r*******y
发帖数: 1081 | 1 执行完毕以后总是出来这几句：
Optimization completed because the size of the gradient at the initial point
is less than the default value of the function tolerance.
有没有办法不让这几句出现？谢谢。难道还必须去修改fminunc.m吗？
谢谢 |
|
m*********s
发帖数: 368 | 2 help 一下这个函数
一般的话, 这些话都是一个返回值
表示这个函数的执行结果
用这样的格式调用
[a,b] = fminunc(...)
b就把这些话个吃掉而不显示出来了
point |
|
l*****a
发帖数: 119 | 3 fminunc是判断FOC的 初始点的导数太接近0了 所以算法觉得初始点就是最优解 停
止了 |
|
a***m
发帖数: 74 | 4 Greetings!
I used fminunc to estimate parameters for a GARCH model,where innovation
process is a mixture distribution. The estimate give back complex parameters
and maximum feval have reached. But I retrieved real part and transformed
the parameter before calculating each likelihood function in the iteration.
I don't understand why matlab kept give me complex estimations. This occurs
to fminsearch as well.
Any suggestions are welcome.Many thanks! |
|
|
|
a***m
发帖数: 74 | 7 Greetings!
My function evaluation sometimes generate complex numbers when I use fmincon
, fminsearch, and fminunc. The optimizations stop or won't converge. Why
this could happen? Can I dump the imaginary part to continue the routine?
Any experience to share?
Many Thanks. |
|
p***c
发帖数: 2403 | 8 自定义了一个函数如 function [out1,out2]=xxx(in1,in2)
我想求out2的极值, 如果只有一个输出的时候，可以用fminunc
现在有两个的时候，怎么用呢？
谢谢 |
|
M********g
发帖数: 183 | 9 没用过fminunc，不过你再定义一个函数调用xxx只返回out2不行吗？
这个难道还要找什么elegant的方法？ |
|
s****h
发帖数: 3979 | 10 Suppose I have a VAR model:
X_t = A * X_(t-1) + B * Y_t,
And I need to estimate the coefficient matrix A & B?
What's the best way to do if A is 3X3 matrix, and B is a 3X5 matrix?
I have only used gradient based method for simpler problems before. For as
many as 3X8 = 24 parameters, should the gradient method works well if I just
use a matlab unction fminunc?
Or, should I estimate A & B separately?
Thanks for your input. |
|
|
r*******y
发帖数: 1081 | 12 那些不是 warning.
warning off all了也没用 |
|
l********a
发帖数: 1154 | 13 Optimization completed because the size of the gradient at the initial point
is less than the default value of the function tolerance.
优化结束,因为初始点的梯度尺寸小于函数容限的默认值
结束啦 |
|
s****h
发帖数: 3979 | 14 【 以下文字转载自 Stock 讨论区 】
发信人: squash (鼻涕泡泡不是传说), 信区: Stock
标 题: a question about VAR
发信站: BBS 未名空间站 (Mon Nov 19 17:05:47 2007)
Suppose I have a VAR model:
X_t = A * X_(t-1) + B * Y_t,
And I need to estimate the coefficient matrix A & B?
What's the best way to do if A is 3X3 matrix, and B is a 3X5 matrix?
I have only used gradient based method for simpler problems before. For as
many as 3X8 = 24 parameters, should the gradient method works well if I just
use a matlab unction fminunc?
Or, should I e |
|
l*****i
发帖数: 3929 | 15 Check your formulation first. Doesn't GARCH have constraints (sum of 3 param
eters = 1 or something like that)? -- I assume you use GARCH(1,1).
parameters
.
occurs |
|
a***m
发帖数: 74 | 16 I copied likelihood function from Bollerslev paper. It used gamma and log
functions. I scaled parameters so that they were bounded. Grad and hessian
was provided by the function. So I wonder about the origin of complex
estimation.Any hints? Many thanks! |
|
f********y
发帖数: 278 | 17 谢谢 DaShagen,这一步我已经解决了。
假定b=[1 x;3 y;x y]， 下一步的问题是求min max{(1-x)^2,(3-y)^2,(x-y)^2}.
如果我直接设置方程，f=max{(1-x)^2,(3-y)^2,(x-y)^2}
然后fminunc(f,x,y,z)这样好象不行，有没有另外的方法？ |
|
