Kaggle Best Practices

[westurner]

• Research Kaggle Best Practices

[westurner]

Research kaggle best practices

• https://www.google.com/search?q=kaggle+best+practices
  • https://www.kaggle.com/wiki/ModelSubmissionBestPractices
    • README.md
    • SETTINGS.json
      • TRAIN_DATA_PATH: input/
      • TEST_DATA_PATH: input/
      • MODEL_PATH: working/
      • SUBMISSION_PATH: working/
    • train.py
    • predict.py
• https://www.google.com/search?q=how+to+structure+a+kaggle+git+repository
  • https://github.com/mikezawitkowski/kaggle_filestructure
    • input/
    • src/
    • working/
• https://www.google.com/search?q=kaggle+scripts
  • https://www.kaggle.com/c/introducing-kaggle-scripts
  • http://blog.kaggle.com/2016/07/08/kaggle-kernel-a-new-name-for-scripts/
    • "Kaggle Scripts" is now "Kaggle Kernels"
• https://www.kaggle.com/kernels
• https://www.google.com/search?kaggle+kernels+github
• https://github.com/Kaggle/docker-python
  • https://github.com/Kaggle/docker-python/blob/master/Dockerfile
    • Ubuntu, Conda, *
  • http://blog.kaggle.com/2016/02/05/how-to-get-started-with-data-science-in-containers/
  • https://www.kaggle.com/wiki/Scripts
    • /kaggle/input
    • /kaggle/working
  • https://www.kaggle.com/forums/f/15/kaggle-forum/t/13628/python-environment-for-kaggle-scripts

So, a git repository which will be cloned or mounted in a docker container (as /kaggle with kaggle/docker-python):

• Dockerfile
  • (FROM continuumio/anaconda3:latest)
  • FROM kaggle/docker-python:latest
  • # RUN conda env update -f=/kaggle/src/[repo/]environment.yml
• input/ -- data {data_description.txt, train.csv, test.csv, }
• src/
  • src/train.py
  • src/predict.py
  • src/repo -- git submodule(s)?
    • src/repo/environment.yml
    • src/repo/Makefile
• working/ -- output {trained models, predicted data, logs}
  • working/ should be added and then .gitignore'd to prevent uploading

[ ] How to handle multiple organization users' approaches with everything in src/?

• git submodules
  • src/repo -- submodule
  • https://github.com/blog/2104-working-with-submodules
  • what about output in working/ ?
    • how to track which code produced which prediction.csv?
      • nothing in working/ should be checked in or shared (at least until after the competition closes)
      • as a team, we commit to always store the commit hash for the entered solution?

        • cd src/..
          git commit
          git rev-parse --short HEAD | tee SUBMISSION_PATH/shortrev.txt;
              cd working/;
          python ../src/train.py;    # > MODEL_PATH/
          python ../srcpredict.py;  # > SUBMISSION_PATH/
             # post (shortrev.txt, prediction.csv, train_MSE) to {Slack,?}
             # post prediction.csv to <https://#kaggle_competition>

[westurner]

From http://blog.kaggle.com/2016/02/05/how-to-get-started-with-data-science-in-containers/ :

kpython(){
  docker run -v $PWD:/tmp/working -w=/tmp/working --rm -it kaggle/python python "$@"
}
ikpython() {
  docker run -v $PWD:/tmp/working -w=/tmp/working --rm -it kaggle/python ipython
}
kjupyter() {
  (sleep 3 && open "http://$(docker-machine ip docker2):8888")&
  docker run -v $PWD:/tmp/working -w=/tmp/working -p 8888:8888 --rm -it kaggle/python jupyter notebook --no-browser --ip="\*" --notebook-dir=/tmp/working
}

• What's a good solution for working with an IDE which supports code-completion which needs filesystem access to the installed libraries within a container?