def generate_CET_forest(ori_graph, graph, explanatory_var, observation, explanadum, path):
if len(explanatory_var) is 0:
return [ExplanationTreeNode()]
forest = []
for x in explanatory_var:
for value in graph.get_node_with_name(x).cpt.values():
intervened_graph = graph.create_graph_with_intervention( {x:value} )
new_forest = generate_CET_forest(ori_graph, intervened_graph, cut(explanatory_var, x), observation, explanadum, path + [(x, value)])
for new_tree in new_forest:
t = ExplanationTreeNode(parent = path[-1][0] if path else None, root = x)
strength = math.log( intervened_graph.prob_given(explanadum, observation) / \
ori_graph.prob_given(explanadum, observation), 2) if intervened_graph.prob_given(explanadum, observation) and ori_graph.prob_given(explanadum, observation) else float("NaN")
t.add_branch(value, new_tree, strength)
forest.append(t)
return forest
def generate_causal_explanation_tree(ori_graph, graph, explanatory_var, observation, explanadum, path, alpha):
x, inf = max_causal_information(graph, explanatory_var, observation, explanadum)
if inf < alpha:
return ExplanationTreeNode()
if len(explanatory_var) is 0:
return ExplanationTreeNode()
t = ExplanationTreeNode(parent = path[-1][0] if path else None, root = x) #new tree with a parent pointer to its parent
for value in graph.get_node_with_name(x).cpt.values():
intervened_graph = graph.create_graph_with_intervention( {x:value} )
new_tree = generate_causal_explanation_tree(ori_graph, intervened_graph, cut(explanatory_var, x), \
observation, explanadum, path + [(x, value)], alpha)
strength = math.log( intervened_graph.prob_given(explanadum, observation) / \
ori_graph.prob_given(explanadum, observation), 2) if intervened_graph.prob_given(explanadum, observation) and ori_graph.prob_given(explanadum, observation) else float("NaN")
t.add_branch(value, new_tree, strength)
return t
def generate_CET_forest(ori_graph, graph, explanatory_var, observation,
explanadum, path):
if len(explanatory_var) is 0:
return [ExplanationTreeNode()]
forest = []
for x in explanatory_var:
for value in graph.get_node_with_name(x).cpt.values():
intervened_graph = graph.create_graph_with_intervention({x: value})
new_forest = generate_CET_forest(ori_graph, intervened_graph,
cut(explanatory_var,
x), observation, explanadum,
path + [(x, value)])
for new_tree in new_forest:
t = ExplanationTreeNode(parent=path[-1][0] if path else None,
root=x)
strength = math.log( intervened_graph.prob_given(explanadum, observation) / \
ori_graph.prob_given(explanadum, observation), 2) if intervened_graph.prob_given(explanadum, observation) and ori_graph.prob_given(explanadum, observation) else float("NaN")
t.add_branch(value, new_tree, strength)
forest.append(t)
return forest
def generate_causal_explanation_tree(ori_graph, graph, explanatory_var,
observation, explanadum, path, alpha):
x, inf = max_causal_information(graph, explanatory_var, observation,
explanadum)
if inf < alpha:
return ExplanationTreeNode()
if len(explanatory_var) is 0:
return ExplanationTreeNode()
t = ExplanationTreeNode(
parent=path[-1][0] if path else None,
root=x) #new tree with a parent pointer to its parent
for value in graph.get_node_with_name(x).cpt.values():
intervened_graph = graph.create_graph_with_intervention({x: value})
new_tree = generate_causal_explanation_tree(ori_graph, intervened_graph, cut(explanatory_var, x), \
observation, explanadum, path + [(x, value)], alpha)
strength = math.log( intervened_graph.prob_given(explanadum, observation) / \
ori_graph.prob_given(explanadum, observation), 2) if intervened_graph.prob_given(explanadum, observation) and ori_graph.prob_given(explanadum, observation) else float("NaN")
t.add_branch(value, new_tree, strength)
return t
