• Divisi2 is embedded in a larger project on knowledge representation (semantic networks and the like) in the MIT, so it is intended to be used in applications in AI and not in linguistics.
  • It is a SVD (singular value decomposition) library for Python.
  • Divisi requires numpy (numerical python, a standard library for scientifical computing with py) and pysparse (a further extension for matrix processing). From my own experience, installing numpy is not trivial.
  • As far as I can see right now, it is necessary to develop or use another application to collect the context vectors, etc..
  • Easy installation (around 5 minutes if you have numpy).
  • Note: In a first run, Divisi2 didn't work.


  • I presuppose that Python is already installed in your computer. Most distributions of linux include in their standard version Python. If you haven't install python yet, please check here.
  • The installation process is by itself quite simple -you will need just two commands. However a number of libraries are necessary to run Divisi2 and their installation is not quite simple, at least that is what I remember from installing numpy. In my case, I had to install (and maybe build) several libraries, in particular, ATLAS.
  • To install:
    1. First install pip
      sudo easy_install pip
    2. This command will also install pysparse (in case you don't have it):
      sudo pip install pysparse
    3. Now install Divisi2
      sudo pip install divisi2 csc-pysparse
  • For a further tutorial using Divisi2

Technical Issues

  • Divisi2 utilizes Pysparce by means of a classes called SparseMatrix- Pysparce is can be seen as an extension of Numpy. It supports three formats for saving sparse matrices:
    1. Linked-list format
    2. Compressed sparse row format
    3. Sparse Skyline format


Eduardo Aponte 2010/11/16 10:38