def test_observed():
with tp.Model() as model:
mu = tp.Parameter()
sigma = tp.Parameter()
X = tp.Normal(mu, sigma)
model.observed(X)
assert(X in model._observed)
def test_observed_in_model():
model = tp.Model()
with model:
mu = tp.Parameter(lower=-5, upper=5)
sigma = tp.Parameter(lower=0)
X = tp.Normal(mu, sigma)
model.observed(X)
def test_initialise_with_list():
with tp.Model() as model:
mu = tp.Parameter()
sigma = tp.Parameter(lower=0)
X = tp.Normal(mu, sigma)
model.observed(X)
model.initialize([5, 2])
def test_assign_empty_dict():
model = tp.Model()
with model:
mu = tp.Parameter(lower=-5, upper=5)
sigma = tp.Parameter(lower=0)
X = tp.Normal(mu, sigma)
model.observed(X)
model.assign({})
def test_assign_wrong_container():
model = tp.Model()
with model:
mu = tp.Parameter(lower=-5, upper=5)
sigma = tp.Parameter(lower=0)
X = tp.Normal(mu, sigma)
model.observed(X)
model.assign([1, 2, 3])
def test_assign_in_before_observed():
model = tp.Model()
with model:
mu = tp.Parameter(lower=-5, upper=5)
sigma = tp.Parameter(lower=0)
tp.Normal(mu, sigma)
model.assign({mu: 1, sigma: 0})
def test_variable_independent():
with tp.Model() as model:
a = tp.Parameter()
b = tp.Parameter()
model.observed(b)
model.initialize({
a: 1,
})
model.fit([1, 2, 3])
def test_initalize_in_model_block():
with tp.Model() as model:
mu = tp.Parameter()
sigma = tp.Parameter(lower=0)
tp.Normal(mu, sigma)
model.initialize({
mu: 5,
sigma: 2
})
def test_initialise_before_observed():
with tp.Model() as model:
mu = tp.Parameter()
sigma = tp.Parameter(lower=0)
tp.Normal(mu, sigma)
model.initialize({
mu: 5,
sigma: 2
})
def test_initialise_with_too_few_variables():
with tp.Model() as model:
mu = tp.Parameter()
sigma = tp.Parameter(lower=0)
X = tp.Normal(mu, sigma)
model.observed(X)
model.initialize({
mu: 5
})
def test_assign_in_model_block():
model = tp.Model()
with model:
mu = tp.Parameter(lower=-5, upper=5)
sigma = tp.Parameter(lower=0)
X = tp.Normal(mu, sigma)
model.assign({mu: 1, sigma: 0})
model.observed(X)
model.initialize({mu: 42, sigma: 1})
def test_initialise_with_invalid_keys():
with tp.Model() as model:
mu = tp.Parameter()
sigma = tp.Parameter(lower=0)
X = tp.Normal(mu, sigma)
model.observed(X)
model.initialize({
'Parameter_1': 5,
'Parameter_2': 2
})
def test_fit():
model = tp.Model()
with model:
mu = tp.Parameter(lower=-5, upper=5)
sigma = tp.Parameter(lower=0)
X = tp.Normal(mu, sigma)
model.observed(X)
model.initialize({mu: 2, sigma: 2})
np.random.seed(0)
data = np.random.normal(0, 1, 100)
results = model.fit(data)
assert results.success
def test_nll():
with tp.Model() as model:
mu = tp.Parameter()
sigma = tp.Parameter()
X = tp.Normal(mu, sigma)
model.observed(X)
xs = np.linspace(-5, 5, 100)
model.initialize({
mu: 0,
sigma: 1
})
out1 = model.nll(xs)
out2 = -sum(st.norm.logpdf(xs, 0, 1))
assert_almost_equal(out1, out2, 10)
def test_assign():
model = tp.Model()
with model:
mu = tp.Parameter(lower=-5, upper=5)
sigma = tp.Parameter(lower=0)
X = tp.Normal(mu, sigma)
model.observed(X)
feed = {mu: 42, sigma: 1}
model.initialize(feed)
model.assign({mu: 1, sigma: 0})
model.assign({mu: -100, sigma: 0})
assert model.state == {mu: -100, sigma: 0}
model.assign(feed)
assert model.state == feed
def test_distribution_creation_global_graph():
# Distribution creation doesn't modify the global graph
before = tf.get_default_graph().as_graph_def()
with tp.Model():
tp.Parameter()
after = tf.get_default_graph().as_graph_def()
assert before == after
def test_pdf():
with tp.Model() as m:
dummy = tp.Parameter()
X = tp.Uniform(lower=dummy, upper=1)
m.observed(X)
m.initialize({dummy: -2})
xs = np.linspace(-1, 1, 1)
out1 = st.uniform.pdf(xs, -2, 3)
out2 = m.pdf(xs)
assert_array_almost_equal(out1, out2, 16)
def test_internal_graph_no_growth():
# Calling assign or fit doesn't grow the execution graph
with tp.Model() as model:
mu = tp.Parameter()
sigma = tp.Parameter(lower=0)
X = tp.Normal(mu, sigma)
model.observed(X)
model.initialize({
mu: 1,
sigma: 1,
})
before = model.session.graph_def
model.assign({
mu: 0,
sigma: 2,
})
model.fit([1, 2, 3])
after = model.session.graph_def
assert before == after
def make_normal():
mu = tp.Parameter(name='mu')
sigma = tp.Parameter(name='sigma', lower=0)
X = tp.Normal(mu, sigma)
return mu, sigma, X
def test_distribution_creation_outside_with():
with tp.Model():
mu = tp.Parameter(name='mu')
sigma = tp.Parameter(name='sigma', lower=0)
tp.Normal(mu, sigma)
def test_scalar_creation_outside_with():
tp.Parameter(name='mu')
def test_observed_erorr_on_non_distribution():
with tp.Model() as model:
mu = tp.Parameter()
sigma = tp.Parameter()
X = tp.Normal(mu, sigma)
model.observed(X, int(42))
def test_prepare_without_observed():
with tp.Model() as model:
mu = tp.Parameter()
model.initialize({mu: 42})
def test_observed_in_model_block():
with tp.Model():
mu = tp.Parameter()
sigma = tp.Parameter(lower=0)
X = tp.Normal(mu, sigma)
tp.observed(X)
