Allows the user to set parameters affecting the estimation of the discrete GLMs implemented in cubinf. Most control parameters are parameters of the ROBETH subroutine GYMAIN (Marazzi, 1993).
cubinf.control(tlo = 0.001, tua = 1e-06, mxx = 30, mxt = 10, mxf = 10, ntm = 0, gma = 1,
iug = 1, ipo = 1, ilg = 2, icn = 1, icv = 1, ufact = 0, cpar = 1.5,
null.dev=TRUE, ...)List of control parameters.
Relative precision for the convergence criterion of the main algorithm (GYMAIN) called by cubinf. The relative precision for the convergence criterion in the lower level steps (theta-step, A-step and c-step) is '10*tlo'.
Tolerance used for the determination of the pseudo-rank.
Maximum number of cycles for the main algorithm.
Maximum number of iterations for the theta-step.
Maximum number of iterations for the A-step.
Parameter to control iteration monitoring. When the number of iterations in the theta-step reaches a multiple of 'ntm', the current parameter values as well as the corresponding value of the objective function are printed.
Relaxation factor for the theta-step.
Parameter for the choice of the u-function in the A-step. See Marazzi, 1993, for details.
Parameter for the choice of the steplength algorithm in the theta-step. If 'ipo=1', a quadratic comparison function is minimized. If 'ipo=2', the Goldstein-Armijo step length algorithm is used.
Parameter for the choice of the algorithm in the c-step. If 'ilg=1', the H-algorithm is used. If 'ilg=2', the W-algorithm is used.
Parameter for the choice of the convergence criterion for the theta-step and the main algorithm. If 'icn=1', convergence is assumed when the change in each coefficient is less than the tolerance ('10*tlo') times an estimate of the coefficient variance. See Marazzi (1993, p. 281), for the other options ('icn=2' and 'icn=3').
Parameter for the choice of the convergence criterion for the A-step. If 'icv=1', convergence is assumed when the norm of the difference between two consecutive values of A is less than the tolerance (10*tol). See Marazzi (1993, p.288 and p. 301), for another option ('icv=2').
The tuning constant b is set equal to ufact*sqrt(p), where p is the dimension of the observation vectors. The default value of b is 1.1*sqrt(p); this value is used when 'ufact=0' on input.
Parameter used in determining an initial value of theta (standard Mallows estimate, see Marazzi, 1993, p281).
If 'null.dev=TRUE', the null deviance is computed. The null deviance is the deviance of the model with no predictors.
Further named control arguments as singular.ok or qr.out used in the case where the x matrix is singular
Marazzi, A. (1993). Algorithms, Routines, and S-functions for robust Statistics. Chapman and Hall, New York.
cubinf
#To compute the classical estimates using cubinf, set:
control <- cubinf.control(ufact=300)
Run the code above in your browser using DataLab