. Can you confirm if this setup works only on a linux machine? I think dopamine is the dependancy I am missing.

Sohojoe commented 5 years ago

@psarras - i was able to get it running on my windows machine. I followed the GCP tutorial. You don't need to install dopamine as the tutorial has you running it from the dopamine directory. But you do need to install dopamine's dependencies (which are listed in the tutorial).

You do not need to worry about the XServer stuff - this is a hack to make rendering work on a headless server

psarras commented 5 years ago

Thanks for you reply! @sohojoe

Yes you are right I was trying that as well. How did you overcome the Atary installation? I think that needed "make" isn't that Linux only? Or is Atary not needed for Obstacle tower?

Sohojoe commented 5 years ago

I followed instructions for MSI - https://github.com/j8lp/atari-py

Arishtanemi2 commented 5 years ago

Hi @awjuliani

Yes I think it does. But I guess this means that free accounts cannot access any GPU resources.

I tried on Friday to install everything locally and do the training on my machine but I am on windows. Can you confirm if this setup works only on a linux machine? I think dopamine is the dependancy I am missing.

Hey, this is just a suggestion.I am using the free tier too and all i had to do was enable billing on my account and i could raise the quota limit.Now my training still consumes those free credits till they run out.

psarras commented 5 years ago

@Sohojoe

I am trying what you suggested, Dopamine still has trouble picking up on the atary_py, but I will try to retrace my steps. Thank you for the suggestion.

@awjuliani

Yes you are right, I just done it and waiting approval. Will give this option a go as well, but I don't think it will be sustainable for over a large period of time for me though. Thank you as well!