Causal Model

class pyagrum.causal.CausalModel(bn, latentVarsDescriptor=None, keepArcs=False)

From an observational BNs and the description of latent variables, this class represent a complet causal model obtained by adding the latent variables specified in latentVarsDescriptor to the Bayesian network bn.

• Parameters:
  • bn (pyagrum.BayesNet) – an observational Bayesian network
  • latentVarsDescriptor (List [ (**str ,**List [**int ] ) ]) – list of couples (, <list of affected variables’ ids>).
  • keepArcs (bool) – By default, the arcs between variables affected by a common latent variable will be removed but this can be avoided by setting keepArcs to True

addCausalArc(x, y)

Add an arc x->y

• Parameters:
  • x (int |**str) – the nodeId or the name of the first node
  • y (int |**str) – the nodeId or the name of the second node
• Return type: None

addLatentVariable(name, lchild, keepArcs=False)

Add a new latent variable with a name, a tuple of children and replacing (or not) correlations between children.

• Parameters:
  • name (str) – the name of the latent variable
  • lchild (Tuple [**str ,**str ]) – the tuple of (2) children
  • keepArcs (bool) – do wee keep (or not) the arc between the children ?
• Return type: None

arcs()

• Return type: Set[Tuple[NewType(NodeId, int), NewType(NodeId, int)]]
• Returns: the set of arcs

backDoor(cause, effect, withNames=True)

Check if a backdoor exists between cause and effect

• Parameters:
  • cause (int |**str) – the nodeId or the name of the cause
  • effect (int |**str) – the nodeId or the name of the effect
  • withNames (bool) – wether we use ids (int) or names (str)
• Returns: None if no found backdoor. Otherwise return the found backdoors as set of ids or set of names.
• Return type: None|Set[str]|Set[int]

causalBN()

• Return type: BayesNet
• Returns: the causal Bayesian network
• Warning: do not infer any computations in this model. It is strictly a structural model

children(x)

From a NodeId, returns its children (as a set of NodeId)

• Parameters: x (int) – the node
• Returns: the set of children
• Return type: Set[int]

connectedComponents()

Return a map of connected components and their nodes.

• Returns: thedisc of connected components
• Return type: Dict[int,NodeSet]

eraseCausalArc(x, y)

Erase the arc x->y

• Parameters:
  • x (int |**str) – the nodeId or the name of the first node
  • y (int |**str) – the nodeId or the name of the second node
• Return type: None

existsArc(x, y)

Does the arc x->y exist ?

• Parameters:
  • x (int |**str) – the nodeId or the name of the first node
  • y (int |**str) – the nodeId or the name of the second node
• Returns: True if the arc exists.
• Return type: bool

frontDoor(cause, effect, withNames=True)

Check if a frontdoor exists between cause and effet

• Parameters:
  • cause (int |**str) – the nodeId or the name of the cause
  • effect (int |**str) – the nodeId or the name of the effect
  • withNames (bool) – wether we use ids (int) or names (str)
• Returns: None if no found frontdoot. Otherwise return the found frontdoors as set of ids or set of names.
• Return type: None|Set[str]|Set[int]

idFromName(name)

• Parameters: name (str) – the name of the variable
• Returns: the id of the variable
• Return type: int

latentVariablesIds()

• Returns: the set of ids of latent variables in the causal model
• Return type: NodeSet

names()

• Returns: the map NodeId,Name
• Return type: Dict[int,str]

nodes()

• Return type: Set[NewType(NodeId, int)]
• Returns: the set of nodes

observationalBN()

• Return type: BayesNet
• Returns: the observational Bayesian network

parents(x)

From a NodeId, returns its parent (as a set of NodeId)

• Parameters: x (int) – the node
• Returns: the set of parents
• Return type: Set[int]

toDot()

Create a dot representation of the causal model

• Return type: str
• Returns: the dot representation in a string