最近在跑一个很大的NLLS, 在用ml maxmize 回归的时候出现了一下问题。
首先,用technique (bfgs)的时候碰到了flat region 报警,只好转用nr 和 bhhh交替, 跑了一会以后,出现了:
Iteration 93: log likelihood = 20955.101 (not concave)
vtol = 3.766e-12
Iteration 94: log likelihood = 20955.101 (not concave)
ptol = 8.464e-07
Iteration 95: log likelihood = 20955.101 (not concave)
vtol = 5.855e-12
Iteration 96: log likelihood = 20955.101 (not concave)
ptol = 9.642e-07
Iteration 97: log likelihood = 20955.101 (not concave)
vtol = 4.223e-12
Iteration 98: log likelihood = 20955.101 (not concave)
ptol = 7.271e-07
Iteration 99: log likelihood = 20955.101 (not concave)
ptol = 8.260e-07
(switching technique to bhhh)
Iteration 100: log likelihood = 20955.101 (not concave)
vtol = 4.304e-12
Iteration 101: log likelihood = 20955.101 (not concave)
ptol = 3.554e-07
Iteration 102: log likelihood = 20955.101 (not concave)
ptol = 6.935e-07
Iteration 103: log likelihood = 20955.101 (not concave)
ptol = 4.119e-07
Iteration 104: log likelihood = 20955.101 (not concave)
ptol = 2.589e-07
Iteration 105: log likelihood = 20955.101 (not concave)
ptol = 4.613e-07
Iteration 106: log likelihood = 20955.101 (not concave)
ptol = 4.523e-07
Iteration 107: log likelihood = 20955.101 (not concave)
无论我是用不用 ml maximize, difficult 还是只用 ml maximize 结果都一样, 甚至设置了 tol(1.0000e-03) ltolerance(1.0000e-03) nrtolerance(1.0000e-03) 结果也没有变化。
不知道该怎么办才好…………