GenExperimental.jl
GenExperimental.jl copied to clipboard
probcomp
Featherweight embedded probabilistic programming language and compositional inference programming library
General meta-programs for VI: - Define a subset of variables that are assumed to always be there. - Target those with the VI - "Gen provides meta-programs for black box...
To draw from the predictive, the current implementation first draws the parameters from the Normal-Inverse-Wishart posterior, and then draws the datum from the multivariate normal with the sampled parameters. It...
Can ideas from this paper be supported in Gen? Divide-and-Conquer with Sequential Monte Carlo https://arxiv.org/pdf/1406.4993.pdf
I am removing these temporarily, in order to allow the major interface update of Gen to be merged into master, without having to patch up that feature. The interface update...
First, do the version that scales O(n) per generation where n is the number of data points (resulting in an O(n^2) Gibbs sweep).
If possible, change the current syntax: ```julia @program foo(args..) begin .. body end ``` to: ```julia @probabilistic function foo(args...) .. body end ``` This syntax be more familiar to Julia...
This conflicts with the usage in https://en.wikipedia.org/wiki/Generator_(computer_programming). This confusion actually visibly arose once during the summer school.
- implement module network DAG - Bayes net of modules with dependency tracking
Making `ProgramTrace` parametric on the address type (e.g. `ProgramTrace{AddressType}`) may be a useful performance optimization in the future, because using the `Any` type for the address, as currently done, probably...
In v0.2.1, calling `generate!` always trigger a backpropagation from the score, even if that isn't necessary. I did this to simplify the code during the earlier refactoring, but it should...