STK: a Small (Matlab/Octave) Toolbox for Kriging
 STK_PARAM_INIT provides a starting point for stk_param_estim

 CALL: PARAM = stk_param_init (MODEL, XI, YI)

   provides a quick and dirty estimate of the parameters of MODEL based on the
   data (XI, YI), that can be used as a starting point for stk_param_estim. It
   selects the maximizer of the ReML criterion out of a list of possible values
   given data (XI, YI). This syntax is appropriate for noiseless observations
   and for noisy observations with known noise variance (i.e., when the
   'lognoisevariance' field in MODEL is either -Inf or has a finite value).

 CALL: [PARAM, LNV] = stk_param_init (MODEL, XI, YI)

   also returns a value for the 'lognoisevariance' field. In the case of
   noiseless observations or noisy observations with known noise variance, this
   is simply the value that was provided by the user in MODEL.lognoisevariance.
   In the case where MODEL.lognoisevariance is NaN (noisy observation with
   unknown noise variance), LNV is estimated by stk_param_init together with
   PARAM.

 CALL: [PARAM, LNV] = stk_param_init (MODEL, XI, YI, BOX)

   takes into account the (hyper-rectangular) domain on which the model is
   going to be used. It is used in the heuristics that determines the list of
   parameter values mentioned above. BOX should be a 2 x DIM matrix with BOX(1,
   j) and BOX(2, j) being the lower- and upper-bound of the interval on the
   j^th coordinate, with DIM being the dimension of XI, DIM = size(XI,2). If
   provided,  If missing or empty, the BOX argument defaults to [min(XI);
   max(XI)].

 See also stk_example_kb02, stk_example_kb03, stk_example_misc03