commons-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Gilles Sadowski <gil...@harfang.homelinux.org>
Subject Re: [Math] "LUDecomposition" in "AbstractLeastSquaresOptimizer"
Date Fri, 09 Sep 2011 14:16:42 GMT
Hi.

> Is the test checked in? I am interested in looking at it.

Thanks for the interest, but it is a unit test for the software which I
work on. I'm afraid that a discussion about this will be fairly OT here.

Basically, I need to compute the covariance matrix, but the model function
is the result of a simulation: thus, currently I compute the Jacobian
numerically (using a trivial approximation of the derivatives).
One of the issues is how to make sure that the resulting matrix is
meaningful.
Another issue, further up, is how to make sure that the Jacobian is
meaningful. We also want to test optimizers that use derivatives (i.e.
"LevenbergMarquardtOptimizer") but, since the parameters are subject to
(simple) boundary conditions (e.g. orbit eccentricity, temperature, ...),
I use a "logit" transform when passing the values to the optimizer. This
leads to a (transformed) Jacobian whose columns have very different orders
of magnitudes. So I wonder whether the "transform" approach is usable at
all...


Best,
Gilles

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
For additional commands, e-mail: dev-help@commons.apache.org


Mime
View raw message