Break Manual into Parts

[bob-carpenter]

Summary:

Split the current manual into multiple documents. This in part mirrors how we're going to split the repos into a Stan language and Stan algorithms repo.

Reference Manual

• part II. Stan Modeling Language
• part VI. Algorithms and Implementations
• Reproducibility
• ch 30. Markov Chain Monte Carlo Sampling 368
• ch 67. Reproducibility
• app A. Licensing (and dependencies)
• app C. Modeling Language Syntax
• (part of) app D. Warning and Error Messages
• app E. Deprecated Features
• app F. Mathematical Functions
• part V. Algorithms & Implementations

Function Reference

• part VI. Built-In Functions
• part VII. Discrete Distributions
• part VIII. Continuous Distributions

User's Guide

• ch 65. Model Building as Software Development
• part III. Example Models
• part IV. Programming Techniques
• app B. Stan for Users of BUGS
• ch 69. Stan Program Style Guide

Stan Software Developer's Guide

• ch 66. Software Development Lifecycle
• ch. 68. Contributed Modules

Introduction Statistical Inference

• ch 27. Bayesian Data Analysis
• ch 29. Penalized Maximum Likelihood Point Estimation
• ch 30. Bayesian Point Estimation
• ch 31. Variational Inference

Notes

• Everything not mentioned disappears (intro, warning messages [never filled out]
• Stan Software goes to Wiki
• everything presupposes the intro stat inference
  • some may beuseful in the reference manaul for precise definitions
  • some might be useful in users guide, but perhaps simpler and example-driven

Current Version:

v2.12.0

[skanskan]

I was just going to say the same, the manual may be split into several parts. Or its parts could be reordered. The Modelling language Reference should come first (though some of its content is repeated there).

[bob-carpenter]

If they're separate docs, there's no order. Aki just rearranged the manual in pull request #2102.

[bob-carpenter]

From @betanalpha in #354

Roadmap

Bayesian Inference in Practice

• Uncertainty as Probability, Bayes Theorem
  • Likelihood, Prior, Posterior
• Employing the Posterior with Expectations
  • Measures and Densities
  • Typical Set, Integrals Hard
• Computational Techniques
  • Approximating Expectations with Point Estimates
  • Markov Chain Monte Carlo
  • Ergodicity, Geometric Ergodicity
    • Convergence and Mixing
    • The Monte Carlo Central Limit Theorem
  • Convergence in Practice
    • Finding the typical set
    • Necessary conditions as proxy for sufficient conditions
    • Split Rhat
  • Analysis
    • Autocorrelations
    • Effective Sample Size
    • Monte Carlo Standard Error
  • Markov Kernels
    • Random Walk Metropolis
    • Gibbs (BUGS, JAGS)
    • Hamiltonian Monte Carlo Stan
• The Stan Modeling Language
  • Example
  • Types and Declarations
  • Expressions
  • Statements
  • Program Blocks
  • Functions
• Automatic Differentiation
  • Factoring algebra from partials with dual numbers
  • Examples
  • Implementation
• The Stan Algorithms
  • Optimization
  • BFGS, L-BFGS
  • MCMC
  • HMC

Stan in Practice

• Best Practices
  • Development
  • Style
• Optimization
  • Model Implementation
  • Rao-Blackwellization
  • Reparameterization
  • Stan Implementation
  • Linear Algebra
  • Vectorization
• Modeling Techniques
  • Dump from Stan

The Stan Reference Guide

• Modeling Language
  • BNF Syntax
  • Functions
  • Distributions
  • Continuous
  • Discrete
• Algorithms
  • Optimization
  • BFGS, L-BFGS
    • Configurations
  • MCMC
  • HMC
    • Configurations

The Stan Interfaces

• CmdStan, RStan, PyStan, MStan
  • Data Format
  • Compiling Models
  • Running Models
  • Tools
  • Quick Start Guide

The Stan Development Process

• Tools
• Issues
• Pull Requests
• Submitting a Patch

[skanskan]

Nice, though maybe you are trying to embrace too much and you should leave most of the theory to any of the already existing books, and just focus on Stan.

[bob-carpenter]

A couple of different pieces of theory are going into separate docs. We're also writing some other introductions to methodology.