Usage: 
mh = regestp(x {,h {,K {,d}}})

Input: 
 x  n x (p+1), the data. In the first p columns the
independent, in the last column the dependent
variable.

 h  scalar or p x 1 vector, bandwidth. If not
given, 20% of the volume of x[,1:p] is used.

 K  string, kernel function on [1,1]^p. If not given,
the product Quartic kernel "qua" is used.

 d  scalar, discretization binwidth. d[i] must be
smaller than h[i]. If not given, the minimum of h/3
and (max(x)min(x))'/r, with r=100 for p=1, and
r=(1000^(1/p)) for p>1 is used.

Output: 
 mh  m x (p+1) matrix, the first p columns constitute
a grid and the last column contains the regression
estimate on that grid. 