Friday, November 4, 2011

OSCATS 0.6 Released

The overdue version 0.6 of OSCATS is now available in source and Windows binaries (32- and 64-bit). New 32- and 64-bit Windows binaries of GSL are also available.  While functional, this release should still be considered another in a line of Beta releases leading up to version 1.0.  Testers wanted!  Send feedback to the mailing list or file an issue for coding errors.

Major changes in this release include:

  • New OscatsSpace and OscatsPoint classes for a unified representation of continuous, binary, and ordinal latent spaces
  • Unification of the OscatsContModel and OscatsDiscrModel into a single OscatsModel based on the new latent space representation
  • Generalization of OscatsItem to allow any arbitrary number of models for complex simulation studies
Other highlights include:

  • Implementation of the a-Stratified item selection algorithm
  • Newly implemented models:
    • Partial Credit, Generalized Partial Credit
    • Graded Response (Homogenous and Heterogenous Logistic)
  • New examples, including implementation of a custom algorithm in Python
For a complete list of changes, see the repository commit log.

Plans for upcoming development include:
  • Support for simultaneous testing of multiple examinees
  • Exposure control and item selection constraints
  • R Bindings
  • Support for GObject Introspection
  • Example integration with Concerto for CAT administration

Tuesday, August 30, 2011

Upcoming Developments

After a hiatus, development on OSCATS is continuing.  Below are listed some changes to appear in the Hg repository soon, and the next release is planned for 1-2 months from now.

  Also, there are some new (and rather experimental) introspection-based MATLAB bindings for GObject at the gmatlab project.  They're ready for general use, but those interested who have some stamina may want to take a look.

  • GObject-Introspection support for liboscats
  • Shift language bindings to GO-I methods (perl, python, java)
  • Unified latent space object for continuous and discrete latent variables
  • Unified model support, allowing each Administrand to have an arbitrary number of models (for different purposes, e.g. simulation, estimation, calibration, statistical interest)
  • Support sub-classing models and algorithms in bound languages
  • More common IRT/CD models
  • More item selection algorithms, including a-Stratified and Dual-purpose methods
  • Simpson-Hetter exposure control