def gaussian_univariate_lc_y_dimension_test(self):
"""Test if the Normal LC DLM observation has the expected dimension.
"""
lc = UnivariateStructure.locally_constant(1.0)
dlm = NormalDLM(structure=lc, V=1.4)
y = dlm.observation(np.array([0]))
assert_true(np.isscalar(y), "Observation is not a scalar")
def univariate_lc_state_dimension_mismatch_test(self):
"""
Test if a state of wrong dimensions throws an exception
"""
lc = UnivariateStructure.locally_constant(1.0)
dlm = NormalDLM(structure=lc, V=1.4)
state0 = np.array([0, 0])
dlm.observation(state0)
def composite_lcll_state_dimension_test(self):
"""Test if the composite LC/LL state has the expected dimensions."""
lc = UnivariateStructure.locally_constant(1.0)
ll = UnivariateStructure.locally_linear(np.eye(2))
ndlm1 = NormalDLM(lc, 1.0)
ndlm2 = NormalDLM(ll, 1.0)
composite = CompositeDLM(ndlm1, ndlm2)
m0 = np.array([0, 0, 0])
state = composite.state(m0)
assert_equals(state.shape, (3,), "state dimensions not correct")
def composite_lc_state_dimension_test(self):
"""Test if the composite LC state has the expected dimensions."""
lc1 = UnivariateStructure.locally_constant(1.0)
lc2 = UnivariateStructure.locally_constant(1.0)
ndlm1 = NormalDLM(lc1, 1.0)
ndlm2 = NormalDLM(lc2, 1.0)
composite = CompositeDLM(ndlm1, ndlm2)
m0 = np.array([0, 0])
state = composite.state(m0)
shape = state.shape
assert_equals(shape, (2,), "state dimensions not correct")
def univariate_ll_prior_dimensions_test(self):
"""The default LL DLM state prior should have the correct dimensions
"""
ll = UnivariateStructure.locally_linear(np.eye(2))
dlm = NormalDLM(structure=ll, V=1.4)
assert_equal(dlm.current_state.shape, (2, ),
"Prior has the wrong dimensions")
def univariate_lc_prior_dimensions_test(self):
"""The default LC DLM state prior should have the correct dimensions
"""
lc = UnivariateStructure.locally_constant(1.0)
dlm = NormalDLM(structure=lc, V=1.4)
assert_equal(dlm.current_state.shape, (1, ),
"Prior has the wrong dimensions")
def composite_lcll_structure_dimension_test(self):
"""Test if the composite LC/LL structure has the expected dimensions."""
lc = UnivariateStructure.locally_constant(1.0)
ll = UnivariateStructure.locally_linear(np.eye(2))
ndlm1 = NormalDLM(lc, 1.0)
ndlm2 = NormalDLM(ll, 1.0)
composite = CompositeDLM(ndlm1, ndlm2)
shape = composite.structure.F.shape
assert_equals(shape, (3, 2), "F dimensions not correct")
shape = composite.structure.G.shape
assert_equals(shape, (3, 3), "G dimensions not correct")
shape = composite.structure.W.shape
assert_equals(shape, (3, 3), "W dimensions not correct")
def composite_lc_structure_dimension_test(self):
"""Test if the composite LC structure has the expected dimensions."""
lc1 = UnivariateStructure.locally_constant(1.0)
lc2 = UnivariateStructure.locally_constant(1.0)
ndlm1 = NormalDLM(lc1, 1.0)
ndlm2 = NormalDLM(lc2, 1.0)
composite = CompositeDLM(ndlm1, ndlm2)
shape = composite.structure.F.shape
assert_equals(shape, (2, 2), "F dimensions not correct")
shape = composite.structure.G.shape
assert_equals(shape, (2, 2), "G dimensions not correct")
shape = composite.structure.W.shape
assert_equals(shape, (2, 2), "W dimensions not correct")
def main(args):
logging.basicConfig(level=args.logging)
logging.info('brokers={}'.format(args.brokers))
logging.info('topic={}'.format(args.topic))
logging.info('conf={}'.format(args.conf))
if args.conf:
model, state, period, name, anomalies = parse_configuration(
_read_conf(args.conf))
else:
state = np.array([0])
lc = UnivariateStructure.locally_constant(1.0)
model = NormalDLM(structure=lc, V=1.4)
period = 2.0
name = 'data'
anomalies = [lambda x: x]
logging.info('creating kafka producer')
producer = KafkaProducer(bootstrap_servers=args.brokers)
logging.info('sending lines (frequency = {})'.format(period))
while True:
dimensions = np.size(state)
if dimensions == 1:
logging.debug("state = {}".format(state))
_state = anomalies[0](state)
logging.debug("anomaly = {}".format(_state))
else:
_state = np.copy(state)
for i in range(dimensions):
logging.debug("state {} = {}".format(i, state[i]))
_state[i] = anomalies[i](state[i])
logging.debug("anomaly {} = {}".format(i, state[i]))
y = model.observation(_state)
state = model.state(state)
message = build_message(name, y)
logging.info("message = {}".format(message))
producer.send(args.topic, message)
time.sleep(period)
def univariate_ll_iterator_dimensions_test(self):
"""The LL DLM iterator result should have the correct dimensions"""
ll = UnivariateStructure.locally_linear(np.eye(2))
dlm = NormalDLM(structure=ll, V=1.4)
items = [next(dlm) for _ in range(10)]
print(items)
assert_equal(items[0][0].shape, (2, ),
"State has the wrong dimensions")
assert_true(np.isscalar(items[0][1]),
"Observation has the wrong dimensions")
def setUpClass(cls):
# Locally constant
lc = {"structure": UnivariateStructure.locally_constant(1.0)}
lc["model"] = NormalDLM(structure=lc["structure"], V=1.4)
lc["nobs"] = 100
# the initial state prior
m0 = np.array([0])
C0 = np.matrix([[1]])
state0 = rand.multivariate_normal(m0, C0)
lc["states"] = [state0]
for t in range(1, lc["nobs"]):
lc["states"].append(lc["model"].state(lc["states"][t - 1]))
lc["obs"] = [None]
for t in range(1, lc["nobs"]):
lc["obs"].append(lc["model"].observation(lc["states"][t]))
cls._lc = lc
# Locally linear
ll = {"structure": UnivariateStructure.locally_linear(np.identity(2))}
ll["model"] = NormalDLM(structure=ll["structure"], V=1.4)
ll["nobs"] = 100
# the initial state prior
m0 = np.array([0, 0])
C0 = np.identity(2)
state0 = rand.multivariate_normal(m0, C0)
ll["states"] = [state0]
for t in range(1, ll["nobs"]):
ll["states"].append(ll["model"].state(ll["states"][t - 1]))
ll["obs"] = [None]
for t in range(1, ll["nobs"]):
ll["obs"].append(ll["model"].observation(ll["states"][t]))
cls._ll = ll
def _parse_observations(obs, structure):
if obs['type'] == 'continuous':
model = NormalDLM(structure=structure, V=obs['noise'])
elif obs['type'] == 'discrete':
model = PoissonDLM(structure=structure)
elif obs['type'] == 'categorical':
if 'values' in obs:
values = obs['values'].split(',')
model = BinomialTransformer(structure=structure, source=values)
elif 'categories' in obs:
model = BinomialDLM(structure=structure,
categories=obs['categories'])
else:
raise ValueError("Categorical models must have either 'values' "
"or 'categories'")
else:
raise ValueError("Model type {} is not valid".format(obs['type']))
return model