Compute composite loss value
loss3(y, mu, theta, weights, cfun, family, s, delta)response variable values, 0/1 if family=2, or binomial
response prediction of y. If mu is linear predictor, use function loss2 instead
scale parameter for family=4, negative binomial
observation weights, same length as y
integer from 1-8, concave function as in ccglm_fit
integer 2, 3 or 4, convex function binomial, Poisson or negative binomial, respectively
tuning parameter of cfun. s > 0 and can be equal to 0 for cfun="tcave".
a small positive number provided by user only if cfun="gcave" and 0 < s <1
Weighted loss values
For large s values, the loss can be 0 with cfun=2,3,4, or "acave", "bcave", "ccave".
Zhu Wang (2020) Unified Robust Estimation via the COCO, arXiv e-prints, https://arxiv.org/abs/2010.02848