Open cboettig opened 11 years ago
set zcov=TRUE, and the *.s2 and, instead of being vectors, the outputs will be matrices containing the posterior mean covariance matrices at the X and XX locations respectively. It is, unfortunately, not possible to obtain samples from the posterior covariance matrix directly from tgp
This is now done for Figure 1b, the "goodness-of-fit", e.g.:
Could be done on Figure 1a (state-space) but less interesting, and already cluttered.
To illustrate the point that the GP posterior contains a family of possible curves, might be useful to include a few example draws from the posterior rather than just the mean path and variance.
Need to reconstruct the covariance kernel from bgp first...