Turing.jl

Researchers and statisticians define probabilistic models in plain Juliaspecifying prior distributions, likelihood functions, and conditional structureand Turing.jl handles inference automatically using a range of algorithms including Hamiltonian Monte Carlo (HMC/NUTS), Sequential Monte Carlo, and variational inference.

The Julia ecosystem gives Turing.jl a performance advantage over Python-based PPLs like PyMC while maintaining the expressiveness of a high-level statistical language.

Turing.jl's composable design allows complex hierarchical and mixture models to be built from simpler components, and its integration with Julia's automatic differentiation ecosystem means custom model components receive gradient-based inference automatically without manual derivative implementation.

The library supports both simple conjugate models and research-frontier applications like neural network weight uncertainty, Gaussian process regression, and state space modelsmaking it useful across a spectrum from applied statistics to ML research.

Bayesian statisticians, quantitative researchers, and ML practitioners who need principled uncertainty quantification use Turing.jl when PyMC or Stan's performance is insufficient for their model complexity or dataset size.

Julia's just-in-time compilation means Turing.jl models often run 10-100x faster than equivalent Python implementations, making it practical for models that would be computationally prohibitive in other environments.

Academic researchers in computational statistics, epidemiology, ecology, and finance publish Turing.jl as the implementation language for their Bayesian methods, establishing it as a credible reference implementation alongside Stan in the probabilistic programming literature.