MacMCMC (v1.5)

State-of-the-art Data Analysis for Mac OS X™

To see MacMCMC in action, consider this famous example from the literature (Arnold and Libby, 1949):

Carbon-14 Dating

Input Data Model

Output Report Trace Marginals

Given the MacMCMC report, any graphing software may be used to prepare a plot showing model versus data.
Note: The blue line in this plot uses mean estimates; the red line shows the prior uncertainty for parameter A.

Principal Features

General

• Complete, standalone Mac application
• 100% Bayesian inference
• 100% ensemble MCMC
• Access to low-level options
• Parallelized for maximum speed

Specifics

• Input data and model with ASCII plaintext (UTF-8)
• Live editing of model
• 27 built-in distributions
• 16 continous (incl. BivariateNormal)
• 8 discrete
• 3 (homogeneous) mixtures
• Normal and BivariateNormal
• Poisson
• Generic (user-defined) distribution
• 15 built-in functions
• [Optional] "Extras" category for ancillary quantities
• Output files for monitored variables
• Report
• MAP, Mean, Median, Mode and Gelman-Rubin statistic
• 90, 95 and 99-percent credible (HPD) intervals
• [Optional] Computation of log(marginal likelihood)
• Trace (thinned, tabular posterior)
• Plots of marginals
• Output as pdf and/or png
• [Optional] Editing of axis labels
• [Optional] Fourier smoothing of plot
• Trace comparison for selected chains
• [Optional] Special handling for built-in mixtures
• Automatic relabeling
• Output of class probabilities for datapoints
• [Optional] Goodness-of-fit plots (in most cases)
• Equations: X-Y plot
• Continuous distributions: quantile-quantile (q-q) plot
• Discrete distributions: PDF histogram
• [Optional] 95-percent posterior-predictive credible interval band
• Examples:
  • 15 ready-to-run examples supplied with package. See descriptions here.
  • A complete and detailed case study. See here.

Access

Platform Requirements

For those who prefer a simpler, frequentist application, Regress+ 2.8 is still available here.

Feedback