def test_categorical_leaf_serialization(tmpdir):
"""Tests the binary serialization of two SPFlow Categorical leaf nodes
by round-tripping and comparing the parameters before and after serialization
& deserialization"""
c1 = Categorical(p=[0.35, 0.55, 0.1], scope=1)
c2 = Categorical(p=[0.25, 0.625, 0.125], scope=2)
p = Product(children=[c1, c2])
binary_file = os.path.join(tmpdir, "test.bin")
print(f"Test binary file: {binary_file}")
model = SPNModel(p, "uint8", "test")
query = JointProbability(model)
BinarySerializer(binary_file).serialize_to_file(query)
deserialized = BinaryDeserializer(binary_file).deserialize_from_file()
assert (isinstance(deserialized, JointProbability))
assert (isinstance(deserialized.graph, SPNModel))
assert (deserialized.graph.featureType == model.featureType)
assert (deserialized.graph.name == model.name)
deserialized = deserialized.graph.root
assert isinstance(deserialized, Product)
assert (len(deserialized.children) == 2)
assert len(c1.p) == len(deserialized.children[0].p)
for i, p in enumerate(c1.p):
assert p == deserialized.children[0].p[i]
assert len(c2.p) == len(deserialized.children[1].p)
for i, p in enumerate(c2.p):
assert p == deserialized.children[1].p[i]
def test_conditional_probability(self):
# test if conditional probability is correct
# same spn as in entropy test
# only for generating the ds_context
train_data = np.array([[0.0, 1.0, 0.0], [1.0, 0.0, 1.0], [2.0, 0.0, 1.0]])
# spn
ds_context = Context(meta_types=[MetaType.DISCRETE] * 3)
ds_context.add_domains(train_data)
ds_context.parametric_type = [Categorical] * 3
spn = 0.64 * (
(
Categorical(p=[0.25, 0.75, 0.0], scope=0)
* (
0.34 * ((Categorical(p=[7 / 34, 27 / 34], scope=1) * Categorical(p=[1.0, 0.0], scope=2)))
+ 0.66 * ((Categorical(p=[21 / 22, 1 / 22], scope=1) * Categorical(p=[0.0, 1.0], scope=2)))
)
)
) + 0.36 * (
(
Categorical(p=[0.0, 0.0, 1.0], scope=0)
* (
0.34 * ((Categorical(p=[7 / 34, 27 / 34], scope=1) * Categorical(p=[1.0, 0.0], scope=2)))
+ 0.66 * ((Categorical(p=[21 / 22, 1 / 22], scope=1) * Categorical(p=[0.0, 1.0], scope=2)))
)
)
)
# tests
x_instance = np.array([1, 1, 0], dtype=float).reshape(1, -1)
self.assertAlmostEqual(conditional_probability(spn, 2, x_instance)[0][0], 0.9)
self.assertAlmostEqual(conditional_probability(spn, 0, x_instance)[0][0], 0.48)
x_instance = np.array([2, 1, 0], dtype=float).reshape(1, -1)
self.assertAlmostEqual(conditional_probability(spn, 0, x_instance)[0][0], 0.36)
def test_we_score(self):
# test if we_score is correct
"""
# explain how training data and the spn comes
# number of RVs
M = 3
# table of probabilities
p1 = 0.6
p2 = 0.3
p31 = 0.1
p32 = 0.9
# generate x1 and x2
x1 = np.random.binomial(1, p1, size=N) + np.random.binomial(1, p1, size=N)
x2 = np.random.binomial(1, p2, size=N)
x3 = np.zeros(N)
# generate x3
for i in range(N):
if x2[i] == 1:
x3[i] = np.random.binomial(1, p31, size=1)
else:
x3[i] = np.random.binomial(1, p32, size=1)
# form a matrix, rows are instances and columns are RVs
train_data = np.concatenate((x1, x2, x3)).reshape((M, N)).transpose()
"""
# only for generating the ds_context
train_data = np.array([[0.0, 1.0, 0.0], [1.0, 0.0, 1.0], [2.0, 0.0, 1.0]])
# spn
ds_context = Context(meta_types=[MetaType.DISCRETE] * 3)
ds_context.add_domains(train_data)
ds_context.parametric_type = [Categorical] * 3
spn = 0.64 * (
(
Categorical(p=[0.25, 0.75, 0.0], scope=0)
* (
0.34 * ((Categorical(p=[7 / 34, 27 / 34], scope=1) * Categorical(p=[1.0, 0.0], scope=2)))
+ 0.66 * ((Categorical(p=[21 / 22, 1 / 22], scope=1) * Categorical(p=[0.0, 1.0], scope=2)))
)
)
) + 0.36 * (
(
Categorical(p=[0.0, 0.0, 1.0], scope=0)
* (
0.34 * ((Categorical(p=[7 / 34, 27 / 34], scope=1) * Categorical(p=[1.0, 0.0], scope=2)))
+ 0.66 * ((Categorical(p=[21 / 22, 1 / 22], scope=1) * Categorical(p=[0.0, 1.0], scope=2)))
)
)
)
# test
n = 40000
x_instance = np.array([1, 1, 0], dtype=float).reshape(1, -1)
y_index = 0
we = weight_of_evidence(spn, 0, x_instance, n, ds_context.domains[y_index].shape[0])
we_true = np.array([[np.nan, 0, 0]])
we = we[~np.isnan(we)]
we_true = we_true[~np.isnan(we_true)]
self.assertTrue((we == we_true).all())
def get_gender_spn():
from spn.structure.leaves.parametric.Parametric import Categorical, Gaussian
spn1 = Categorical(p=[0.0, 1.0], scope=[0]) * Categorical(p=[0.2, 0.8],
scope=[1])
spn2 = Categorical(p=[1.0, 0.0], scope=[0]) * Categorical(p=[0.7, 0.3],
scope=[1])
spn3 = 0.4 * spn1 + 0.6 * spn2
spn = spn3 * Gaussian(mean=20, stdev=3, scope=[2])
spn.scope = sorted(spn.scope)
return spn
def test_induced_trees(self):
spn = 0.5 * (Gaussian(mean=10, stdev=1, scope=0) * Categorical(p=[1.0, 0], scope=1)) + \
0.5 * (Gaussian(mean=50, stdev=1, scope=0) * Categorical(p=[0, 1.0], scope=1))
data = np.zeros((2, 2))
data[1, 1] = 1
data[:, 0] = np.nan
mpevals = mpe(spn, data)
self.assertAlmostEqual(mpevals[0, 0], 10)
self.assertAlmostEqual(mpevals[1, 0], 50)
def create_SPN2():
from spn.structure.Base import assign_ids
from spn.structure.Base import rebuild_scopes_bottom_up
from spn.algorithms.Validity import is_valid
from spn.structure.leaves.parametric.Parametric import Categorical
from spn.structure.Base import Sum, Product
p0 = Product(children=[
Categorical(p=[0.3, 0.7], scope=1),
Categorical(p=[0.4, 0.6], scope=2)
])
p1 = Product(children=[
Categorical(p=[0.5, 0.5], scope=1),
Categorical(p=[0.6, 0.4], scope=2)
])
s1 = Sum(weights=[0.3, 0.7], children=[p0, p1])
p2 = Product(children=[Categorical(p=[0.2, 0.8], scope=0), s1])
p3 = Product(children=[
Categorical(p=[0.2, 0.8], scope=0),
Categorical(p=[0.3, 0.7], scope=1)
])
p4 = Product(children=[p3, Categorical(p=[0.4, 0.6], scope=2)])
spn = Sum(weights=[0.4, 0.6], children=[p2, p4])
assign_ids(spn)
rebuild_scopes_bottom_up(spn)
val, msg = is_valid(spn)
assert val, msg
return spn
def test_induced_trees(self):
spn = 0.5 * (Gaussian(mean=10, stdev=0.000000001, scope=0) * Categorical(p=[1.0, 0], scope=1)) + \
0.5 * (Gaussian(mean=50, stdev=0.000000001, scope=0) * Categorical(p=[0, 1.0], scope=1))
rand_gen = np.random.RandomState(17)
data = np.zeros((2, 2))
data[1, 1] = 1
data[:, 0] = np.nan
samples = sample_instances(spn, data, rand_gen)
self.assertAlmostEqual(samples[0, 0], 10)
self.assertAlmostEqual(samples[1, 0], 50)
def test_eval_parametric(self):
data = np.array([1, 1, 1, 1, 1, 1, 1], dtype=np.float32).reshape(
(1, 7))
spn = (Gaussian(mean=1.0, stdev=1.0, scope=[0]) *
Exponential(l=1.0, scope=[1]) *
Gamma(alpha=1.0, beta=1.0, scope=[2]) *
LogNormal(mean=1.0, stdev=1.0, scope=[3]) *
Poisson(mean=1.0, scope=[4]) * Bernoulli(p=0.6, scope=[5]) *
Categorical(p=[0.1, 0.2, 0.7], scope=[6]))
ll = log_likelihood(spn, data)
tf_ll = eval_tf(spn, data)
self.assertTrue(np.all(np.isclose(ll, tf_ll)))
spn_copy = Copy(spn)
tf_graph, data_placeholder, variable_dict = spn_to_tf_graph(
spn_copy, data, 1)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
tf_graph_to_spn(variable_dict)
str_val = spn_to_str_equation(spn)
str_val2 = spn_to_str_equation(spn_copy)
self.assertEqual(str_val, str_val2)
def create_leaf(data, ds_context, scope):
idx = scope[0]
meta_type = ds_context.meta_types[idx]
if meta_type == MetaType.REAL:
if identity_numeric:
return create_identity_leaf(data, scope)
if prior_weight == 0.:
return create_piecewise_leaf(data, ds_context, scope, prior_weight=None)
else:
return create_piecewise_leaf(data, ds_context, scope, prior_weight=prior_weight)
elif meta_type == MetaType.DISCRETE:
unique, counts = np.unique(data[:,0], return_counts=True)
sorted_counts = np.zeros(len(ds_context.domains[idx]), dtype=np.float64)
for i, x in enumerate(unique):
sorted_counts[int(x)] = counts[i]
p = sorted_counts / data.shape[0]
#Do regularization
if prior_weight > 0.:
p += prior_weight
p = p/np.sum(p)
return Categorical(p, scope)
else:
raise Exception("Mehtod learn_mspn_for_aqp(...) cannot create leaf for " + str(meta_type))
def getSpn1():
spn = 0.4 * (Categorical(p=[0.2, 0.8], scope=0) *
(0.3 * (Categorical(p=[0.3, 0.7], scope=1) *
Categorical(p=[0.4, 0.6], scope=2))
+ 0.7 * (Categorical(p=[0.5, 0.5], scope=1) *
Categorical(p=[0.6, 0.4], scope=2)))) \
+ 0.6 * (Categorical(p=[0.2, 0.8], scope=0) *
Categorical(p=[0.3, 0.7], scope=1) *
Categorical(p=[0.4, 0.6], scope=2))
return spn
def test_mutual_info(self):
# test if mutual info is correct
# same spn as in entropy test
# only for generating the ds_context
train_data = np.array([[0.0, 1.0, 0.0], [1.0, 0.0, 1.0], [2.0, 0.0, 1.0]])
# spn
ds_context = Context(meta_types=[MetaType.DISCRETE] * 3)
ds_context.add_domains(train_data)
ds_context.parametric_type = [Categorical] * 3
spn = 0.64 * (
(
Categorical(p=[0.25, 0.75, 0.0], scope=0)
* (
0.34 * ((Categorical(p=[7 / 34, 27 / 34], scope=1) * Categorical(p=[1.0, 0.0], scope=2)))
+ 0.66 * ((Categorical(p=[21 / 22, 1 / 22], scope=1) * Categorical(p=[0.0, 1.0], scope=2)))
)
)
) + 0.36 * (
(
Categorical(p=[0.0, 0.0, 1.0], scope=0)
* (
0.34 * ((Categorical(p=[7 / 34, 27 / 34], scope=1) * Categorical(p=[1.0, 0.0], scope=2)))
+ 0.66 * ((Categorical(p=[21 / 22, 1 / 22], scope=1) * Categorical(p=[0.0, 1.0], scope=2)))
)
)
)
# real mutual info
p2 = 0.3
p3 = 0.66
h_x2 = -p2 * np.log(p2) - (1 - p2) * np.log(1 - p2)
h_x3 = -p3 * np.log(p3) - (1 - p3) * np.log(1 - p3)
h_x2x3 = -(p2 * np.log(p2) + (1 - p2) * np.log(1 - p2) + 0.9 * np.log(0.9) + 0.1 * np.log(0.1))
mi_x2x3 = h_x2 + h_x3 - h_x2x3
self.assertAlmostEqual(mi_x2x3, mutual_information(spn, ds_context, {1}, {2}))
mi_x1x2 = 0
self.assertAlmostEqual(mi_x1x2, mutual_information(spn, ds_context, {1}, {0}))
# test symmetry
self.assertAlmostEqual(
mutual_information(spn, ds_context, {2}, {1}), mutual_information(spn, ds_context, {1}, {2})
)
self.assertAlmostEqual(
mutual_information(spn, ds_context, {0, 2}, {1}), mutual_information(spn, ds_context, {1}, {0, 2})
)
# rest 0
self.assertAlmostEqual(0, mutual_information(spn, ds_context, {2, 1}, {0}))
def test_conditional_mutual_info(self):
# test if conditional mutual info is correct
# same spn as in entropy test
# only for generating the ds_context
train_data = np.array([[0.0, 1.0, 0.0], [1.0, 0.0, 1.0], [2.0, 0.0, 1.0]])
# spn
ds_context = Context(meta_types=[MetaType.DISCRETE] * 3)
ds_context.add_domains(train_data)
ds_context.parametric_type = [Categorical] * 3
spn = 0.64 * (
(
Categorical(p=[0.25, 0.75, 0.0], scope=0)
* (
0.34 * ((Categorical(p=[7 / 34, 27 / 34], scope=1) * Categorical(p=[1.0, 0.0], scope=2)))
+ 0.66 * ((Categorical(p=[21 / 22, 1 / 22], scope=1) * Categorical(p=[0.0, 1.0], scope=2)))
)
)
) + 0.36 * (
(
Categorical(p=[0.0, 0.0, 1.0], scope=0)
* (
0.34 * ((Categorical(p=[7 / 34, 27 / 34], scope=1) * Categorical(p=[1.0, 0.0], scope=2)))
+ 0.66 * ((Categorical(p=[21 / 22, 1 / 22], scope=1) * Categorical(p=[0.0, 1.0], scope=2)))
)
)
)
# real mutual info
p2 = 0.3
p3 = 0.66
h_x1 = -(0.16 * np.log(0.16) + 0.36 * np.log(0.36) + 0.48 * np.log(0.48))
h_x2x1 = -(0.7 * np.log(0.7) + 0.3 * np.log(0.3)) + h_x1
h_x3x1 = -(0.66 * np.log(0.66) + 0.34 * np.log(0.34)) + h_x1
h_x2x3 = -(p2 * np.log(p2) + (1 - p2) * np.log(1 - p2) + 0.9 * np.log(0.9) + 0.1 * np.log(0.1))
h_x2x3x1 = h_x1 + h_x2x3
cmi_x2x3_x1 = h_x2x1 + h_x3x1 - h_x2x3x1 - h_x1
self.assertAlmostEqual(cmi_x2x3_x1, conditional_mutual_information(spn, ds_context, {1}, {2}, {0}))
h_x1x3 = h_x3x1
h_x1x2x3 = h_x2x3x1
h_x3 = -p3 * np.log(p3) - (1 - p3) * np.log(1 - p3)
cmi_x1x2_x3 = h_x1x3 + h_x2x3 - h_x1x2x3 - h_x3
self.assertAlmostEqual(cmi_x1x2_x3, conditional_mutual_information(spn, ds_context, {1}, {0}, {2}))
h_x1x2x3 = h_x2x3x1
h_x2 = -p2 * np.log(p2) - (1 - p2) * np.log(1 - p2)
cmi_x1x3_x2 = h_x2x1 + h_x2x3 - h_x1x2x3 - h_x2
self.assertAlmostEqual(cmi_x1x3_x2, conditional_mutual_information(spn, ds_context, {2}, {0}, {1}))
def create_SPN():
from spn.structure.leaves.parametric.Parametric import Categorical
spn = 0.4 * (Categorical(p=[0.2, 0.8], scope=0) * \
(0.3 * (Categorical(p=[0.3, 0.7], scope=1) * Categorical(p=[0.4, 0.6], scope=2)) + \
0.7 * (Categorical(p=[0.5, 0.5], scope=1) * Categorical(p=[0.6, 0.4], scope=2)))) \
+ 0.6 * (Categorical(p=[0.2, 0.8], scope=0) * \
Categorical(p=[0.3, 0.7], scope=1) * \
Categorical(p=[0.4, 0.6], scope=2))
return spn
def mix_categorical(weighted_nodes):
assert sum([weight for (weight, node) in weighted_nodes]) == 1
p = np.zeros(len(weighted_nodes[0][1].p))
scope = weighted_nodes[0][1].scope
for (weight, node) in weighted_nodes:
assert isinstance(node, Categorical)
for i in range(len(p)):
p[i] += weight * node.p[i]
return Categorical(p=p, scope=scope)
def test_cuda_categorical():
# Construct a minimal SPN
c1 = Categorical(p=[0.35, 0.55, 0.1], scope=0)
c2 = Categorical(p=[0.25, 0.625, 0.125], scope=1)
c3 = Categorical(p=[0.5, 0.2, 0.3], scope=2)
c4 = Categorical(p=[0.6, 0.15, 0.25], scope=3)
c5 = Categorical(p=[0.7, 0.11, 0.19], scope=4)
c6 = Categorical(p=[0.8, 0.14, 0.06], scope=5)
p = Product(children=[c1, c2, c3, c4, c5, c6])
# Randomly sample input values.
inputs = np.column_stack((
np.random.randint(3, size=30),
np.random.randint(3, size=30),
np.random.randint(3, size=30),
np.random.randint(3, size=30),
np.random.randint(3, size=30),
np.random.randint(3, size=30),
)).astype("float64")
if not CUDACompiler.isAvailable():
print("Test not supported by the compiler installation")
return 0
# Execute the compiled Kernel.
results = CUDACompiler().log_likelihood(p, inputs, supportMarginal=False)
# Compute the reference results using the inference from SPFlow.
reference = log_likelihood(p, inputs)
reference = reference.reshape(30)
# Check the computation results against the reference
# Check in normal space if log-results are not very close to each other.
assert np.all(np.isclose(results, reference)) or np.all(
np.isclose(np.exp(results), np.exp(reference)))
def create_SPN():
from spn.algorithms.Validity import is_valid
from spn.structure.leaves.parametric.Parametric import Categorical
spn = 0.4 * (Categorical(p=[0.2, 0.8], scope=0) * \
(0.3 * (Categorical(p=[0.3, 0.7], scope=1) * Categorical(p=[0.4, 0.6], scope=2)) + \
0.7 * (Categorical(p=[0.5, 0.5], scope=1) * Categorical(p=[0.6, 0.4], scope=2)))) \
+ 0.6 * (Categorical(p=[0.2, 0.8], scope=0) * \
Categorical(p=[0.3, 0.7], scope=1) * \
Categorical(p=[0.4, 0.6], scope=2))
assert is_valid(spn)
return spn
def get_credit_spn():
from spn.structure.Base import Product
from spn.structure.leaves.parametric.Parametric import Categorical
spn1 = Categorical(p=[0.0, 1.0], scope=[2]) * Categorical(p=[0.5, 0.5],
scope=[3])
spn2 = Categorical(p=[1.0, 0.0], scope=[2]) * Categorical(p=[0.1, 0.9],
scope=[3])
spn3 = 0.3 * spn1 + 0.7 * spn2
spn4 = Categorical(p=[0.0, 1.0], scope=[1]) * spn3
spn6 = Product([
Categorical(p=[1.0, 0.0], scope=[1]),
Categorical(p=[0.0, 1.0], scope=[2]),
Categorical(p=[1.0, 0.0], scope=[3])
])
spn6.scope = [1, 2, 3]
spn7 = 0.8 * spn4 + 0.2 * spn6
spn = spn7 * Categorical(p=[0.2, 0.8], scope=[0])
spn.scope = sorted(spn.scope)
return spn
def __init__(self):
p0 = Product(children=[
Categorical(p=[0.3, 0.7], scope=1),
Categorical(p=[0.4, 0.6], scope=2)
])
p1 = Product(children=[
Categorical(p=[0.5, 0.5], scope=1),
Categorical(p=[0.6, 0.4], scope=2)
])
s1 = Sum(weights=[0.3, 0.7], children=[p0, p1])
p2 = Product(children=[Categorical(p=[0.2, 0.8], scope=0), s1])
p3 = Product(children=[
Categorical(p=[0.2, 0.8], scope=0),
Categorical(p=[0.3, 0.7], scope=1)
])
p4 = Product(children=[p3, Categorical(p=[0.4, 0.6], scope=2)])
self.spn = Sum(weights=[0.4, 0.6], children=[p2, p4])
assign_ids(self.spn)
rebuild_scopes_bottom_up(self.spn)
def test_spn_to_torch(self):
# SPFLow implementation
n0 = Gaussian(mean=0.0, stdev=1.0, scope=0)
n1 = Categorical(p=[0.1, 0.3, 0.6])
n2 = Sum(weights=[0.1, 0.2, 0.3, 0.4], children=[n0, n1])
n3 = Product(children=[n0, n1])
torch_n0 = GaussianNode.from_spn(n0)
torch_n1 = CategoricalNode.from_spn(n1)
torch_n2 = SumNode.from_spn(n2)
torch_n3 = ProductNode.from_spn(n3)
self.assertEqual(torch_n0.mean, n0.mean)
self.assertEqual(torch_n0.std, n0.stdev)
self.assertTrue(
np.isclose(torch_n1.p.detach().numpy(), n1.p, atol=DELTA).all())
self.assertTrue(
np.isclose(torch_n2.weights.detach().numpy(),
n2.weights,
atol=DELTA).all())
def test_cpu_categorical():
# Construct a minimal SPN
c1 = Categorical(p=[0.35, 0.55, 0.1], scope=0)
c2 = Categorical(p=[0.25, 0.625, 0.125], scope=1)
c3 = Categorical(p=[0.5, 0.2, 0.3], scope=2)
c4 = Categorical(p=[0.6, 0.15, 0.25], scope=3)
c5 = Categorical(p=[0.7, 0.11, 0.19], scope=4)
c6 = Categorical(p=[0.8, 0.14, 0.06], scope=5)
p = Product(children=[c1, c2, c3, c4, c5, c6])
# Randomly sample input values.
inputs = np.column_stack((
np.random.randint(3, size=30),
np.random.randint(3, size=30),
np.random.randint(3, size=30),
np.random.randint(3, size=30),
np.random.randint(3, size=30),
np.random.randint(3, size=30),
)).astype("float64")
# Insert some NaN in random places into the input data.
inputs.ravel()[np.random.choice(inputs.size, 10, replace=False)] = np.nan
if not CPUCompiler.isVectorizationSupported():
print("Test not supported by the compiler installation")
return 0
# Execute the compiled Kernel.
results = CPUCompiler(computeInLogSpace=False,
vectorize=False).log_likelihood(p,
inputs,
batchSize=10)
# Compute the reference results using the inference from SPFlow.
reference = log_likelihood(p, inputs)
reference = reference.reshape(30)
# Check the computation results against the reference
# Check in normal space if log-results are not very close to each other.
assert np.all(np.isclose(results, reference)) or np.all(
np.isclose(np.exp(results), np.exp(reference)))
def test_vector_slp_escaping_users():
g0 = Gaussian(mean=0.00, stdev=1, scope=0)
g1 = Gaussian(mean=0.01, stdev=0.75, scope=1)
g2 = Gaussian(mean=0.02, stdev=0.5, scope=2)
g3 = Gaussian(mean=0.03, stdev=0.25, scope=3)
g4 = Gaussian(mean=0.04, stdev=1, scope=4)
g5 = Gaussian(mean=0.05, stdev=0.25, scope=5)
g6 = Gaussian(mean=0.06, stdev=0.5, scope=6)
g7 = Gaussian(mean=0.07, stdev=0.75, scope=7)
g8 = Gaussian(mean=0.08, stdev=1, scope=8)
g9 = Gaussian(mean=0.09, stdev=0.75, scope=9)
g10 = Gaussian(mean=0.10, stdev=1, scope=10)
g11 = Gaussian(mean=0.11, stdev=1, scope=11)
h0 = Histogram([0., 1., 2.], [0.1, 0.9], [1, 1], scope=12)
h1 = Histogram([0., 1., 2.], [0.2, 0.8], [1, 1], scope=13)
h2 = Histogram([0., 1., 2.], [0.3, 0.7], [1, 1], scope=14)
h3 = Histogram([0., 1., 2.], [0.4, 0.6], [1, 1], scope=15)
h4 = Histogram([0., 1., 2.], [0.5, 0.5], [1, 1], scope=16)
h5 = Histogram([0., 1., 2.], [0.6, 0.4], [1, 1], scope=17)
h6 = Histogram([0., 1., 2.], [0.7, 0.3], [1, 1], scope=18)
h7 = Histogram([0., 1., 2.], [0.8, 0.2], [1, 1], scope=19)
c0 = Categorical(p=[0.1, 0.1, 0.8], scope=20)
c1 = Categorical(p=[0.2, 0.2, 0.6], scope=21)
c2 = Categorical(p=[0.3, 0.3, 0.4], scope=22)
c3 = Categorical(p=[0.4, 0.4, 0.2], scope=23)
c4 = Categorical(p=[0.5, 0.4, 0.1], scope=24)
c5 = Categorical(p=[0.6, 0.3, 0.1], scope=25)
c6 = Categorical(p=[0.7, 0.2, 0.1], scope=26)
c7 = Categorical(p=[0.8, 0.1, 0.1], scope=27)
s0 = Sum(children=[g8, h4], weights=[0.5, 0.5])
s1 = Sum(children=[g9, h5], weights=[0.5, 0.5])
s2 = Sum(children=[g10, c6], weights=[0.5, 0.5])
s3 = Sum(children=[g11, h7], weights=[0.5, 0.5])
s4 = Sum(children=[s0, c4], weights=[0.5, 0.5])
s5 = Sum(children=[s1, c5], weights=[0.5, 0.5])
s6 = Sum(children=[s2, g6], weights=[0.5, 0.5])
s7 = Sum(children=[s3, c7], weights=[0.5, 0.5])
s8 = Sum(children=[s4, g4], weights=[0.5, 0.5])
s9 = Sum(children=[s5, g5], weights=[0.5, 0.5])
s10 = Sum(children=[s6, h6], weights=[0.5, 0.5])
s11 = Sum(children=[s7, g7], weights=[0.5, 0.5])
p0 = Product(children=[h0, s8])
p1 = Product(children=[c1, s9])
p2 = Product(children=[c2, s10])
p3 = Product(children=[g3, s11])
p4 = Product(children=[p0, g0])
p5 = Product(children=[p1, g1])
p6 = Product(children=[p2, h2])
p7 = Product(children=[p3, c3])
p8 = Product(children=[p4, c0])
p9 = Product(children=[p5, h1])
p10 = Product(children=[p6, g2])
p11 = Product(children=[p7, h3])
s12 = Sum(children=[p8, p9], weights=[0.5, 0.5])
s13 = Sum(children=[p10, p11], weights=[0.5, 0.5])
s14 = Sum(children=[s12, p2], weights=[0.5, 0.5])
s15 = Sum(children=[s13, s2], weights=[0.5, 0.5])
spn = Product(children=[s14, s15])
# Randomly sample input values from Gaussian (normal) distributions.
num_samples = 100
inputs = np.column_stack((
# gaussian
np.random.normal(loc=0.5, scale=1, size=num_samples),
np.random.normal(loc=0.125, scale=0.25, size=num_samples),
np.random.normal(loc=0.345, scale=0.24, size=num_samples),
np.random.normal(loc=0.456, scale=0.1, size=num_samples),
np.random.normal(loc=0.94, scale=0.48, size=num_samples),
np.random.normal(loc=0.56, scale=0.42, size=num_samples),
np.random.normal(loc=0.76, scale=0.14, size=num_samples),
np.random.normal(loc=0.32, scale=0.58, size=num_samples),
np.random.normal(loc=0.58, scale=0.219, size=num_samples),
np.random.normal(loc=0.14, scale=0.52, size=num_samples),
np.random.normal(loc=0.24, scale=0.42, size=num_samples),
np.random.normal(loc=0.34, scale=0.1, size=num_samples),
# histogram
np.random.randint(low=0, high=2, size=num_samples),
np.random.randint(low=0, high=2, size=num_samples),
np.random.randint(low=0, high=2, size=num_samples),
np.random.randint(low=0, high=2, size=num_samples),
np.random.randint(low=0, high=2, size=num_samples),
np.random.randint(low=0, high=2, size=num_samples),
np.random.randint(low=0, high=2, size=num_samples),
np.random.randint(low=0, high=2, size=num_samples),
# categorical
np.random.randint(low=0, high=3, size=num_samples),
np.random.randint(low=0, high=3, size=num_samples),
np.random.randint(low=0, high=3, size=num_samples),
np.random.randint(low=0, high=3, size=num_samples),
np.random.randint(low=0, high=3, size=num_samples),
np.random.randint(low=0, high=3, size=num_samples),
np.random.randint(low=0, high=3, size=num_samples),
np.random.randint(low=0, high=3, size=num_samples))).astype("float64")
# Compute the reference results using the inference from SPFlow.
reference = log_likelihood(spn, inputs)
reference = reference.reshape(num_samples)
# Compile the kernel with batch size 1 to enable SLP vectorization.
compiler = CPUCompiler(vectorize=True,
computeInLogSpace=True,
vectorLibrary="LIBMVEC")
kernel = compiler.compile_ll(spn=spn, batchSize=1, supportMarginal=False)
# Execute the compiled Kernel.
time_sum = 0
for i in range(len(reference)):
# Check the computation results against the reference
start = time.time()
result = compiler.execute(kernel, inputs=np.array([inputs[i]]))
time_sum = time_sum + time.time() - start
print(
f"evaluation #{i}: result: {result[0]:16.8f}, reference: {reference[i]:16.8f}",
end='\r')
if not np.isclose(result, reference[i]):
print(
f"\nevaluation #{i} failed: result: {result[0]:16.8f}, reference: {reference[i]:16.8f}"
)
raise AssertionError()
print(f"\nExecution of {len(reference)} samples took {time_sum} seconds.")
y = np.array(y).reshape(-1, )
z = np.random.choice([1, 2], int(1e4), replace=True, p=[0.4, 0.6])
df = pd.DataFrame(dict(zip(['X', 'Y', 'Z'], [x, y, z]))).astype(str)
df, vd, pars = fn.transform_dataset(df)
spn = spn_handler.load_or_create_spn(df,
vd,
pars,
'mini_example',
0.4,
0.5,
nrows=None,
seed=1,
force_create=True,
clustering='km_rule_clustering')
spn = spn.children[1]
manspn = ( 0.3 * (Categorical(p=[0.9, 0.1], scope=0) * Categorical(p=[0.55, 0.4, 0.05], scope=1))
+ 0.7 * (Categorical(p=[0., 1.], scope=0) * Categorical(p=[0.1, 0.2, 0.7], scope=1)) ) \
* (Categorical(p=[0.4, 0.6], scope=2))
# plot leaves from example
p = [[0.9, 0.1], [0.4, 0.55, 0.05], [0., 1.], [0.1, 0.2, 0.7], [0.4, 0.6]]
y = 2
size = (2.88 * y, y)
fig, axes = plt.subplots(1, 4, sharey=True, squeeze=True, figsize=size)
for i, var in enumerate(['X', 'Y', 'X', 'Y']):
currp = p[i]
ax = axes[i]
# if i in [1,2]:
# d = df[var].value_counts(sort=False).divide(len(df))
# if i in [3,4]:
ticks = list(range(len(currp)))
#
# geometric
geometric = Geometric(p=.025, scope=[0])
pdf_x, pdf_y = approximate_density(geometric, x_range)
fig, ax = plt.subplots(1, 1)
ax.plot(pdf_x, pdf_y, label="geometric")
print('Geometric Mode:', geometric.mode)
plt.axvline(x=geometric.mode, color='r')
if show_plots:
plt.show()
#
# categorical
categorical = Categorical(p=[0.1, 0.05, 0.3, 0.05, 0.2, 0.2, 0.1],
scope=[0])
pdf_x, pdf_y = approximate_density(categorical, np.arange(categorical.k))
fig, ax = plt.subplots(1, 1)
ax.plot(pdf_x, pdf_y, label="categorical")
print('Categorical Mode:', categorical.mode)
plt.axvline(x=categorical.mode, color='r')
if show_plots:
plt.show()
#
# exponential
exponential = Exponential(l=5, scope=[0])
pdf_x, pdf_y = approximate_density(exponential, x_range)
fig, ax = plt.subplots(1, 1)
"""
=================================
Domain Specific Language for SPNs
=================================
We start by creating an SPN. Using a Domain-Specific Language (DSL), we can
quickly create an SPN of categorical leave nodes like this:
"""
from spn.structure.leaves.parametric.Parametric import Categorical
from spn.io.Graphics import draw_spn
import matplotlib.pyplot as plt
spn = 0.4 * (
Categorical(p=[0.2, 0.8], scope=0) *
(0.3 *
(Categorical(p=[0.3, 0.7], scope=1) * Categorical(p=[0.4, 0.6], scope=2))
+ 0.7 *
(Categorical(p=[0.5, 0.5], scope=1) * Categorical(p=[0.6, 0.4], scope=2)))
) + 0.6 * (Categorical(p=[0.2, 0.8], scope=0) * Categorical(
p=[0.3, 0.7], scope=1) * Categorical(p=[0.4, 0.6], scope=2))
ax = draw_spn(spn)
from spn.structure.leaves.parametric.Parametric import Categorical
from spn.structure.Base import Sum, Product
from spn.structure.Base import assign_ids, rebuild_scopes_bottom_up
p0 = Product(children=[
Categorical(p=[0.3, 0.7], scope=1),
Categorical(p=[0.4, 0.6], scope=2)
])
p1 = Product(children=[
Categorical(p=[0.5, 0.5], scope=1),
Categorical(p=[0.6, 0.4], scope=2)
])
s1 = Sum(weights=[0.3, 0.7], children=[p0, p1])
p2 = Product(children=[Categorical(p=[0.2, 0.8], scope=0), s1])
p3 = Product(children=[
Categorical(p=[0.2, 0.8], scope=0),
Categorical(p=[0.3, 0.7], scope=1)
])
p4 = Product(children=[p3, Categorical(p=[0.4, 0.6], scope=2)])
spn = Sum(weights=[0.4, 0.6], children=[p2, p4])
assign_ids(spn)
rebuild_scopes_bottom_up(spn)
import numpy as np
test_data = np.array([1.0, 0.0, 1.0]).reshape(-1, 3)
from spn.algorithms.Inference import log_likelihood
ll = log_likelihood(spn, test_data)
print(ll, np.exp(ll))
def test_entropy(self):
# test if entropy is correct
"""
# explain how training data and the spn comes
# number of RVs
M = 3
# table of probabilities
p1 = 0.6
p2 = 0.3
p31 = 0.1
p32 = 0.9
# generate x1 and x2
x1 = np.random.binomial(1, p1, size=N) + np.random.binomial(1, p1, size=N)
x2 = np.random.binomial(1, p2, size=N)
x3 = np.zeros(N)
# generate x3
for i in range(N):
if x2[i] == 1:
x3[i] = np.random.binomial(1, p31, size=1)
else:
x3[i] = np.random.binomial(1, p32, size=1)
# form a matrix, rows are instances and columns are RVs
train_data = np.concatenate((x1, x2, x3)).reshape((M, N)).transpose()
"""
# only for generating the ds_context
train_data = np.array([[0.0, 1.0, 0.0], [1.0, 0.0, 1.0], [2.0, 0.0, 1.0]])
# spn
ds_context = Context(meta_types=[MetaType.DISCRETE] * 3)
ds_context.add_domains(train_data)
ds_context.parametric_type = [Categorical] * 3
spn = 0.64 * (
(
Categorical(p=[0.25, 0.75, 0.0], scope=0)
* (
0.34 * ((Categorical(p=[7 / 34, 27 / 34], scope=1) * Categorical(p=[1.0, 0.0], scope=2)))
+ 0.66 * ((Categorical(p=[21 / 22, 1 / 22], scope=1) * Categorical(p=[0.0, 1.0], scope=2)))
)
)
) + 0.36 * (
(
Categorical(p=[0.0, 0.0, 1.0], scope=0)
* (
0.34 * ((Categorical(p=[7 / 34, 27 / 34], scope=1) * Categorical(p=[1.0, 0.0], scope=2)))
+ 0.66 * ((Categorical(p=[21 / 22, 1 / 22], scope=1) * Categorical(p=[0.0, 1.0], scope=2)))
)
)
)
# real entropy
p2 = 0.3
h_x2 = -p2 * np.log(p2) - (1 - p2) * np.log(1 - p2)
self.assertAlmostEqual(h_x2, entropy(spn, ds_context, {1}))
h_x2x3 = -(p2 * np.log(p2) + (1 - p2) * np.log(1 - p2) + 0.9 * np.log(0.9) + 0.1 * np.log(0.1))
self.assertAlmostEqual(h_x2x3, entropy(spn, ds_context, {1, 2}))
h_x1 = -(0.16 * np.log(0.16) + 0.36 * np.log(0.36) + 0.48 * np.log(0.48))
self.assertAlmostEqual(h_x1, entropy(spn, ds_context, {0}))
h_x2x1 = -(0.7 * np.log(0.7) + 0.3 * np.log(0.3)) + h_x1
self.assertAlmostEqual(h_x2x1, entropy(spn, ds_context, {1, 0}))
h_x3x1 = -(0.66 * np.log(0.66) + 0.34 * np.log(0.34)) + h_x1
self.assertAlmostEqual(h_x3x1, entropy(spn, ds_context, {2, 0}))
h_x2x3x1 = h_x1 + h_x2x3
self.assertAlmostEqual(h_x2x3x1, entropy(spn, ds_context, {1, 2, 0}))
# test symmetry
self.assertAlmostEqual(entropy(spn, ds_context, {0, 2}), entropy(spn, ds_context, {2, 0}))
self.assertAlmostEqual(entropy(spn, ds_context, {1, 2}), entropy(spn, ds_context, {2, 1}))
def _deserialize_categorical(self, node, node_map):
probabilities = node.categorical.probabilities
cat = Categorical(p=probabilities, scope=node.categorical.scope)
cat.id = node.id
return cat
node = Gaussian(np.inf, np.inf)
data = np.array([1, 2, 3, 4, 5]).reshape(-1, 1)
update_parametric_parameters_mle(node, data)
assert np.isclose(node.mean, np.mean(data))
assert np.isclose(node.stdev, np.std(data))
node = Gamma(np.inf, np.inf)
data = np.array([1, 2, 3, 4, 5]).reshape(-1, 1)
update_parametric_parameters_mle(node, data)
assert np.isclose(node.alpha / node.beta, np.mean(data)), node.alpha
node = LogNormal(np.inf, np.inf)
data = np.array([1, 2, 3, 4, 5]).reshape(-1, 1)
update_parametric_parameters_mle(node, data)
assert np.isclose(node.mean, np.log(data).mean(), atol=0.00001)
assert np.isclose(node.stdev, np.log(data).std(), atol=0.00001)
node = Poisson(np.inf)
data = np.array([1, 2, 3, 4, 5]).reshape(-1, 1)
update_parametric_parameters_mle(node, data)
assert np.isclose(node.mean, np.mean(data))
node = Categorical(np.array([1, 1, 1, 1, 1, 1]) / 6)
data = np.array([0, 0, 1, 3, 5]).reshape(-1, 1)
update_parametric_parameters_mle(node, data)
assert np.isclose(node.p[0], 2 / 5)
assert np.isclose(node.p[1], 1 / 5)
assert np.isclose(node.p[2], 0)
assert np.isclose(node.p[3], 1 / 5)
assert np.isclose(node.p[4], 0)
from spn.structure.leaves.parametric.Parametric import Categorical from spn.structure.leaves.parametric.SamplingRange import sample_categorical_node from spn.structure.leaves.parametric.InferenceRange import categorical_likelihood_range from spn.structure.Base import Context from spn.structure.StatisticalTypes import MetaType from spn.experiments.AQP.Ranges import NominalRange, NumericRange from spn.algorithms import SamplingRange rand_gen = np.random.RandomState(100) #Create SPN node1 = Categorical(p=[0.9, 0.1], scope=[0]) node2 = Categorical(p=[0.1, 0.9], scope=[0]) x = [0., 1., 2., 3., 4.] y = [0., 10., 0., 0., 0.] x, y = np.array(x), np.array(y) auc = np.trapz(y, x) y = y / auc node3 = PiecewiseLinear(x_range=x, y_range=y, bin_repr_points=x[1:-1], scope=[1]) x = [0., 1., 2., 3., 4.] y = [0., 0., 0., 10., 0.] x, y = np.array(x), np.array(y) auc = np.trapz(y, x) y = y / auc
pass
if __name__ == '__main__':
from spn.structure.Base import Sum, Product, Leaf
from spn.structure.leaves.parametric.Parametric import Categorical
spn1 = Categorical(p=[0.0, 1.0], scope=[2]) * Categorical(p=[0.5, 0.5], scope=[3])
spn2 = Categorical(p=[1.0, 0.0], scope=[2]) * Categorical(p=[0.1, 0.9], scope=[3])
spn3 = 0.3 * spn1 + 0.7 * spn2
spn4 = Categorical(p=[0.0, 1.0], scope=[1]) * spn3
spn6 = Product([Categorical(p=[1.0, 0.0], scope=[1]), Categorical(p=[0.0, 1.0], scope=[2]), Categorical(p=[1.0, 0.0], scope=[3])])
spn6.scope = [1,2,3]
spn7 = 0.8 * spn4 + 0.2 * spn6
spn = spn7 * Categorical(p=[0.2, 0.8], scope=[0])
#spn_util.plot_spn(spn, "rule_spn.pdf")
extract_rules(spn)
