def test_constraints_greater_than_works(self):
"""
Ensures that setting a constraint actually limits the trace
"""
# Database size
N = 10
# Specify distributions
age = pv.Normal("age", mu=55.2, std=3.5, num_elements=N)
# Create dataset. Refer to "age_alice" as "age1" in the trace and "age" as "age2" in the trace. This is the general naming convention.
ds = pv.Dataset(input_specs=[age])
# For now output type can be: Int, Float, List(Float), List(Int)
program = pv.Program("output",
dataset=ds,
output_type=pv.Float,
function=program_alpha)
# Add observations
program.add_observation("57>output>56")
# Call infer and specify program output
trace = pv.infer(program, cores=2, draws=1000)
os.remove("typed.py")
self.assertTrue(
all(57 > (np.array(trace.posterior["output"] > 56).flatten())))
def test_sampling_samples_correct_order(self):
"""
Ensures that when sampling for distribution A and B then output is made of [(a_0,b_0), ..., (a_i, b_i)]
"""
a = pv.Normal("age", mu=10, std=3.5)
b = pv.Normal("height", mu=40, std=3.5)
# Create dataset and specify program output
ds = pv.Dataset(input_specs=[a, b])
prog = pv.Program("output",
dataset=ds,
output_type=pv.Float,
function=program_addition)
#Program
# self.create_file(lambda a,b: a+b)
# Call infer
trace = pv.infer(prog, draws=1000, cores=1)
os.remove("typed.py")
for a, b, o in zip(trace.posterior["age"].values,
trace.posterior["height"].values,
trace.posterior["output"].values):
for i in range(len(a)):
self.assertEqual(a[i] + b[i], o[i])
def test_uniform_cutoff(self):
"""
Ensures that when sampling for continuous uniform, no value exceeds domain
"""
a = pv.Uniform("age", 10, 50)
# Create dataset and specify program output
ds = pv.Dataset(input_specs=[a])
prog = pv.Program("output",
dataset=ds,
output_type=pv.Float,
function=program_identity)
#Program
# self.create_file(lambda a: a)
# Call infer
trace = pv.infer(prog, draws=1000, cores=1)
os.remove("typed.py")
for ai, oi in zip(trace.posterior["age"], trace.posterior["output"]):
for i in range(len(ai)):
self.assertTrue(50 >= ai[i] >= 10)
self.assertTrue(50 >= oi[i] >= 10)
def test_k_samples_gives_k_samples(self):
"""
Ensures that the number of samples given also corresponds to the number of samples returned
"""
sample_size = 1000
# Specify distributions
age = pv.Normal("age", mu=55.2, std=3.5)
# Create dataset and specify program output
ds = pv.Dataset(input_specs=[age])
#Program
prog = pv.Program("output",
dataset=ds,
output_type=pv.Float,
function=program_identity)
# Call infer
trace = pv.infer(prog, draws=sample_size, cores=1, chains=1)
os.remove("typed.py")
self.assertEqual(len(trace.posterior["age"][0]), sample_size)
self.assertEqual(len(trace.posterior["output"][0]), sample_size)