Welcome to chronowords’s documentation!

chronowords is a Python package for training small PPMI-based language models, making topic models via non-negative matrix factorization, and detecting semantic shift over time using the Procrustes method.

Built and maintained by Crow Intelligence.

Install from PyPI.

New here? Start with the Quickstart, then follow the Tutorial: Detecting Semantic Shift Over Time to detect how a word’s meaning shifts across time.

Contents:

• Installation
  • Install from source
  • Development install with uv
• Quickstart
  • Install
  • Build embeddings from a corpus
  • Query the embeddings
  • Save and reload a model
  • Extract topics
  • Next steps
• Tutorial: Detecting Semantic Shift Over Time
  • 1. Train one embedding space per slice
  • 2. Align the two spaces
  • 3. Score how much each word shifted
  • 4. Interpret a shift
  • 5. (Optional) Track topics across slices
  • Where to go next
• Troubleshooting
  • Training
  • Memory and accuracy (Count-Min Sketch)
  • Saving and loading
  • Alignment
  • Topic modeling
  • Build / install
• Examples
  • The worked example
  • Short snippets
• API Reference
  • Word Embeddings
  • Alignment
  • Topic Modeling
  • Utilities

Indices and tables

• Index

• Module Index

• Search Page