Default control parameters for 'serp' fit. User-supplied control parameters could be specified in the main function.
serp.control(
maxits = 5e01,
eps = 1e-07,
maxpen = 1e07,
trace = 0L,
maxAdjIter = 5e0,
max.half.iter = 1e01,
relTol = 1e-03,
nrFold = 5e0,
cv.seed = 1e01,
grid.length = 5e01,
misclass.thresh = 5e-01,
minP = .Machine$double.eps,
...)the maximum number of Newton's iterations. Default to 100.
threshold value during optimization at which the iteration routine terminates. In other words, when the reported change in the log-likelihood goes below this threshold, convergence is achieved.
the upper end point of the interval from zero to be searched for a tuning parameter.
prints the Newton's fitting process at each iteration step.If 0 (default) no information is printed, if 1, 2 or 3 different shades of information are printed.
the maximum allowable number of Newton step adjustment to forestall an early optimization failure. Defaults to 5.
the maximum number of iteration step-halfings. Defaults to 10.
relative convergence tolerance, defaults to 1e-03. checks relative changes in the parameter estimates between Newton iterations.
the number of k-fold cross validation for the CV tuning method. Default to k = 5.
single numeric value to change the random seed in CV tuning.
the length of the discrete lambda grid for the penalty method.
to reset the classification threshold in
errorMetrics when type is 'misclass'.
A near zero minimum value the fitted probabilities are allowed to get during iteration to prevent numerical instability .
additional arguments.
a list of control parameters.
library(serp)
serp(rating ~ contact, slope = "parallel", link = "logit",
control = list(maxits = 2e01, eps=1e-05, trace = 2),
data = wine)
#> iter: max|grad| logLik
#> 1: 6.333e+00 100.12594
#> delta: -2.021e+00 6.378e-02 1.600e+00 2.803e+00 -1.149e+00
#> step: -5.746e-01 -5.746e-01 -5.746e-01 -5.746e-01 1.149e+00
#> 2: 1.232e+00 99.95626
#> delta: -2.133e+00 4.191e-02 1.711e+00 2.969e+00 -1.203e+00
#> step: 1.120e-01 2.188e-02 -1.102e-01 -1.663e-01 5.389e-02
#> 3: 4.850e-02 99.95591
#> delta: -2.139e+00 4.258e-02 1.714e+00 2.979e+00 -1.207e+00
#> step: 6.676e-03 -6.785e-04 -3.863e-03 -9.377e-03 3.824e-03
#> 4: 3.209e-04 99.95591
#> delta: -2.139e+00 4.257e-02 1.714e+00 2.979e+00 -1.207e+00
#> step: 3.566e-05 1.308e-05 2.791e-06 -2.395e-05 2.596e-05
#>
#> Successful convergence!
#> Absolute and relative convergence criteria satisfied
#>
#> call:
#> serp(formula = rating ~ contact, link = "logit", slope = "parallel",
#> data = wine, control = list(maxits = 20, eps = 1e-05, trace = 2))
#>
#> coefficients:
#> (Intercept):1 (Intercept):2 (Intercept):3 (Intercept):4 contactyes
#> -2.13933037 0.04257146 1.71448721 2.97874484 -1.20695403
#>
#> loglik: -99.95591 aic: 209.9118