Page History

• Do we want to continue using docker containers? Yes

Was able to implement a front end that asks the user for a docker run command: Image Added Implemented a worker queue that allows a separate VM to run the docker job so that a large worker VM with GPUs can be used sparingly versus constantly if it hosted the website. Things that could be improved:

• The participant dockers seemed to be able to delete a couple of images. I think participants shouldn’t be able to adjust files their containers are mounted to. This might be as simple as read-only permissions. Cuong says we can mount files to dockers with read-only permissions.

• We should finalize metrics for the evaluate.py script so that the final leaderboard columns are as we expected.

• We should know if we will need to use shm (shared memory size) before hand:

$ sudo docker run --runtime=nvidia --shm-size 50G -v /home/bbearce/input_images:/data/images -v /home/bbearce/yunzhe_docker_run/output_result:/data/output xyzacademic/njit_datasci:cpmv

• Remember to delete the index.html files that come with a curl or wget command. This caused some errors and confusion for participants.

• Docker worker container needs updated python for certain scikit-learn methods.

• Is there any way to check if a team has multiple screen names or accounts ahead of time?

• FTP:

• Server on same VM as CodaLab.

• Have data available all year so participants have more time.