[SUPPORT] Workbench Example Project: Competition Kernel

Hi! This is the support thread for the Competition Kernel project, an example project that lets users bring their own compute instance to the Kaggle competition platform. Any major updates we push to the project will be announced here. Further, feel free to discuss, raise issues, and ask for assistance in this thread.

Please keep discussion in this thread project-related. Any issues with the Workbench application should be raised as a standalone thread. Thanks!

This project sounds interesting. Can’t wait to see the details.

Project is live! Note any bugs or issues you run into in this thread. Thanks!

The project configuration implies you must set your host mounts for Kaggle input and output directories.

The README.md mentions setting the Kaggle keys. It does not mention setting the mount targets; it just moves directly to running the Jupyter Notebook. workbench-example-competition-kernel/README.md at main · NVIDIA/workbench-example-competition-kernel · GitHub

  1. The Windows file systems are auto-mounted into the Linux WSL instances in /mnt/ as /mnt/c, /mnt/d, etc.
  2. Which Linux instance is being used for the /mnt/c passthrough? The AI-workbench instance or the docker WSL instance?

The README could also include the path to the jpynb examples. There are in /kaggle/working. I’m not a kaggle person maybe that is a standard format
¯\(ツ)

Get an error 01-data.ipynb when running the download step. The error doesn’t appear to affect the download.

Error in sitecustomize; set PYTHONVERBOSE for traceback:
PermissionError: [Errno 13] Permission denied: '/tmp/kaggle.log'
Downloading digit-recognizer.zip to /project/kaggle/input
 72%|███████████████████████████▎          | 11.0M/15.3M [00:00<00:00, 52.4MB/s]
100%|██████████████████████████████████████| 15.3M/15.3M [00:00<00:00, 53.7MB/s]
Archive:  digit-recognizer.zip
  inflating: digit-recognizer/sample_submission.csv  
  inflating: digit-recognizer/test.csv  
  inflating: digit-recognizer/train.csv  

This project is well organized. The plots in the steps are really useful visualizations.

40s to train on an NVidia Titan RTX (not supported by the project)

Comparing GPU and CPU using the Kaggle Competition Container in NVidia AI Workbench