Corncob is a collection of C classes (yep, you read that right) for building probabilistic models based on corpus counts, and other compling-y things. It includes homebrew utilities for all kinds of things, including:

• C-string to integer maps
• integer to integer maps
• Sparse count vectors
• Wordmaps
• Task-specific linked lists

As well as full-blown implementations built on these tools for a growing number of existing models, including:

• Blei et al.'s Latent Dirichlet Allocation (lda)
• Anderson's Rational Model of Categorization (rmc)

I'm not claiming that any of this is good design, by the way. I'd probably be better off using C++ and the STL, or maybe Boost, or something -- but this way lets me exercise some rusty data-structures-and-algorithms parts of my brain, and gets my C hackery back up to speed. Good times.

UML-type documentation for these classes can be generated using crud.