def test_bad_names2():
"""
Now the distributions have bad names.
"""
x = nx.DiGraph()
d = dit.example_dists.Xor()
cdist, dists = d.condition_on([0, 1])
x.add_edge(0, 2)
x.add_edge(1, 2)
# Node 2 should have more than one dist. If we pass just a distribution in,
# as if it had no parents, then an exception should raise.
# x.nodes[2]['dist'] = (cdist.outcomes, dists)
x.nodes[2]['dist'] = cdist.marginal([0])
x.nodes[0]['dist'] = cdist.marginal([0])
x.nodes[1]['dist'] = cdist.marginal([1])
with pytest.raises(Exception):
dit.distribution_from_bayesnet(x)
# If they don't have the same length, it's invalid too.
x.nodes[2]['dist'] = (cdist.outcomes, dists[:0])
with pytest.raises(Exception):
dit.distribution_from_bayesnet(x)
def test_distribution_from_bayesnet_func():
# Smoke test with distributions as functions.
x = nx.DiGraph()
x.add_edge('A', 'C')
x.add_edge('B', 'C')
d = dit.example_dists.Xor()
sample_space = d._sample_space
def uniform(node_val, parents):
return 0.5
def xor(node_val, parents):
if parents['A'] != parents['B']:
output = '1'
else:
output = '0'
# If output agrees with passed in output, p = 1
p = int(output == node_val)
return p
x.nodes['C']['dist'] = xor
x.nodes['A']['dist'] = uniform
x.nodes['B']['dist'] = uniform
# Samplespace is required when functions are callable.
with pytest.raises(ValueError):
dit.distribution_from_bayesnet(x)
d2 = dit.distribution_from_bayesnet(x, sample_space=sample_space)
assert d.is_approx_equal(d2)
ss = ['000', '001', '010', '011', '100', '101', '110', '111']
d3 = dit.distribution_from_bayesnet(x, sample_space=ss)
# Can't test using is_approx_equal, since one has SampleSpace and the
# other has CartesianProduct as sample spaces. So is_approx_equal would
# always be False.
assert np.allclose(d.pmf, d3.pmf)
assert list(d.sample_space()) == list(d3.sample_space())
def test_distribution_from_bayesnet_nonames():
# Smoke test without rv names.
x = nx.DiGraph()
d = dit.example_dists.Xor()
cdist, dists = d.condition_on([0, 1])
x.add_edge(0, 2)
x.add_edge(1, 2)
x.node[2]['dist'] = (cdist.outcomes, dists)
x.node[0]['dist'] = cdist.marginal([0])
x.node[1]['dist'] = cdist.marginal([1])
d2 = dit.distribution_from_bayesnet(x)
d3 = dit.distribution_from_bayesnet(x, [0, 1, 2])
assert_true( d.is_approx_equal(d2) )
assert_true( d.is_approx_equal(d3) )
# Use a dictionary too
x.node[2]['dist'] = dict(zip(cdist.outcomes, dists))
d4 = dit.distribution_from_bayesnet(x)
assert_true( d.is_approx_equal(d4) )
del x.node[1]['dist']
assert_raises(ValueError, dit.distribution_from_bayesnet, x)
def test_distribution_from_bayesnet_func():
# Smoke test with distributions as functions.
x = nx.DiGraph()
x.add_edge('A', 'C')
x.add_edge('B', 'C')
d = dit.example_dists.Xor()
sample_space = d._sample_space
def uniform(node_val, parents):
return 0.5
def xor(node_val, parents):
if parents['A'] != parents['B']:
output = '1'
else:
output = '0'
# If output agrees with passed in output, p = 1
p = int(output == node_val)
return p
x.node['C']['dist'] = xor
x.node['A']['dist'] = uniform
x.node['B']['dist'] = uniform
# Samplespace is required when functions are callable.
assert_raises(ValueError, dit.distribution_from_bayesnet, x)
d2 = dit.distribution_from_bayesnet(x, sample_space=sample_space)
assert_true( d.is_approx_equal(d2) )
ss = ['000', '001', '010', '011', '100', '101', '110', '111']
d3 = dit.distribution_from_bayesnet(x, sample_space=ss)
# Can't test using is_approx_equal, since one has SampleSpace and the
# other has CartesianProduct as sample spaces. So is_approx_equal would
# always be False.
assert_true( np.allclose(d.pmf, d3.pmf) )
assert_equal( list(d.sample_space()), list(d3.sample_space()) )
def test_distribution_from_bayesnet_nonames():
# Smoke test without rv names.
x = nx.DiGraph()
d = dit.example_dists.Xor()
cdist, dists = d.condition_on([0, 1])
x.add_edge(0, 2)
x.add_edge(1, 2)
x.node[2]['dist'] = (cdist.outcomes, dists)
x.node[0]['dist'] = cdist.marginal([0])
x.node[1]['dist'] = cdist.marginal([1])
d2 = dit.distribution_from_bayesnet(x)
d3 = dit.distribution_from_bayesnet(x, [0, 1, 2])
assert_true(d.is_approx_equal(d2))
assert_true(d.is_approx_equal(d3))
# Use a dictionary too
x.node[2]['dist'] = dict(zip(cdist.outcomes, dists))
d4 = dit.distribution_from_bayesnet(x)
assert_true(d.is_approx_equal(d4))
del x.node[1]['dist']
assert_raises(ValueError, dit.distribution_from_bayesnet, x)
def test_distribution_from_bayesnet_names():
# Smoke test with rv names.
x = nx.DiGraph()
d = dit.example_dists.Xor()
d.set_rv_names(['A', 'B', 'C'])
cdist, dists = d.condition_on(['A', 'B'])
x.add_edge('A', 'C')
x.add_edge('B', 'C')
x.nodes['C']['dist'] = (cdist.outcomes, dists)
x.nodes['A']['dist'] = cdist.marginal(['A'])
x.nodes['B']['dist'] = cdist.marginal(['B'])
d2 = dit.distribution_from_bayesnet(x)
assert d.is_approx_equal(d2)
# Specify names
d3 = dit.distribution_from_bayesnet(x, ['A', 'B', 'C'])
assert d.is_approx_equal(d2)
# Test with a non-Cartesian product distribution
dd = x.nodes['B']['dist']
dd._sample_space = dit.SampleSpace(list(dd.sample_space()))
d3 = dit.distribution_from_bayesnet(x)
assert d.is_approx_equal(d3)
def test_distribution_from_bayesnet_error():
# Test distribution_from_bayesnet with functions and distributions.
# This is not allowed and should fail.
x = nx.DiGraph()
x.add_edge('A', 'C')
x.add_edge('B', 'C')
d = dit.example_dists.Xor()
sample_space = d._sample_space
def uniform(node_val, parents):
return 0.5
unif = dit.Distribution('01', [.5, .5])
unif.set_rv_names('A')
x.nodes['C']['dist'] = uniform
x.nodes['A']['dist'] = unif
x.nodes['B']['dist'] = uniform
with pytest.raises(Exception):
dit.distribution_from_bayesnet(x, sample_space=sample_space)
def test_distribution_from_bayesnet_names():
# Smoke test with rv names.
x = nx.DiGraph()
d = dit.example_dists.Xor()
d.set_rv_names(['A', 'B', 'C'])
cdist, dists = d.condition_on(['A', 'B'])
x.add_edge('A', 'C')
x.add_edge('B', 'C')
x.node['C']['dist'] = (cdist.outcomes, dists)
x.node['A']['dist'] = cdist.marginal(['A'])
x.node['B']['dist'] = cdist.marginal(['B'])
d2 = dit.distribution_from_bayesnet(x)
assert_true( d.is_approx_equal(d2) )
# Specify names
d3 = dit.distribution_from_bayesnet(x, ['A', 'B', 'C'])
assert_true( d.is_approx_equal(d2) )
# Test with a non-Cartesian product distribution
dd = x.node['B']['dist']
dd._sample_space = dit.SampleSpace(list(dd.sample_space()))
d3 = dit.distribution_from_bayesnet(x)
assert_true( d.is_approx_equal(d3) )
def test_bad_names1():
x = nx.DiGraph()
d = dit.example_dists.Xor()
cdist, dists = d.condition_on([0, 1])
x.add_edge(0, 2)
x.add_edge(1, 2)
x.nodes[2]['dist'] = (cdist.outcomes, dists)
x.nodes[0]['dist'] = cdist.marginal([0])
x.nodes[1]['dist'] = cdist.marginal([1])
with pytest.raises(ValueError):
dit.distribution_from_bayesnet(x, [0, 1])
with pytest.raises(ValueError):
dit.distribution_from_bayesnet(x, [0, 1, 1])
with pytest.raises(ValueError):
dit.distribution_from_bayesnet(x, [0, 1, 2, 3])
with pytest.raises(ValueError):
dit.distribution_from_bayesnet(x, [0, 1, 2, 2])
