def model_exposed():
return ExposedSumSPE(
spe_weights=(Nationality >> choice({
'India': 0.5,
'USA': 0.5
})),
children={
# American student.
'USA':
ExposedSumSPE(spe_weights=(Perfect >> choice({
'True': 0.01,
'False': 0.99
})),
children={
'False': GPA >> uniform(loc=0, scale=4),
'True': GPA >> atomic(loc=4),
}),
# Indian student.
'India':
ExposedSumSPE(spe_weights=(Perfect >> choice({
'True': 0.01,
'False': 0.99
})),
children={
'False': GPA >> uniform(loc=0, scale=10),
'True': GPA >> atomic(loc=10),
})
},
)
def model_perfect_nested():
Nationality = Id('Nationality')
Perfect = Id('Perfect')
GPA = Id('GPA')
command = Sequence(
Sample(Nationality, choice({
'India': 0.5,
'USA': 0.5
})),
IfElse(
Nationality << {'India'},
Sequence(
Sample(Perfect, choice({
'True': 0.01,
'False': 0.99
})),
IfElse(Perfect << {'True'}, Sample(GPA, atomic(loc=10)), True,
Sample(GPA, uniform(scale=10)))),
Nationality << {'USA'},
Sequence(
Sample(Perfect, choice({
'True': 0.01,
'False': 0.99
})),
IfElse(
Perfect << {'True'},
Sample(GPA, atomic(loc=4)),
True,
Sample(GPA, uniform(scale=4)),
))))
return command.interpret()
def model_no_latents():
return \
0.5 * ( # American student
0.99 * (GPA >> uniform(loc=0, scale=4)) | \
0.01 * (GPA >> atomic(loc=4))) | \
0.5 * ( # Indian student
0.99 * (GPA >> uniform(loc=0, scale=10)) | \
0.01 * (GPA >> atomic(loc=10)))
def test_logpdf_lexicographic_both():
spe = .75*(X >> norm() & Y >> atomic(loc=0) & Z >> discrete({1:.2, 2:.8})) \
| .25*(X >> discrete({1:.5, 2:.5}) & Y >> norm() & Z >> atomic(loc=2))
# Lexicographic, Mix
assignment = {X:1, Y:0, Z:2}
assert allclose(
spe.logpdf(assignment),
logsumexp([
log(.75) + norm().dist.logpdf(1) + log(1) + log(.8),
log(.25) + log(.5) + norm().dist.logpdf(0) + log(1)]))
assert isinstance(spe.constrain(assignment), SumSPE)
def test_logpdf_lexicographic_either():
spe = .75*(X >> norm() & Y >> atomic(loc=0) & Z >> discrete({1:.1, 2:.9})) \
| .25*(X >> atomic(loc=0) & Y >> norm() & Z >> norm())
# Lexicographic, Branch 1
assignment = {X:0, Y:0, Z:2}
assert allclose(
spe.logpdf(assignment),
log(.75) + norm().dist.logpdf(0) + log(1) + log(.9))
assert isinstance(spe.constrain(assignment), ProductSPE)
# Lexicographic, Branch 2
assignment = {X:0, Y:0, Z:0}
assert allclose(
spe.logpdf(assignment),
log(.25) + log(1) + norm().dist.logpdf(0) + norm().dist.logpdf(0))
assert isinstance(spe.constrain(assignment), ProductSPE)
def model_ifelse_exhuastive():
command = Sequence(
Sample(Nationality, choice({
'India': 0.5,
'USA': 0.5
})), Sample(Perfect, choice({
'True': 0.01,
'False': 0.99
})),
IfElse((Nationality << {'India'}) & (Perfect << {'False'}),
Sample(GPA, uniform(loc=0, scale=10)),
(Nationality << {'India'}) & (Perfect << {'True'}),
Sample(GPA, atomic(loc=10)),
(Nationality << {'USA'}) & (Perfect << {'False'}),
Sample(GPA, uniform(loc=0, scale=4)),
(Nationality << {'USA'}) & (Perfect << {'True'}),
Sample(GPA, atomic(loc=4))))
return command.interpret()