MariaGorinova
summary
Claims that Stan's block syntax is unnecessary and implements a prototype slick Stan compiler that shreds blockless programs back into Stan, demonstrating actual code transformation. Uses information-flow level inference to assign variables to data/model/generated-quantity levels and a performance ordering to pick the most efficient assignment. Implements static elaboration via unrolling of user-defined functions so local parameters work, reports it runs and shreds to Stan but currently lacks vectorized user functions.
AI-written summary · about
chapters (7)
- 0:00Start
- 1:34What is the problem?
- 7:43Information Flow in Stan
- 8:16Key idea
- 9:31Performance ordering
- 12:14What comes next?
- 16:02Neal's Funnel
related talks
33mScala DSLs and Probabilistic Programming
28mStan: why does it exist? when is it useful? why do I use it?
31mProbabilistic Programming: The What, Why and How
27mParsing, Compiling, and Static Metaprogramming
28mFunctional Sparse Tensor Compilation
41mIdealized Commit Logs: Code Simplification via Program Slicing
24mSoss - Lightweight Probabilistic Programming with No Performance Ceiling
01h 01mProgram Obesity: A Code Health Epidemic