GP-DGEN:  Generate values for target variables given latent values

Gp-dgen randomly samples from the distribution for targets values in
training cases given the latent values for these cases.  This may be
of interest in itself, and can also be used as part of Geweke's
procedure for testing the correctness of Markov chain sampling methods
(see xxx-mc-test.doc).  Gp-dgen cannot be used if latent values are
not stored in the log file.

Note that the gp-eval program does related, but more specific, things.

Usage:

    gp-dgen log-file [ index ] / targets-file

Latent values are read from the record in the log file with the
specified index.  Target values are then sampled from the distribution
for targets in each training case defined by these latent values, and
written to the targets file, whose previous contents are overwritten.
For models of real data, the noise standard deviation hyperparameter
is also taken from the specified index in the log file.  The random
number seed used is that saved for the index specified.

The default if no index is specified is the last index in the log
file.  When the last index is used by default, the random number state
is appended to the file after being updated during generation of the
targets.

            Copyright (c) 1995-2004 by Radford M. Neal