CrossValidation Platform

Hold local Machine Learning contests in a kaggle like format.

How it works

• participants create accounts
• participants login
• participants navigate to the /contests link and see the list of available contests.
• participants click on the contest they wish to participate in
• partici.. ok. They sign contract
• download resources
• run computations on their machines
• create a submission file
• upload to website via submit option
• Get a score on their answers.

Leaderboard displays rankings and scores.

Installation

1. setup.sh will install all the prerequisites.
2. serve.sh will start serving the development server. This is not recommended for actual use.

Usage

• Set up
• go to /admin
  • add contest
  • Fill in the details
    • Ground Truth: this is the file expected from the perfect player submission
    • Check script: This is used to score the submission. It is provided with ground_path, given_path, pd, metrics in it's namespaces It must set the score variable
    • Resources:
      • you are expected to provide a train.csv and test.csv file.
      • The participant trains her model on train.csv and predicts for test.csv.
      • The predictions are uploaded and scored agains ground_truth

Contribution

Please do so!

TODO

• Cleaning and validating the submissions before running scoring
• Public Private lb split
• Submission limit must be counted in today's date
• UI change
• Systems Hardening
• Splitting the code check and web server code

Screenshots

These change on a regular basis.