def test_rvs_backend_arithmetic(tf_seed):
"""Test backend arithmetic implemented by the `WithBackendArithmetic` class."""
x = pm.Normal("NormA", mu=0, sigma=1)
y = pm.Normal("NormB", mu=1, sigma=2)
assert x + y is not None
assert x - y is not None
assert x * y is not None
# TODO test __matmul__ once random variables support shapes.
# assert x @ y is not None
assert x / y is not None
assert x // y is not None
assert x % y is not None
assert x**y is not None
assert -x is not None
def test_wrong_distribution_argument_in_list_fails():
with pytest.raises(TypeError, match=r"every element in 'distribution' "):
pm.Mixture(
"mix",
p=[0.5, 0.5],
distributions=[pm.Normal("comp1", loc=0.0, scale=1.0), "not a distribution"],
)
def model_with_yields_and_other_assigns():
N = 10
N += 10
x = yield pm.Normal(loc=0, scale=1)
s: str = "A test string."
y = yield pm.HalfCauchy(scale=2)
a, b, c = (1, 2, 3)
z = yield pm.Binomial(total_count=10, probs=0.2)
def model():
x = yield pm.Normal(
"x",
np.zeros(n_obs, dtype="float32"),
1,
observed=np.zeros(n_obs, dtype="float32"),
conditionally_independent=True,
reinterpreted_batch_ndims=1,
)
beta = yield pm.Normal("beta", 0, 1, conditionally_independent=True)
bias = yield pm.Normal("bias", 0, 1, conditionally_independent=True)
mu = beta[..., None] * x + bias[..., None]
yield pm.Normal("obs",
mu,
1,
observed=np.ones(n_obs, dtype="float32"),
reinterpreted_batch_ndims=1)
def model():
mu = yield pm.Normal(
"mu",
tf.zeros(4),
1,
conditionally_independent=True,
reinterpreted_batch_ndims=1,
)
scale = yield pm.HalfNormal("scale",
1,
conditionally_independent=True)
x = yield pm.Normal(
"x",
mu,
scale[..., None],
observed=observed,
reinterpreted_batch_ndims=1,
event_stack=5,
)
def model():
beta = yield pm.Normal(
"beta",
tf.zeros((n_features, )),
1,
conditionally_independent=True,
reinterpreted_batch_ndims=1,
)
bias = yield pm.Normal("bias",
0,
1,
conditionally_independent=True)
scale = yield pm.HalfNormal("scale",
1,
conditionally_independent=True)
mu = tf.linalg.matvec(regressors, beta) + bias[..., None]
y = yield pm.Normal(
"y",
mu,
scale[..., None],
observed=observed,
reinterpreted_batch_ndims=1,
)
def model():
yield pm.Normal("x", 0, 1)
def test_as_sampling_state_failure_on_dangling_distribution():
st = pm.flow.executor.SamplingState(
distributions={"bla": pm.Normal("bla", 0, 1)})
with pytest.raises(TypeError):
st.as_sampling_state()
def test_initialized_distribution_cant_be_transformed_into_a_new_prior():
with pytest.raises(TypeError) as e:
pm.Normal("m", 0, 1).prior("n")
assert e.match("already not anonymous")
def unvectorized_model():
norm = yield pm.Normal("norm", 0, 1, batch_stack=norm_shape)
determ = yield pm.Deterministic("determ", tf.reduce_max(norm))
output = yield pm.Normal("output", determ, 1, observed=observed)
def model(df):
"""Create I_0 with
shape:
number_of_countries*number_of_age_groups
"""
batch_shape = len(countries) * len(age_groups)
I_0 = yield pm.HalfCauchy(loc=[10] * batch_shape, name="I_0")
""" Create Contact number matrix
with LKJ prior and
shape:
[
number_of_age_groups
x
number_of_age_groups
]
x
number_of_countries
"""
batch_shape = len(countries)
C = yield pm.LKJ(
dimension=len(age_groups),
concentration=[2] * batch_shape, # eta
name="Contact_matrix",
)
""" Create Reproduction number matrix,
should be diagonal matrix
shape:
number_of_age_groups
x
number_of_age_groups
"""
batch_shape = len(age_groups) # *time?
# Create RV with shape number_of_age_groups and convert to diag matrix
R = yield pm.Normal(loc=[2] * batch_shape, scale=2.5, name="R_age_groups")
R_diag = tf.linalg.diag(R)
""" Create generation interval RV
see function documentation for more informations
"""
g = covid19_npis.model._construct_generation_interval_gamma()
"""
Get RV for new_cases from SIR model
it should have the shape:
(data_end-data_begin).days+fcast
x
number_of_countries*number_of_age_groups
"""
new_cases = covid19_npis.model.NewCasesModel(I_0=I_0, R=R_diag,
g=g) # TODO
"""
Delay new cases via convolution
"""
# 1. Delay RV
delay = yield pm.Normal(loc=4, scale=3, name="D",
batch_shape=1) # TODO shape
# 2. Do convolution https://www.tensorflow.org/probability/api_docs/python/tfp/experimental/nn/Convolution
new_cases_delayed = yield tfp.experimental.nn.Convolution() # TODO
# supprisingly df.to_numpy() gives us the right numpy array i.e. with shape [time,countries*age_groups]
likelihood = pm.NegativeBinomial(new_cases, observed=df.to_numpy())
def model():
a = yield pm.Normal("a", 0, 1)
b = yield pm.HalfNormal("b", 1)
c = yield pm.Normal("c", loc=a, scale=b, plate=len(observed_value))
def model():
x = yield pm.Normal("x", 0, 1)
det = yield pm.Deterministic("det", x)
y = yield pm.Normal("det", det, 1)
return y
def model():
x = yield pm.Normal("x", tf.ones(5), 1, reinterpreted_batch_ndims=1)
y = yield pm.Normal("y", x, 1)
def model():
beta = yield pm.Normal("beta", tf.zeros((n_features, )), 1)
bias = yield pm.Normal("bias", 0, 1)
scale = yield pm.HalfNormal("scale", 1)
mu = tf.linalg.matvec(regressors, beta) + bias
y = yield pm.Normal("y", mu, scale, observed=observed)
def model_with_only_good_yields():
x = yield pm.Normal(loc=0, scale=1)
y = yield pm.HalfCauchy(scale=2)
z = yield pm.Binomial(total_count=10, probs=0.2)
def model_with_yield_tuple():
x, y = yield pm.Normal(loc=0, scale=1), pm.HalfCauchy(scale=2)
def model():
x = yield pm.Normal("x", 0, 1)
yield pm.Deterministic(None, x)
def model():
x = yield pm.Normal("x", 0, 1)
det = yield pm.Deterministic("x", x)
return det
def model():
a = yield pm.Normal("a", 0, 1, conditionally_independent=True)
b = yield pm.Normal("b", a, 1)
def model(observed):
a = yield pm.Normal("a", 0, [1, 2], observed=observed)
def model():
mu = yield pm.Normal("mu", prior_mean, prior_sigma)
ll = yield pm.Normal("ll", mu, known_sigma, observed=data)
def create_rvs(**kwargs):
# AST parsers should just add name kwargs to pymc4 RVs
assert "name" not in kwargs.keys()
return pm.Normal(0, 1, name="inside_function")
def invdalid_model():
yield pm.HalfNormal("n", 1, transform=pm.distributions.transforms.Log())
yield pm.Normal("n", 0, 1)
def simple_model():
norm = yield pm.Normal("norm", 0, 1)
return norm
def model():
x = yield pm.Normal("x", tf.ones(5), 1)
y = yield pm.Normal("y", x, 1)
def simple_model_no_free_rvs():
norm = yield pm.Normal("norm", 0, 1, observed=1)
return norm
def model():
sd = yield pm.HalfNormal("sd", 1.0)
mu = yield pm.Deterministic("mu", tf.convert_to_tensor(1.0))
x = yield pm.Normal("x", mu, sd, observed=observed)
y = yield pm.Normal("y", x, 1e-9)
dy = yield pm.Deterministic("dy", 2 * y)
def nested_model(cond):
x = yield pm.Normal("x", cond, 1)
dx = yield pm.Deterministic("dx", x + 1)
return dx
def class_model_method(self):
norm = yield pm.Normal("n", 0, 1)
return norm
