Re: how to project multivariate Gaussian to a lower dimension?

hi, there
If I have a n-dimensional Gaussian and I want to
project it to n-1
dimension (and I know which variable needs to drop
off), do I just
need to delete the row and column where the drop off
variable stands
in the n by n covariant matrix? (of cause drop that
from the mean
too). Thanks for your help.
zl2k



Yes. What you've described gives you the (n-1)-dimensional normal marginal distribution. If, however, you were interested in the conditional distribution of X1, ..., Xn-1 given Xn = xn, that would be a different solution.

Jack
.

Relevant Pages
Quantcast