Superintendent documentation

Practical data labelling and active learning in Jupyter notebooks.

superintendent is a set of ipywidget-based interactive labelling tools for your data. It allows you to flexibly label all kinds of data.

It also allows you to combine your data-labelling task with a statistical or machine learning model to enable quick and practical active learning.

For example:

Note

Throughout the documentation, all widgets will look like they do in a jupyter notebook. However, clicking on any of the buttons will not trigger anything. This is because there is no running backend. To really get a feel for how it all works, you should install superintendent and try it out in a notebook.

Contents:

• Installation
  • Development installation
• Labelling data
  • When do you need data labelling?
  • Quickstart
  • Necessary components to label data
  • Elements of a Superintendent Widget
• Active learning
  • What is active learning?
  • Active learning in superintendent
  • Active learning strategies
  • Active learning for multi-output widgets
  • Preprocessing data before passing it to the model
  • Which model to choose
• Distributed labelling
  • Distributing the labelling of images across people
  • Retrieving data from the distributed widget
  • Doing active learning during distributed labelling
• API Reference
  • Acquisition functions

Example gallery:

• Examples
  • Labelling text
  • Labelling images
    • Labelling images randomly
    • Labelling images with active learning
  • Custom pre-processing steps as part of superintendent
    • Data: Emails with metadata
    • Display function and annotation widget
    • Model Pipeline
    • The widget
  • Using docker-compose to serve a superintendent interface to many users
    • docker-compose
    • The notebook (our webapp)
    • The orchestration script (our machine learning model)
    • Dockerfiles
      • Web application (voila) dockerfile
      • Model training dockerfile
    • Starting

Indices and tables

• Index

• Module Index

• Search Page