def test_last_item_retrieved_correctly() -> None:
x_label = "x"
def factory(sequence_item):
x = sequence_item.add_double_proxy_for(x_label)
x_out = x * Const(2.0)
x_out.set_label(x_label)
sequence_item.add(x_out)
x_start = ConstantDouble(1.0)
initial_state: Optional[Dict[str, vertex_constructor_param_types]] = {
x_label: x_start
}
sequence = Sequence(count=2,
factories=factory,
initial_state=initial_state)
sequence_item_contents = sequence.get_last_item().get_contents()
x_output = sequence_item_contents.get(x_label)
x_proxy = sequence_item_contents.get(Sequence.proxy_label_for(x_label))
assert x_output is not None
assert x_proxy is not None
assert x_output.get_value() == 4
assert x_proxy.get_value() == 2
def test_you_can_name_a_sequence() -> None:
x_label = "x"
def factory(sequence_item):
x = sequence_item.add_double_proxy_for(x_label)
x_out = x * Const(2.0)
x_out.set_label(x_label)
sequence_item.add(x_out)
x_start = ConstantDouble(1.0)
initial_state: Optional[Dict[str, vertex_constructor_param_types]] = {
x_label: x_start
}
sequence_name = "My_Awesome_Sequence"
sequence = Sequence(count=2,
factories=factory,
initial_state=initial_state,
name=sequence_name)
sequence_item_contents = sequence.get_last_item().get_contents()
x_output = sequence_item_contents.get(x_label)
x_proxy = sequence_item_contents.get(Sequence.proxy_label_for(x_label))
assert x_output is not None
assert x_proxy is not None
assert x_output.get_value() == 4
assert x_proxy.get_value() == 2
x_output_label = x_output.get_label()
assert x_output_label is not None
assert re.match("My_Awesome_Sequence.Sequence_Item_1.\d+.x",
x_output_label) is not None
def test_you_can_build_a_time_series() -> None:
"""
This is a Hidden Markov Model -
see for example http://mlg.eng.cam.ac.uk/zoubin/papers/ijprai.pdf
... --> X[t-1] --> X[t] --> ...
| |
Y[t-1] Y[t]
"""
x_label = "x"
y_label = "y"
num_items = 10
initial_x = 1.
def create_time_step(sequence_item):
x_previous = sequence_item.add_double_proxy_for(x_label)
x = Exponential(x_previous)
y = Poisson(x)
sequence_item.add(x, label=x_label)
sequence_item.add(y, label=y_label)
sequence = Sequence(initial_state={x_label: initial_x},
count=num_items,
factories=create_time_step)
assert sequence.size() == num_items
x_from_previous_step = None
x_previous_label = Sequence.proxy_label_for(x_label)
for item in sequence:
x_previous_proxy = item.get(x_previous_label)
x = item.get(x_label)
y = item.get(y_label)
if x_from_previous_step is None:
assert [p.get_value()
for p in x_previous_proxy.iter_parents()] == [initial_x]
else:
assert [p.get_id() for p in x_previous_proxy.iter_parents()
] == [x_from_previous_step.get_id()]
assert [p.get_id()
for p in x.iter_parents()] == [x_previous_proxy.get_id()]
assert [p.get_id() for p in y.iter_parents()] == [x.get_id()]
x_from_previous_step = x
def test_you_can_use_multiple_factories_to_build_sequences() -> None:
x1_label = "x1"
x2_label = "x2"
x3_label = "x3"
x4_label = "x4"
two = ConstantDouble(2)
half = ConstantDouble(0.5)
def factory1(sequence_item):
x1_input = sequence_item.add_double_proxy_for(x1_label)
x2_input = sequence_item.add_double_proxy_for(x2_label)
x1_output = x1_input * two
x1_output.set_label(x1_label)
x3_output = x2_input * two
x3_output.set_label(x3_label)
sequence_item.add(x1_output)
sequence_item.add(x3_output)
def factory2(sequence_item):
x3_input = sequence_item.add_double_proxy_for(x3_label)
x4_input = sequence_item.add_double_proxy_for(x4_label)
x2_output = x3_input * half
x2_output.set_label(x2_label)
x4_output = x4_input * half
x4_output.set_label(x4_label)
sequence_item.add(x2_output)
sequence_item.add(x4_output)
x1_start = ConstantDouble(4)
x2_start = ConstantDouble(4)
x3_start = ConstantDouble(4)
x4_start = ConstantDouble(4)
initial_state: Optional[Dict[str, vertex_constructor_param_types]] = {
x1_label: x1_start,
x2_label: x2_start,
x3_label: x3_start,
x4_label: x4_start
}
factories = [factory1, factory2]
sequence = Sequence(count=5,
factories=factories,
initial_state=initial_state)
assert sequence.size() == 5
for item in sequence:
__check_sequence_output_links_to_input(
item, Sequence.proxy_label_for(x1_label), x1_label)
__check_sequence_output_links_to_input(
item, Sequence.proxy_label_for(x2_label), x3_label)
__check_sequence_output_links_to_input(
item, Sequence.proxy_label_for(x3_label), x2_label)
__check_sequence_output_links_to_input(
item, Sequence.proxy_label_for(x4_label), x4_label)
__check_output_equals(sequence, x1_label, 128)
__check_output_equals(sequence, x2_label, 2)
__check_output_equals(sequence, x3_label, 8)
__check_output_equals(sequence, x4_label, 0.125)