 Usage:  mh = lregestp(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. 
                         
--------------------------------------------------------------
(C) MD*TECH Method and Data Technologies, 17.8.2000
