Paramerator: Parameter Management System

Whats the issue?

Keeping track of all of the things you’ve tried and tested when training a machine learning model is tedious.

A secondary issue is that uploading files with sensitive data to Github is a bad idea, but if those files are needed for your project, what do you do?

How do you fix it?

With Paramerator! <- download link

  • Load hyper-parameters (or any options really) from disk
  • Save to disk
  • Build a dictionary of ‘default files’ and their parameters
  • Force user input for missing required parameters like IP addresses, passwords and API keys
  • Works great in Jupyter Notebooks!
  • Parameters become nested attributes so you can “tab-complete” all of them
  • Type-aware.  Store any standard python dtype
  • Parameter files are human-readable and editable

The Notebook.

Leave a Reply

Your email address will not be published. Required fields are marked *