def test_2_multinom_RVs(df_2_multinomial_indep_RVs):
df = df_2_multinomial_indep_RVs(size=10000)
graph = Graph(variables=list(df.columns), complete=True)
skeleton_finder = PCSkeletonFinder(data=df, graph=graph)
cond_sets_satisfying_cond_indep = skeleton_finder.find()
assert graph.get_edges() == set({})
assert cond_sets_satisfying_cond_indep['x _||_ y'] == set({frozenset({})})
def test_2_deterministic_and_3rd_var_caused_by_one_of_them(
df_2_deterministic_and_3rd_var_caused_by_one_of_them):
df = df_2_deterministic_and_3rd_var_caused_by_one_of_them(size=1000)
graph = Graph(variables=list(df.columns), complete=True)
skeleton_finder = PCSkeletonFinder(data=df, graph=graph)
skeleton_finder.find()
assert graph.has_adjacency(('x', 'y'))
assert graph.get_nodes() == set({'x', 'y', 'z'}) # pylint: disable='no-member'
def test_skeleton_finder_X_causes_Y(df_X_causes_Y):
df = df_X_causes_Y(size=1000)
graph = Graph(variables=list(df.columns), complete=True)
skeleton_finder = PCSkeletonFinder(data=df, graph=graph)
cond_sets_satisfying_cond_indep = skeleton_finder.find()
assert graph.has_adjacency(('x', 'y'))
assert cond_sets_satisfying_cond_indep == {}
def test_skeleton_finder_Z_causes_X_and_Y(df_Z_causes_X_and_Y):
df = df_Z_causes_X_and_Y(size=1000)
graph = Graph(variables=list(df.columns), complete=True)
skeleton_finder = PCSkeletonFinder(data=df, graph=graph)
cond_sets_satisfying_cond_indep = skeleton_finder.find()
assert graph.has_adjacency(('x', 'z'))
assert graph.has_adjacency(('y', 'z'))
assert cond_sets_satisfying_cond_indep == \
{'x _||_ y': set({frozenset({'z'})})}
def test_chain_and_collider_without_MI(df_chain_and_collider_without_MI):
size = 10000
df = df_chain_and_collider_without_MI(size=size)
graph = Graph(variables=list(df.columns), complete=True)
skeleton_finder = PCSkeletonFinder(data=df, graph=graph)
skeleton_finder.find()
assert graph.has_adjacency(('a', 'b'))
assert graph.has_adjacency(('b', 'c'))
assert graph.has_adjacency(('a', 'd'))
assert graph.has_adjacency(('c', 'd'))
def test_3_multinom_RVs_MAR(df_Z_causes_X_Y_and_X_Z_causes_MI_Y):
size = 70000
df = df_Z_causes_X_Y_and_X_Z_causes_MI_Y(size=size)
graph = Graph(variables=list(df.columns), complete=True)
skeleton_finder = PCSkeletonFinder(data=df, graph=graph)
skeleton_finder.find()
assert set(graph.get_nodes()).intersection(set(['x', 'y', 'MI_x'])) # pylint: disable='no-member'
assert graph.has_adjacency(('x', 'z'))
assert graph.has_adjacency(('y', 'z'))
def test_chain_and_collider_with_MI(df_chain_and_collider_with_MI):
size = 20000
df = df_chain_and_collider_with_MI(size=size)
graph = Graph(variables=list(df.columns), complete=True)
skeleton_finder = PCSkeletonFinder(data=df, graph=graph)
skeleton_finder.find()
# we expect a-c in this intermediate stage. a-c is spurious, due to
# collider bias.
assert graph.has_adjacency(('a', 'c'))
assert graph.has_adjacency(('a', 'b'))
assert graph.has_adjacency(('b', 'c'))
assert graph.has_adjacency(('a', 'd'))
assert graph.has_adjacency(('c', 'd'))
def test_init_complete_then_add_edge():
graph = PartialAncestralGraph(complete=True, variables=['A', 'B', 'C'])
graph.add_edge('E o-> D')
assert graph.has_edge('E o-> D')
assert graph.has_edge('D <-o E')
assert graph.has_edge('A o-o B')
assert graph.has_edge('C o-o B')
assert graph.has_edge('A o-o C')
assert graph.has_edge('D o-o A') is False
def test_simple_chain():
graph = PAG(variables=['X', 'Y', 'Z'])
graph.add_edge('X o-o Y')
graph.add_edge('Z o-o Y')
sep_sets = SepSets()
sep_sets.add(
node_1='X',
node_2='Z',
cond_set=set({'Y'})
)
immoralities = ImmoralitiesFinder(
graph=graph,
sep_sets=sep_sets
).find()
assert len(immoralities) == 0
def test_long_chains_collider_bias_with_MI(
df_long_chains_and_collider_with_MI):
size = 10000
df = df_long_chains_and_collider_with_MI(size=size, proba_noise=0.7)
graph = Graph(variables=list(df.columns), complete=True)
skeleton_finder = PCSkeletonFinder(data=df, graph=graph)
skeleton_finder.find()
# we expect b-d in this intermediate stage. b-d is spurious, due to
# collider bias.
assert graph.has_adjacency(('a', 'b'))
assert graph.has_adjacency(('b', 'd'))
assert graph.has_adjacency(('d', 'e'))
assert graph.has_adjacency(('b', 'c'))
assert graph.has_adjacency(('c', 'd'))
def test_simple_collider():
graph = PAG(
variables=['Parent 1', 'Parent 2', 'collider'],
)
graph.add_edge('Parent 1 o-o collider')
graph.add_edge('Parent 2 o-o collider')
sep_sets = SepSets()
sep_sets.add(
node_1='Parent 1',
node_2='Parent_2',
cond_set=set({'collider'})
)
immoralities = ImmoralitiesFinder(
graph=graph,
sep_sets=sep_sets
).find()
assert ('Parent 1', 'collider', 'Parent 2') in immoralities
def test_init_complete_graph():
graph = PartialAncestralGraph(complete=True, variables=['A', 'B', 'C'])
assert graph.has_edge('A o-o B')
assert graph.has_edge('C o-o B')
assert graph.has_edge('A o-o C')
assert set({'A', 'C'}) == graph.get_neighbors('B')
assert set({'B', 'C'}) == graph.get_neighbors('A')
assert Edge('A o-o B') in graph.get_edges()
assert Edge('A o-o A') not in graph.get_edges()
assert Edge('B o-o B') not in graph.get_edges()
def test_get_edge():
graph = PartialAncestralGraph()
assert graph.get_edge('A', 'B') is None
graph.add_edge('A o-o B')
result = graph.get_edge('A', 'B')
assert result[0] == 'A'
assert result[1] == 'o-o'
assert result[2] == 'B'
def test_firing_squad_example():
graph = PAG(
variables=[
'Captain ordered to shoot',
'Rifleman 1 shot',
'Rifleman 2 shot',
'Prisoner hit by bullet',
'Prisoner dead'
]
)
graph.add_edge('Rifleman 1 shot o-o Prisoner hit by bullet')
graph.add_edge('Rifleman 2 shot o-o Prisoner hit by bullet')
graph.add_edge('Captain ordered to shoot o-o Rifleman 1 shot')
graph.add_edge('Captain ordered to shoot o-o Rifleman 2 shot')
graph.add_edge('Prisoner hit by bullet o-o Prisoner dead')
sep_sets = SepSets()
sep_sets.add(
node_1='Captain ordered to shoot',
node_2='Prisoner hit by bullet',
cond_set=set({'Rifleman 1 shot', 'Rifleman 2 shot'})
)
sep_sets.add(
node_1='Prisoner dead',
node_2='Rifleman 1 shot',
cond_set=set({'Prisoner hit by bullet', 'Rifleman 2 shot'})
)
sep_sets.add(
node_1='Prisoner dead',
node_2='Rifleman 2 shot',
cond_set=set({'Prisoner hit by bullet', 'Rifleman 1 shot'})
)
sep_sets.add(
node_1='Rifleman 1 shot',
node_2='Rifleman 2 shot',
cond_set=set({'Captain ordered to shoot'})
)
immoralities = ImmoralitiesFinder(
graph=graph,
sep_sets=sep_sets
).find()
assert ('Rifleman 1 shot', 'Prisoner hit by bullet', 'Rifleman 2 shot') \
in immoralities
def test_dog_pee():
size = 100000
# Sometimes cloudy
cloudy = np.random.binomial(n=1, p=0.5, size=size)
# Cloudyness causes rain, but sometimes it rains even when it's not cloudy.
rain = cloudy * np.random.binomial(n=1, p=0.7, size=size) \
+ (1 - cloudy) * np.random.binomial(n=1, p=0.1, size=size)
# Sprinkler generally turns on when it isn't cloudy.
sprinkler = (cloudy == 0) * np.random.binomial(n=1, p=0.8, size=size) \
+ cloudy * np.random.binomial(n=1, p=0.1, size=size)
# Grass is generally wet whenever it rained or the sprinkler is on.
wet_grass = (rain | sprinkler) * np.random.binomial(n=1, p=0.90, size=size)
# Dog doesn't like to get rained on
# Dog goes out more frequently when it's not raining
dog_goes_out_to_pee = rain * np.random.binomial(n=1, p=0.2, size=size) \
+ (1 - rain) * np.random.binomial(n=1, p=0.9, size=size)
df = pd.DataFrame({
'cloudy': cloudy,
'sprinkler': sprinkler,
'rain': rain,
'wet_grass': wet_grass,
'dog_goes_out_to_pee': dog_goes_out_to_pee
})
graph = Graph(variables=list(df.columns), complete=True)
skeleton_finder = PCSkeletonFinder(data=df, graph=graph)
skeleton_finder.find()
assert graph.has_adjacency(('cloudy', 'rain'))
assert graph.has_adjacency(('cloudy', 'sprinkler'))
assert graph.has_adjacency(('rain', 'dog_goes_out_to_pee'))
assert graph.has_adjacency(('rain', 'wet_grass'))
assert graph.has_adjacency(('sprinkler', 'wet_grass'))
def test_remove_edge():
graph = PartialAncestralGraph()
graph.add_edge('A o-o B')
assert graph.has_adjacency(('A', 'B')) is True
graph.remove_edge(('A', 'B'))
assert graph.has_adjacency(('A', 'B')) is False
graph = PartialAncestralGraph()
graph.add_edge('A <-o B')
assert graph.has_adjacency(('A', 'B')) is True
graph.remove_edge(('A', 'B'))
assert graph.has_adjacency(('A', 'B')) is False
def test_ancestral_validation():
# No directed cycles
graph = PartialAncestralGraph()
graph.add_edge('A --> B')
graph.add_edge('B --> C')
with pytest.raises(NotAncestralError):
graph.add_edge('C --> A')
graph = PartialAncestralGraph()
graph.add_edge('C --> A')
graph.add_edge('B --> C')
with pytest.raises(NotAncestralError):
graph.add_edge('A --> B')
# No almost-directed cycles
graph = PartialAncestralGraph()
graph.add_edge('A --> B')
graph.add_edge('B --> C')
with pytest.raises(NotAncestralError):
graph.add_edge('C <-> A')
graph = PartialAncestralGraph()
graph.add_edge('A --> B')
graph.add_edge('C <-> A')
with pytest.raises(NotAncestralError):
graph.add_edge('B --> C')
# Nodes of undirected edges can't have siblings.
graph = PartialAncestralGraph()
graph.add_edge('A --- B')
with pytest.raises(NotAncestralError):
graph.add_edge('B <-> C')
graph = PartialAncestralGraph()
graph.add_edge('B <-> C')
with pytest.raises(NotAncestralError):
graph.add_edge('A --- B')
def test_dog_example():
df = dog_example(size=100000)
graph = Graph(variables=list(df.columns), complete=True)
skeleton_finder = PCSkeletonFinder(data=df, graph=graph)
cond_sets_satisfying_cond_indep = \
skeleton_finder.find()
assert cond_sets_satisfying_cond_indep[key_for_pair(
('activity', 'dog_tired'))].intersection(
set({frozenset({'exercise_levels'})
})) == set({frozenset({'exercise_levels'})})
assert set({frozenset({'best_friends_visit', 'activity'})}) not in \
cond_sets_satisfying_cond_indep[
key_for_pair(('weekend', 'mentally_exhausted_before_bed'))
]
assert graph.has_adjacency(('rain', 'best_friends_visit'))
assert graph.has_adjacency(('weekend', 'best_friends_visit'))
assert graph.has_adjacency(('rain', 'activity'))
assert graph.has_adjacency(('exercise_levels', 'best_friends_visit'))
assert graph.has_adjacency(('exercise_levels', 'activity'))
assert graph.has_adjacency(('mentally_exhausted_before_bed', 'activity'))
assert graph.has_adjacency(('exercise_levels', 'dog_tired'))
assert graph.has_adjacency(
('best_friends_visit', 'mentally_exhausted_before_bed'))
assert graph.has_adjacency(
('mentally_exhausted_before_bed', 'dog_teeth_brushed'))
assert graph.has_adjacency(('dog_tired', 'dog_teeth_brushed'))
def test_add_edge():
graph = PartialAncestralGraph()
assert graph.has_edge('A o-o B') is False
assert graph.has_edge('B o-o A') is False
graph.add_edge('A o-o B')
assert graph.has_edge('A o-o B') is True
assert graph.has_edge('B o-o A') is True
graph = PartialAncestralGraph()
assert graph.has_edge('A --o B') is False
assert graph.has_edge('B o-- A') is False
graph.add_edge('A --o B')
assert graph.has_edge('B o-- A') is True
assert graph.has_edge('A --o B') is True
graph = PartialAncestralGraph()
assert graph.has_edge('A o-- B') is False
assert graph.has_edge('B --o A') is False
graph.add_edge('A o-- B')
assert graph.has_edge('B --o A') is True
assert graph.has_edge('A o-- B') is True
graph = PartialAncestralGraph()
assert graph.has_edge('A --- B') is False
assert graph.has_edge('B --- A') is False
graph.add_edge('A --- B')
assert graph.has_edge('B --- A') is True
assert graph.has_edge('A --- B') is True
graph = PartialAncestralGraph()
assert graph.has_edge('A o-> B') is False
assert graph.has_edge('B <-o A') is False
graph.add_edge('A o-> B')
assert graph.has_edge('A o-> B') is True
assert graph.has_edge('B <-o A') is True
graph = PartialAncestralGraph()
assert graph.has_edge('A <-> B') is False
assert graph.has_edge('B <-> A') is False
graph.add_edge('A <-> B')
assert graph.has_edge('A <-> B') is True
assert graph.has_edge('B <-> A') is True
