JuliaBUGS Meta Issue

[sunxd3]

One of the fundamental difference between JuliaBUGS and DynamicPPL is that JuliaBUGS' simpler syntax makes it amendable to more precise analysis. In fact, JuliaBUGS compiles a probabilistic programming language to graphical models.

Code Quality

Goal: Ensure better testing and documentation.
Tasks:

• Integrate and test functionality with other packages.
  Refer to: Issue #245 (this is not a Turing 1.0 task, but rather a task part of the JuliaBUGS research project)

---

Integration with Turing

Goal: Enhance integration with the Turing framework.
Tasks:

• Address compatibility issues with Turing.
  Refer to: Issue #2425

---

Performance Improvements

• Parallel Computing Potentials
  Explore parallel computing capabilities to enhance performance.
  Refer to: Issue #209 (again, not a Turing 1.0 task, but rather a future direction for JuliaBUGS)
• Gibbs Sampling
  Ensure correctness through more rigorous testing and improve performance.

[yebai]

This needs more work. The above todo items are vague and too ambitious.

The core goals are to improve the performance of JuliaBUGS for log density and gradient evaluation and provide a to_distribution / to_sampleable interface so they can be used as submodels / customised distributions in Turing.

[yebai]

Closed in favour of #2425