def test_PoissonZeroProcess_point():
test_states = np.r_[0, 0, 1, 1, 0, 1]
with pm.Model():
test_mean = pm.Constant("c", 1000.0)
test_point = {"c": 100.0}
test_sample = PoissonZeroProcess.dist(
test_mean, test_states).random(point=test_point)
assert np.all(0 < test_sample[..., test_states > 0])
assert np.all(test_sample[..., test_states > 0] < 200)
def test_multiplicative_seasonality():
z = np.sin(np.linspace(0, 200, 200)) * np.linspace(0, 200, 200)
df = pd.DataFrame()
df["ds"] = pd.date_range(start="2018-01-01", periods=200)
df["y"] = z
m = PMProphet(df, auto_changepoints=False, growth=True, intercept=False, name="model")
with m.model:
m.priors['growth'] = pm.Constant('growth_model', 1)
m.add_seasonality(seasonality=3.14 * 2, fourier_order=3, mode=Seasonality.MULTIPLICATIVE)
m.fit()
m.predict(60, alpha=0.2, include_history=True, plot=True)
m.plot_components(intercept=False)
def test_pymc_pydantic():
"""
Test that a simple PyMC3 model is successfully serialized/deserialized
when part of a Pydantic Model.
"""
import mackelab_toolbox.typing as mtbtyping
import mackelab_toolbox.theano
mackelab_toolbox.theano.freeze_theano_types()
## Simplistic scalar model – Test basic serialize/deserialize concept
with PyMC_Model() as model:
x = pm.Normal('x', mu=0, sigma=2)
y = pm.Deterministic('y', x**2)
class Foo(BaseModel):
model: PyMC_Model
class Config:
json_encoders = mtbtyping.json_encoders
foo = Foo(model=model)
## Following is useful to inspect the serialized data
# import json
# json.loads(foo.json())
# Smoke test
foo2 = Foo.parse_raw(foo.json())
# Confirm that the deserialized y is bound to the deserialized x
symbolic_inputs = [
v for v in theano.graph.basic.graph_inputs([foo2.model.y])
if isinstance(v, tt.Variable) and not isinstance(v, tt.Constant)
]
assert len(symbolic_inputs) == 1
assert foo2.model.x is symbolic_inputs[0]
assert foo2.model.y.eval({foo2.model.x: 4}) == 16
## Test serialization of more ops: Reduce (Sum), DimShuffle
with PyMC_Model() as model_b:
x = pm.Normal('x', mu=0, sigma=2, shape=(3, 3))
y = pm.Deterministic('y', x**2)
z = pm.Deterministic('z', y.sum(axis=1))
# Smoke test
foo_b = Foo(model=model_b)
foo2_b = Foo.parse_raw(foo_b.json())
## Test with bigger model; incl. sqrt, abs
with PyMC_Model() as prior:
M = 2
Mtilde = 2
pm.Constant('M', M)
pm.Constant('Mtilde', Mtilde)
μ = pm.Normal('μtilde', mu=0, sigma=2, shape=(M, ))
logτ = pm.Normal('τtilde', mu=0, sigma=1, shape=(M, ))
W = pm.Normal('Wtilde', mu=0, sigma=1, shape=(Mtilde, M))
σ = pm.Deterministic('σtilde', tt.sqrt(abs(W.sum(axis=1))))
foo_prior = Foo(model=prior)
foo2_prior = Foo.parse_raw(foo_prior.json())
assert isinstance(foo2_prior.model.M.distribution, pm.Constant)