Re: [ojAlgo-user] MatrixStore to BasicMatrix (or Cholesky factorization -> BasicMatrix)

[Anders P. <an...@op...>]

On 2 jul 2012, at 22:46, Chris Lucas wrote:

> Again, thanks for such a quick reply.
> 
>> That link doesn't work...
> 
> Perhaps it was down temporarily or there's some sort of IP-based
> restriction?  I was able to download the chapter pdf just now.

Yes, some temporary problem. I've got that pdf now. Interesting subject, but a little too much for me to read right now.


>> Did you find the org.ojalgo.random.process package? Cannot that be adapted or extended to do what you want?
> 
> I did find it, but sadly Gaussian processes are somewhat ill-suited to
> the current interface, as I understand it: in a GP, all y are jointly
> normal, where y_{t+1} depends on y_{1...t} via a covariance matrix
> determined by \mathbf{x}_{1...t+1}. In contrast, it seems that the
> interface is best suited for processes where y_{t+1} is independent of
> y_{1...t-1} conditional on y_t, and there's no role for x.


Essentially the current code was written to do monte carlo simulations using (multi-dimensional) geometric brownian motion - not regression.

The RandomProcess interface does not dictate if the process maintains state or not, nor what that state would depend on. That's entirely up to the implementations. The Process1D can be a multi-dimensional version of any RandomProcess implementation. Gaussian1D is a "multivariate Gaussian distribution".

GP is new to me. In-between having RandomProcess and Gaussian1D I feel there can't be that much missing...


/Anders