Quckly set up new machine learning projects that make best practices easier.

Successful machine learning requires a long list of best practices from research and software engineering. This project helps to enforce an opinionated but configurable workflow that supports

• Reproducible research
  • Persistance of all trained models
  • Logging of all training details needed for reproductions
  • Docker integration for reproducing exact versions of all software dependencies
• Systematic experimentation with different modeling choices and hyperparameters
• Configurable interfaces to support the needs of different organizations and individuals

• New projects are set up with a single command, a la Cookie Cutter
• Manage training many different configurations by specifying the desired experiments in configuration files
• Automate the submission of dockerized training jobs
• Manage production deployments of trained models
• Integration with ModelDB so that all experiments are recorded
• Configurable interactions with your organization's computing platform
  • Training job submissions
  • Trained model deployment
  • Database for persisting models and experiment metadata