GS (bnlearn)

Module name	bnlearn_gs
Package	bnlearn
Version	4.8.3
Language	R
Docs	here
Paper	Margaritis[1]
Graph type	DAG
MCMC	No
Edge constraints	Yes
Data type	C, D, M
Data missingness
Intervention type
Docker	bpimages/bnlearn:4.8.3

Grow-shrink

The grow-shrink (GS) algorithm is based on the Markov blanket of the nodes in a DAG. For a specific node, the Markov blanket it the set of nodes which conditioning upon renders it conditionally independent from all other variables Margaritis[1]. It is a constraint-based method which estimates the Markov blanket of a node in a two-stage forward-backward proce- dure using conditional independence tests. The Markov blankets are used to first estimate an undirected graph and then estimate a DAG in a four-step procedure.

Some fields described

edgeConstraints Name of the JSON file containing background knowledge

Example JSON

[
  {
    "id": "gs-mi",
    "alpha": [
      0.01,
      0.05,
      0.1,
      0.2
    ],
    "test": "mi",
    "B": null,
    "maxsx": null,
    "debug": false,
    "undirected": false,
    "timeout": null,
    "edgeConstraints": "edgeConstraints.json"
  },
  {
    "id": "gs-zf",
    "alpha": [
      0.01,
      0.05
    ],
    "test": "zf",
    "B": null,
    "maxsx": null,
    "debug": false,
    "undirected": false,
    "timeout": null,
    "edgeConstraints": "edgeConstraints.json"
  }
]