

Minor:

subset/na.action args

print: do smooths and ranefs separately

coef method

varying coefficient models: ensure identifiability for f1(x1)*u + f2(x2)*u

by-terms: check for/enforce treatment contasts?

Done(?): expandBasis: replace for-loops (and list str) with sth moore efficient, 
	takes forever if by has many levels

change formula evaluation etc to sth like used in mboost/lm (Manuel?)

add general GMRF-constructor (based on graph/adjacency matrix?) 

add Jan's ordinal predictor penalization (/Misc/linearDummies.r)

put all info on basisGenerators in a single helpfile?

reparameterizeDesign: use chol for full rank penalties to preserve sparsity

################

MAJOR:

2D: reparameterizeDesign: decompose into marginal smooths and interaction surface,  
	example in vignette,
	getF/plotF, 

class Term with Subclasses Smooth(1D/2D), Linear, Random (Spatial) --> predict, plot methods
(S4 or proto?)amer

make compatible to lme4a-branch (lo prio until DB gets active again, stops fiddling with Rcpp)
methods for dense RanEf-Designs/ backfitting?
   