Problems with fmincon when trying to compute transition probabilities
in [Matlab]
|
From: Tara Reale on 30 Jun 2010 12:28 Hi all, I keep encountering a problem with fmincon when I use it to solve Markov transition probability matrices. I have no inequality constraints in my program, just boxed constraints, namely that the resulting answers are probabilities, therefore there is an upper bound of 1 and a lower bound of 0 on my variables. The problem isn't large in size and there are only 6 values to be optimized but I keep finding that the function is giving me ridiculous values and clearly not suitable for my problem. In particular it is extremely dependant on teh initail guesses - sometimes a variation of 200% can be encountered... Basically I am wondering if anybody has used fmincon when dealing with transition probability matrices before or if maybe my problem is as simple as fmincon cannot be used when you ahve box constraints alone? If anybody could offer me any advice on the above issues I would greatly appreciate it, Thanks, Tara
From: Alan Weiss on 30 Jun 2010 13:25 I'm not sure what you are trying to do. Are you trying to find a local minimum to some function that depends on the calculated Markov transition probabilities? Are you trying to find a global minimum of such a function? Or are you doing something else? fmincon is a tool that finds local minima. It can indeed be sensitive to initial conditions. For a discussion, see http://www.mathworks.com/access/helpdesk/help/toolbox/optim/ug/br44i40.html#brhkghv-65 If you want a global minimum, you might do better with tools from the Global Optimization Toolbox, such as GlobalSearch or MultiStart. For more information, see http://www.mathworks.com/access/helpdesk/help/toolbox/gads/bsa_e70.html#bsbjh03 Alan Weiss MATLAB mathematical toolbox documentation On 6/30/2010 12:28 PM, Tara Reale wrote: > Hi all, > > I keep encountering a problem with fmincon when I use it to solve Markov > transition probability matrices. I have no inequality constraints in my > program, just boxed constraints, namely that the resulting answers are > probabilities, therefore there is an upper bound of 1 and a lower bound > of 0 on my variables. The problem isn't large in size and there are only > 6 values to be optimized but I keep finding that the function is giving > me ridiculous values and clearly not suitable for my problem. In > particular it is extremely dependant on teh initail guesses - sometimes > a variation of 200% can be encountered... > > Basically I am wondering if anybody has used fmincon when dealing with > transition probability matrices before or if maybe my problem is as > simple as fmincon cannot be used when you ahve box constraints alone? > > If anybody could offer me any advice on the above issues I would greatly > appreciate it, > > Thanks, > > Tara
|
Pages: 1 Prev: can u show me how Next: UITree Refresh without collapsing |