问题描述
我正在尝试将Beta几何回归模型拟合到我的数据中， 我尝试了以下代码：
fitb <- vglm(monch ~ AGW1+WEDU+PLACE+REGION1+WORK+MRS+PARITY,betageometric,data = data,trace = TRUE)
结果是：
Applying Greenstadt modification to 1299 matrices
VGLM linear loop 1 : loglikelihood = -3166.83441
VGLM linear loop 2 : loglikelihood = -3051.30638
Applying Greenstadt modification to 17 matrices
VGLM linear loop 3 : loglikelihood = -3096.65251
Taking a modified step..
VGLM linear loop 3 : loglikelihood = -3040.66509
Applying Greenstadt modification to 4 matrices
VGLM linear loop 4 : loglikelihood = -3028.5109
Applying Greenstadt modification to 1 matrix
VGLM linear loop 5 : loglikelihood = NaN
Taking a modified step....................
警告消息：
1: In checkwz(wz,M = M,trace = trace,wzepsilon = control$wzepsilon) :
3 diagonal elements of the working weights variable 'wz' have been replaced by 1.819e-12
2: In checkwz(wz,wzepsilon = control$wzepsilon) :
1 diagonal elements of the working weights variable 'wz' have been replaced by 1.819e-12
3: In vchol(wz,n = n,silent = !trace) :
weight matrices 421,635,759,824 not positive-definite
4: In checkwz(wz,wzepsilon = control$wzepsilon) :
1958 diagonal elements of the working weights variable 'wz' have been replaced by 1.819e-12
5: In vchol(wz,silent = !trace) :
weight matrix 635 not positive-definite
6: In vglm.fitter(x = x,y = y,w = w,offset = offset,Xm2 = Xm2,:
iterations terminated because half-step sizes are very small
7: In vglm.fitter(x = x,:
some quantities such as z,residuals,SEs may be inaccurate due to convergence at a half-step
所以问题是，在中间收敛的情况下，我一次删除并添加了一些变量，但是问题仍然存在。
我该如何解决这类问题？
解决方法
暂无找到可以解决该程序问题的有效方法，小编努力寻找整理中！
如果你已经找到好的解决方法，欢迎将解决方案带上本链接一起发送给小编。
小编邮箱:dio#foxmail.com (将#修改为@）