nlsr.control function

nlsr.control

Set and provide defaults of controls for package nlsr

nlsr.control(control)

Arguments

• control: A list of controls. If missing, the defaults are provided. See below. If a named control is provided, e.g., via a call ctrllist<- nlsr.control(japprox="jand"), then that value is substituted for the default of the control in the FULL list of controls that is returned.

  NOTE: at 2022-6-17 there is NO CHECK FOR VALIDITY

  The set of possible controls to set is as follows, and is returned.

Returns

• femax: INTEGER limit on the number of evaluations of residual function Default 10000.

• japprox: CHARACTER name of the Jacobian approximation to use Default NULL, since we try to use analytic gradient

• jemax: INTEGER limit on the number of evaluations of the Jacobian Default 5000

• lamda: REAL initial value of the Marquardt parameter Default 0.0001 Note: mis-spelling as in JNMWS, kept as historical serendipity.

• lamdec: REAL multiplier used to REDUCE lambda (0 < lamdec < laminc) Default 4, so lamda <- lamda * (lamdec/laminc)

• laminc: REAL multiplier to INCREASE lambda (1 < laminc

  Default 10

• nbtlim: if stepredn > 0, then maximum number of backtrack loops (in addition to default evaluation); Default 6

• ndstep: REAL stepsize for numerical Jacobian approximation Default 1e-7

• offset: REAL A value used to test for numerical equality, i.e. a and b are taken equal if (a + offset) == (b + offset)

  Default 100.

• phi: REAL Factor used to add unit Marquardt stabilization matrix in Nash modification of LM method. Default 1

• prtlvl: INTEGER The higher the value, the more intermediate output is provided. Default 0

• psi: REAL Factor used to add scaled Marquardt stabilization matrix in Nash modification of LM method. Default 0

• rofftest: LOGICAL If TRUE, perform (safeguarded) relative offset convergence test Default TRUE

• scaleOffset: a positive constant to be added to the denominator sum-of-squares in the relative offset convergence criteria. Default 0

• smallsstest: LOGICAL. If TRUE tests sum of squares and terminates if very small. Default TRUE

• stepredn: REAL Factor used to reduce the stepsize in a Gauss-Newton algorithm (Hartley's method). 0 means NO backtrack. Default 0

• watch: LOGICAL to provide a pause at the end of each iteration for user to monitor progress. Default FALSE

Author(s)

J C Nash 2014-7-16 revised 2022-11-22 nashjc at uottawa.ca