- allow for arbitrary weights; 
  possible missings could then enter the model with weight zero and be easily
  estimated due to the use of basis functions.

- add calculation of confidence intervals for effects and fitted values;
  determination of the variances involve matrix inversion which might become
  problematic for huge data sets if tensor product approach applied.

- allow 1d case

- the sequential variant of 'svcm' can be extended to *any* dimension  

- introduce memory check before increasing the resolution