Resets a model file to the zero classifier, and changes the values of
parameters--use gtsvm_optimize to actually find an optimal classifier. The
regularization parameter is that of the C-formulation, while the kernel
parameter must be one of "gaussian", "polynomial" or "sigmoid", for which the
kernel functions are:
	gaussian    ==>  K( x, y ) = exp( -p1 * || x - y ||^2 )
	polynomial  ==>  K( x, y ) = ( p1 * <x,y> + p2 )^p3
	sigmoid     ==>  K( x, y ) = tanh( p1 * <x,y> + p2 )
Here, "p1", "p2" and "p3" are the values given for parameter1, parameter2 and
parameter3, respectively. The biased parameter selects whether the optimization
problem should include an unregularized bias.

Allowed options:
  -h [ --help ]               display this help
  -i [ --input ] arg          input model file
  -o [ --output ] arg         output model file (may be same as input)
  -C [ --regularization ] arg regularization parameter
  -k [ --kernel ] arg         kernel
  -1 [ --parameter1 ] arg     first kernel parameter
  -2 [ --parameter2 ] arg     second kernel parameter
  -3 [ --parameter3 ] arg     third kernel parameter
  -b [ --biased ] arg (=0)    include an unregularized bias?