To reduce the computational burden, we will add options for
selecting a subset of representers among all
representers 's in the solution ().
Inferences on nonparametric functions are usually accomplished
using Bayesian confidence intervals. However, they do not
provide pointwise coverage, nor a single p-value. Hypothesis
tests are only available for simple spline models
(Liu et al., 2004; Liu and Wang, 2004). Further research on model selection is also needed.
One of our future task is to extend the `anova` function
to perform hypothesis tests for more complicated models, and
to compare different models.

Karcher and Wang (2002) proposed the SLM models for
correlated non-Gaussian data. Thus we can extend the `slm`
function for non-Gaussian families.

Yuedong Wang 2004-05-19