NOTES ON THE VERSION OF 1999-12-06 This is a fairly minor release, with a few bug fixes, a couple of which might be important, some new experimental facilities that aren't of general interest (at least not yet), and a few added conveniences and other minor updates. Changes in this version: 1) Several new ways of computing trajectories have been implemented, in addition to the old leapfrog method. See mc-spec.doc for details. The new methods may be better for some applications, but they aren't uniformly better than the leapfrog method, and they are not recommended for general use. 2) New "spiral" and "double-spiral" operations were added. See mc-spec.doc for details. These are experimental. Quantities m0, m1, m2, m3, m4, and m5 were added to support this feature. See mc-quantities.doc for details. 3) The mc-spec program now lets you see all the specifications in the log file, at any iteration, with "mc-spec log-file all". See mc-spec.doc for details. 4) New "multiply-momentum" and "set-momentum" operations were added, for testing and research purposes. See mc-spec.doc for details. These are probably of no interest to most users. 5) An install-arch shell file for conveniently maintaining separate versions of the compiled programs for different machine architectures has been added. 6) The documentation has been updated and corrected in various fairly minor ways. 7) Compute times quoted for the demos are now for a Linux system with a 550MHz Pentium III processor. Bug fixes. 1) A bug was fixed that caused derivatives to sometimes be calculated incorrectly when the "omit" option was used with Gaussian process models. The effect was to increase the rejection rate for "hybrid" operations, but the final answer (if convergence was reached) would still be correct. 2) A bug was fixed that caused the "first:last" option of the met-1 operation to be ignored. 3) A bug was fixed that sometime caused data-spec to report spurious errors when targets were log-transformed. 4) A bug was fixed that caused problems with automatic scaling of target variables when they were all the same (ie, had standard deviation of 0). 5) The xxx-stepsizes and xxx-grad-test programs were changed to not try to open the log file for writing (which might not be allowed), seeing as they only need to read it. Known bugs and other deficiencies. 1) The facility for plotting quantities using "plot" operations in xxx-mc doesn't always work for the first run of xxx-mc (before any iterations exist in the log file). A work-around is to do a run of xxx-mc to produce just one iteration before attempting a run of xxx-mc that does any "plot" operations. 2) The CPU time features (eg, the "k" quantity) will not work correctly if a single iteration takes more than about 71 minutes. 3) The latent value update operations for Gaussian processes may recompute the inverse covariance matrix even when an up-to-date version was computed for the previous Monte Carlo operation. 4) Covariance matrices are stored in full, even though they are symmetric, which sometimes costs a factor of two in memory usage. 5) The "p" option of net-pred isn't allowed for survival models with piecewise-constant hazard. There's no big reason for this - I just haven't gotten around to it.