mcmc_heatmaps

MCMC mean graphs

Module

mcmc_heatmaps

Description

For Bayesian inference it is custom to use MCMC methods to simulate a Markov chain of graphs \(\{G^l\}_{l=0}^\infty\) having the graph posterior as stationary distribution. Suppose we have a realisation of length \(M + 1\) of such chain, then the posterior edge probability of an edge e is estimated by \(\frac{1}{M+1-b} \sum_{l=b}^{M} \mathbf{1}_{e}(e^l)\), where the first \(b\) samples are disregarded as a burn-in period.

This module has a list of objects, where each object has

  • Fields

    • burn_in percent [0, 1] to burn of the number of samples.

The estimated probabilities are plotted in heatmaps using seaborn which are saved in results/mcmc_heatmaps/ and copied to results/output/mcmc_heatmaps/ for easy reference.

The Alarm network

Fig. 45 Mean graph estimate of the Alarm network using order MCMC with startspace from iterative MCMC

Example

[
  {
    "id": "omcmc_itsample-bge",
    "burn_in": 0,
    "active": true
  }
]