You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
A few enhancement requests/ideas for blang. These arose from some work I have been doing on IID collections and IID distributions. Conceptually, these collections and distributions are meant to encapsulate for loops in, say, JAGS or Stan. These loops allow the user to capture plates used in the graphical model. That is, many observations are distributed according to the same distribution and share parameters.
The current IID package is a bit clumsy due to some limitations in the interface setup. Two interface changes could prove useful:
Another layer of abstraction above UnivariateRealDistribution to say Distribution. This would include multivariate distributions, or distributions of different datatype (eg. integers). The IID package would then be written abstractly for Distribution;
Distribution would extend GenerativeFactor instead of Factor. This is particularly useful in the case of IID collections (or just independent collections) when the user wants to forward simulate from the model for a number of points.
The text was updated successfully, but these errors were encountered:
A few enhancement requests/ideas for blang. These arose from some work I have been doing on IID collections and IID distributions. Conceptually, these collections and distributions are meant to encapsulate
for loops
in, say, JAGS or Stan. These loops allow the user to capture plates used in the graphical model. That is, many observations are distributed according to the same distribution and share parameters.For an example of IID collections see blang_examples/src/main/java/iid, and for one in action see the challenger model.
The current IID package is a bit clumsy due to some limitations in the interface setup. Two interface changes could prove useful:
UnivariateRealDistribution
to sayDistribution
. This would include multivariate distributions, or distributions of different datatype (eg. integers). The IID package would then be written abstractly forDistribution
;Distribution
would extendGenerativeFactor
instead ofFactor
. This is particularly useful in the case of IID collections (or just independent collections) when the user wants to forward simulate from the model for a number of points.The text was updated successfully, but these errors were encountered: