PyMC’s User Guide

Release:2.1 Date:April 02, 2009

Table of contents

• 1. Introduction
  • 1.1. Purpose
  • 1.2. Features
  • 1.3. What’s new in 2.1
  • 1.4. Usage
  • 1.5. History
  • 1.6. Getting started
• 2. Installation Instructions
  • 2.1. Dependencies
  • 2.2. Installation using EasyInstall
  • 2.3. Installing from pre-built binaries
  • 2.4. Compiling the source code
  • 2.5. Development version
  • 2.6. Running the test suite
  • 2.7. Bugs and feature requests
• 3. Tutorial
  • 3.1. An example statistical model
  • 3.2. Two types of variables
  • 3.3. Parents and children
  • 3.4. Variables’ values and log-probabilities
  • 3.5. Fitting the model with MCMC
  • 3.6. Fine-tuning the MCMC algorithm
  • 3.7. Beyond the basics
• 4. Building models
  • 4.1. The Stochastic class
  • 4.2. Data
  • 4.3. The Deterministic class
  • 4.4. Containers
  • 4.5. The Potential class
  • 4.6. Graphing models
  • 4.7. Class LazyFunction and caching
• 5. Fitting Models
  • 5.1. Creating models
  • 5.2. Maximum a posteriori estimates
  • 5.3. Normal approximations
  • 5.4. Markov chain Monte Carlo: the MCMC class
  • 5.5. Step methods
  • 5.6. Gibbs step methods
  • 5.7. The Model class
  • 5.8. The Sampler class
• 6. Saving and managing sampling results
  • 6.1. Accessing Sampled Data
  • 6.2. Saving Data to Disk
  • 6.3. Loading Back a Database
  • 6.4. Backends Description
  • 6.5. Writing a New Backend
• 7. Model checking and diagnostics
  • 7.1. Convergence Diagnostics
  • 7.2. Autocorrelation Plots
  • 7.3. Goodness of Fit
• 8. Extending PyMC
  • 8.1. Nonstandard Stochastics
  • 8.2. User-defined step methods
  • 8.3. New fitting algorithms
  • 8.4. Don’t update stochastic variables’ values in-place
• 9. Probability distributions
  • 9.1. Discrete distributions
  • 9.2. Continuous distributions
  • 9.3. Multivariate discrete distributions
  • 9.4. Multivariate continuous distributions
• 10. Conclusion
  • 10.1. What’s next?
  • 10.2. How to get involved
• 11. Appendix: Markov Chain Monte Carlo
  • 11.1. Monte Carlo Methods in Bayesian Analysis
  • 11.2. Markov Chains
  • 11.3. Why MCMC Works: Reversible Markov Chains
  • 11.4. Gibbs Sampling
  • 11.5. The Metropolis-Hastings Algorithm
• 12. References

The incomplete laundry list

Todo

Need a better description of what a Potential is. Given the description of Stochastic and Deterministic we have given, its not clear where Potential fits in, as it classifies the world into 2 things – completely determined by parents and not. ..

(The original entry is located in modelbuilding.rst, line 31 and can be found here.)

Indices and tables

• Index
• Module Index
• Search Page
• API documentation