I placed a new version, 4.21, of the Bayesian Analysis software and interface online late yesterday afternoon. The most pressing thing about the update was to fix a model selection problem we were having. Everything was working fine in the enter ASCII model selection program. However, the analysis could not be reproduced using the image model selection program. The problem turned out to be related to the prior probability assigned for marginalized amplitudes. Nonmarginalized amplitude model selection was working fine, but when marginalization was used, the prior was over penalized the amplitudes and consequently pulling them much to strongly toward zero. The problem has been fix and it affected about 5 programs.
The previous update was to add a new version of the user manual. This version covers release 4.10 of the Bayesian Analaysis software. This new version features an extensively updated Analyze Image Pixels package. The primary modification to this package was to modify the Markov chain Monte Carlo routine to make it as similiar to the Enter Ascii model package. In addition, Analyze Image Pixels no longer outputs images that are all zero. Finally, all packages that output probability density functions, now also output a scatter plot of the posterior probability vs the paramter values. This plot turns out to be very useful in understanding how well a parameter has been estimated. For more on this plot see the updated user manual.
This site is being maintained by:
Larry Bretthorst
Dept. Of Chemistry and Radiology
Washington University
St. Louis MO 63130
Phone: 314 362-9994
All questions concerning this site should be directed to him.