Infer.NET User Guide

• Introduction

  • What is Probabilistic Programming?
  • What is Infer.NET?
  • A simple example
  • How Infer.NET works
  • Frequently Asked Questions
  • Resources and References
  • Infer.NET 101 paper, a detailed sample-based introduction to programming with Infer.NET.

• Building probabilistic models

  • Conceptual background: Model-Based Machine Learning Book
  • The Infer.NET modelling API
    • Creating variables
      • Variable types and their distributions
      • Vector and Matrix types
      • Applying functions and operators to variables
    • Attaching constraints to variables
    • Working with arrays and ranges
      • Arrays and ranges
      • ForEach blocks
      • Markov chains and grids
      • Jagged arrays
      • Indexing arrays by observed variables
      • Cloning ranges
    • Branching on variables to create mixture models
      • The importance of using SetTo
    • Repeat blocks
    • Computing model evidence for model selection
    • Increment log density
  • Factors and constraints
    • Boolean factors
    • Integer and enum factors
    • Double factors
    • String and char factors
    • Array and list factors
    • Vector and matrix factors
    • Generic factors
    • Constraints
    • Undirected factors
    • List of factors and constraints
  • Advanced model building
    • Adding attributes to your model
    • The Model Specification Language
    • Transform browser

• Running inference on your model

  • Running inference
    • Inference engine settings
    • Working with different inference algorithms
      • Expectation Propagation
      • Variational Message Passing
      • Gibbs sampling
    • Monitoring the progress of inference
    • Quality bands
  • Controlling how inference is performed
    • Customising the algorithm initialisation
    • Using a precompiled inference algorithm
    • Structure of generated inference code
  • Debugging inference
  • Improving the speed and accuracy of inference
    • Using sparse messages
  • Sharing variables between models
    • Shared variable arrays
    • Defining shared variables within a model
    • Jagged shared variable arrays
    • Computing evidence for models with shared variables
    • Using shared variables to support hybrid algorithms
  • Online learning
  • Distributed inference
  • Computing derivatives of functions

• Extending Infer.NET

  • Infer.NET component architecture
  • Adding a new distribution type
  • Adding a new factor and its message operators
  • Adding a new constraint
  • Modifying the operator search path
  • Automatically computing EP messages: KJIT

• Calling Infer.NET from other languages

  • F#
    • Variable Arrays in F#
    • Inference in F#
    • Operator Overloading in F#
    • Imperative Statement Blocks in F#
  • C++/CLI
  • Mono

• Learners

  • Bayes Point Machine classifiers
  • Matchbox recommender

• Tutorials and examples

  • The Examples Browser
  • Tutorials: Two coins, Clinical trial, Mixture of Gaussians, more…
  • String tutorials: Hello, Strings!, StringFormat operation, and Motif finder.
  • Examples: Latent Dirichlet Allocation, Recommender System, Bayesian PCA, Discrete Bayesian network, more…
  • How-to guides: How to handle missing data, How to build scalable applications, more…
• Infer.NET development
• Code documentation
• Release change history