 |
 |
I received several requests to summarize, so this is an issue that others are experiencing. Thanks to Nicky Best and Bob O'Hara for replies, which I have combined below. _______________________ 1) the summary monitor is really your best option here. The means and sd are exact, it is just the posterior quantiles that are approximate. The best thing to do is to set summary monitors on all the p and rho parameters, and then set samples monitors as well for a small subset of them, and compared the exact and approximate quantiles to check for how well the summary monitor is doing. 2) Increasing thinning on the *update* tool will also help save memory 3) you could try to run the model for, say 7500 iterations, then stop it, save the variables and then clear them, and then start saving them again. You'll have to stick them back together again afterwards, but that can be done in R, for example. If that doesn't work, then you could always run it for 7500 iterations, save state, close BUGS, and then start it again, and run it from the saved states. ______________________ Original question: I am running CAR models in GeoBUGS on maps with fairly large numbers of cells (5000-10000). I have found that setting the program to sample both p and rho (via the inference/sample monitor tool) causes "illegal memory write" at the same iteration (~8000 iterations), at which point the program is using ~ 1.5GB of memory. (This occurs with both WINBUGS 1.4.1 and OpenBUGS 2.2.0). If I set it to only sample either p or rho, it will run for twice as many iterations without the error. I tried switching to WINXP's 64-bit version on a computer with 4GB of RAM. WINBUGS runs well on the 64-bit system, but still gives illegal memory write at the same point (~1.5GB memory usage). Theoretically, a 32-bit application can access up to 2.7GB of RAM on XP64-bit, so I am wondering if: 1) Is there an option for increasing the maximum RAM WinBUGS can access, similar to how maximum object size can be set via options in R? 2) If not, what other workarounds might be possible? I see that the Summary Monitor Tool might provide one option, but the manual says to use with caution as statistics are approximate. Would increasing the thinning value in the sample monitor tool or the update tool help? -- Carlos Carroll, Ph.D. Klamath Center for Conservation Research PO Box 104 Orleans, CA 95556 ------------------------------------------------------------------- This list is for discussion of modelling issues and the BUGS software. For help with crashes and error messages, first mail [log in to unmask] To mail the BUGS list, mail to [log in to unmask] Before mailing, please check the archive at www.jiscmail.ac.uk/lists/bugs.html Please do not mail attachments to the list. To leave the BUGS list, send LEAVE BUGS to [log in to unmask] If this fails, mail [log in to unmask], NOT the whole list