Future Directions for Statistical Modeling Software
Why re-think things now?
- Not all modeling concepts from S/R will translate
well into Omega
- Some modeling concepts from S/R deserve re-thinking
- CORBA encourages thinking of interfaces distinct from implementation
Other initiatives
Numerical Linear Algebra
- Two implementations in Java from NIST in the US -
JAMA
and Jampack
- Neither seem to be "the answer" yet
- Don't seem to be based on interfaces
Components for Statistical Models
- Need data in some form
- Class-wise concepts - fitting method, etc.
- Instance-wise concepts - formula language
Does the formula/data-frame concept hold up?
- Wilkinson and Rogers language for linear models
- General expressions