def test_mech_spec_list(self):
R1 = TransferMechanism(output_ports=['FIRST', 'SECOND'])
T = TransferMechanism(default_variable=[[0]], input_ports=[R1])
np.testing.assert_array_equal(T.defaults.variable, np.array([[0]]))
assert len(T.input_ports) == 1
assert T.input_port.path_afferents[0].sender == R1.output_port
T.execute()
def test_runtime_params_reset_isolated(self):
T = TransferMechanism()
# Intercept attr updated
T.function.intercept = 2.0
assert T.function.intercept == 2.0
# Runtime param used for slope
T.execute(runtime_params={"slope": 10.0}, input=2.0)
assert T.function.slope == 10.0
assert T.parameter_states['slope'].value == 10.0
# Intercept attr NOT affected by runtime params
assert T.function.intercept == 2.0
assert T.value == 22.0
# Runtime param NOT used for slope
T.execute(input=2.0)
assert T.function.slope == 1.0
assert T.parameter_states['slope'].value == 1.0
# Intercept attr NOT affected by runtime params reset
assert T.function.intercept == 2.0
assert T.value == 4.0
def test_mech_spec_standalone(self):
R1 = TransferMechanism(output_ports=['FIRST', 'SECOND'])
# Mechanism outside of list specification
T = TransferMechanism(default_variable=[[0]], input_ports=R1)
np.testing.assert_array_equal(T.defaults.variable, np.array([[0]]))
assert len(T.input_ports) == 1
assert T.input_port.path_afferents[0].sender == R1.output_port
T.execute()
def test_2_item_tuple_value_for_first_item(self):
R2 = TransferMechanism(size=3)
T = TransferMechanism(input_ports=[([0,0], R2)])
np.testing.assert_array_equal(T.defaults.variable, np.array([[0, 0]]))
assert len(T.input_ports) == 1
assert T.input_port.path_afferents[0].sender.socket_width == 3
assert T.input_port.socket_width == 2
T.execute()
def test_projection_tuple_with_matrix_spec(self):
R2 = TransferMechanism(size=3)
T = TransferMechanism(size=2, input_ports=[(R2, None, None, np.zeros((3, 2)))])
np.testing.assert_array_equal(T.defaults.variable, np.array([[0, 0]]))
assert len(T.input_ports) == 1
assert T.input_port.path_afferents[0].sender.defaults.variable.shape[-1] == 3
assert T.input_port.socket_width == 2
T.execute()
def test_output_port_spec_standalone(self):
R1 = TransferMechanism(output_ports=['FIRST', 'SECOND'])
T = TransferMechanism(default_variable=[0],
input_ports=R1.output_ports['FIRST'])
np.testing.assert_array_equal(T.defaults.variable, np.array([[0]]))
assert len(T.input_ports) == 1
assert T.input_ports.names[0] == 'InputPort-0'
T.input_port.path_afferents[0].sender == R1.output_port
T.execute()
def test_2_item_tuple_spec(self):
R2 = TransferMechanism(size=3)
T = TransferMechanism(size=2, input_states=[(R2, np.zeros((3, 2)))])
np.testing.assert_array_equal(T.defaults.variable, np.array([[0, 0]]))
assert len(T.input_states) == 1
assert len(
T.input_state.path_afferents[0].sender.defaults.variable) == 3
assert T.input_state.socket_width == 2
T.execute()
def test_projection_list(self):
R2 = TransferMechanism(size=3)
P = MappingProjection(sender=R2)
T = TransferMechanism(size=2, input_states=[P])
np.testing.assert_array_equal(T.defaults.variable, np.array([[0, 0]]))
assert len(T.input_states) == 1
assert len(
T.input_state.path_afferents[0].sender.defaults.variable) == 3
assert len(T.input_state.defaults.variable) == 2
T.execute()
def test_projection_in_tuple(self):
R2 = TransferMechanism(size=3)
P = MappingProjection(sender=R2)
T = TransferMechanism(size=2, input_ports=[(R2, None, None, P)])
np.testing.assert_array_equal(T.defaults.variable, np.array([[0, 0]]))
assert len(T.input_ports) == 1
assert len(
T.input_port.path_afferents[0].sender.defaults.variable) == 3
assert len(T.input_port.defaults.variable) == 2
T.execute()
def test_inspect_function_params_slope_noise(self):
A = TransferMechanism()
B = TransferMechanism()
assert A.function.slope.base == 1.0
assert B.function.slope.base == 1.0
assert A.function.slope.modulated == [1.0]
assert B.function.slope.modulated == [1.0]
assert A.noise.base == 0.0
assert B.noise.base == 0.0
assert A.noise.modulated == 0.0
assert B.noise.modulated == 0.0
A.function.slope.base = 0.2
assert A.function.slope.base == 0.2
assert B.function.slope.base == 1.0
assert A.function.slope.modulated == [1.0]
assert B.function.slope.modulated == [1.0]
A.noise.base = 0.5
assert A.noise.base == 0.5
assert B.noise.base == 0.0
assert A.noise.modulated == 0.0
assert B.noise.modulated == 0.0
B.function.slope.base = 0.7
assert A.function.slope.base == 0.2
assert B.function.slope.base == 0.7
assert A.function.slope.modulated == [1.0]
assert B.function.slope.modulated == [1.0]
B.noise.base = 0.6
assert A.noise.base == 0.5
assert B.noise.base == 0.6
assert A.noise.modulated == 0.0
assert B.noise.modulated == 0.0
A.execute(1.0)
assert A.function.slope.modulated == [0.2]
B.execute(1.0)
assert A.function.slope.base == 0.2
assert B.function.slope.base == 0.7
assert A.function.slope.modulated == [0.2]
assert B.function.slope.modulated == [0.7]
assert A.noise.base == 0.5
assert B.noise.base == 0.6
assert A.noise.modulated == 0.5
assert B.noise.modulated == 0.6
def test_default_variable_override_mech_list(self):
R2 = TransferMechanism(size=3)
# default_variable override of OutputPort.value
T = TransferMechanism(default_variable=[[0, 0]], input_ports=[R2])
np.testing.assert_array_equal(T.defaults.variable, np.array([[0, 0]]))
assert len(T.input_ports) == 1
assert len(
T.input_port.path_afferents[0].sender.defaults.variable) == 3
assert len(T.input_port.defaults.variable[0]) == 2
T.execute()
def test_output_port_spec_list_two_items(self):
R1 = TransferMechanism(output_ports=['FIRST', 'SECOND'])
T = TransferMechanism(
default_variable=[[0], [0]],
input_ports=[R1.output_ports['FIRST'], R1.output_ports['SECOND']])
np.testing.assert_array_equal(T.defaults.variable, np.array([[0],
[0]]))
assert len(T.input_ports) == 2
assert T.input_ports.names[0] == 'InputPort-0'
assert T.input_ports.names[1] == 'InputPort-1'
for input_port in T.input_ports:
for projection in input_port.path_afferents:
assert projection.sender.owner is R1
T.execute()
def test_projection_in_specification_dict(self):
R1 = TransferMechanism(output_ports=['FIRST', 'SECOND'])
T = TransferMechanism(input_ports=[{
NAME:
'My InputPort with Two Projections',
PROJECTIONS: [R1.output_ports['FIRST'], R1.output_ports['SECOND']]
}])
np.testing.assert_array_equal(T.defaults.variable, np.array([[0]]))
assert len(T.input_ports) == 1
assert T.input_port.name == 'My InputPort with Two Projections'
for input_port in T.input_ports:
for projection in input_port.path_afferents:
assert projection.sender.owner is R1
T.execute()
def test_mechanism(benchmark, executions, mech_mode):
benchmark.group = "TransferMechanism multirun {}".format(executions)
variable = [0 for _ in range(SIZE)]
T = TransferMechanism(
name='T',
default_variable=variable,
integration_rate=1.0,
noise=-2.0,
integrator_mode=True
)
var = [[10.0 for _ in range(SIZE)] for _ in range(executions)]
expected = [[8.0 for i in range(SIZE)]]
if mech_mode == 'Python':
f = lambda x : [T.execute(x[i]) for i in range(executions)]
res = benchmark(f if executions > 1 else T.execute, var)
elif mech_mode == 'LLVM':
e = pnlvm.execution.MechExecution(T, [None for _ in range(executions)])
res = benchmark(e.execute, var)
elif mech_mode == 'PTX':
e = pnlvm.execution.MechExecution(T, [None for _ in range(executions)])
res = benchmark(e.cuda_execute, var)
if executions > 1:
expected = [expected for _ in range(executions)]
assert np.allclose(res, expected)
assert len(res) == executions
def test_function_runtime_param(self):
T = TransferMechanism()
assert T.function.slope.base == 1.0
assert T.parameter_ports['slope'].value == 1.0
# runtime param used for slope
T.execute(runtime_params={"slope": 10.0}, input=2.0)
assert T.value == 20.0
# defalut values are restored
assert T.function.slope.base == 1.0
assert T.parameter_ports['slope'].value == 1.0
T.execute(input=2.0)
assert T.function.slope.base == 1.0
assert T.parameter_ports['slope'].value == 1.0
assert T.value == 2.0
def test_mechanism_runtime_param(self):
T = TransferMechanism()
assert T.noise.base == 0.0
assert T.parameter_ports['noise'].value == 0.0
# runtime param used for noise
T.execute(runtime_params={"noise": 10.0}, input=2.0)
assert T.value == 12.0
# defalut values are restored
assert T.noise.base == 0.0
assert T.parameter_ports['noise'].value == 0.0
T.execute(input=2.0)
assert T.noise.base == 0.0
assert T.parameter_ports['noise'].value == 0.0
assert T.value == 2.0
def test_mechanism_execute_function_param(self):
# Construction
T = TransferMechanism()
assert T.function.slope == 1.0
assert T.parameter_states['slope'].value == 1.0
# Runtime param used for slope
T.execute(runtime_params={"slope": 10.0}, input=2.0)
assert T.function.slope == 10.0
assert T.parameter_states['slope'].value == 10.0
assert T.value == 20.0
# Runtime param NOT used for slope
T.execute(input=2.0)
assert T.function.slope == 1.0
assert T.parameter_states['slope'].value == 1.0
assert T.value == 2.0
def test_mechanism_execute_mechanism_param(self):
# Construction
T = TransferMechanism()
assert T.noise == 0.0
assert T.parameter_states['noise'].value == 0.0
# Runtime param used for noise
T.execute(runtime_params={"noise": 10.0}, input=2.0)
assert T.noise == 10.0
assert T.parameter_states['noise'].value == 10.0
assert T.value == 12.0
# Runtime param NOT used for noise
T.execute(input=2.0)
assert T.noise == 0.0
assert T.parameter_states['noise'].value == 0.0
assert T.value == 2.0
def test_transfer_mech_process_matrix_change(self):
from psyneulink.core.components.projections.pathway.mappingprojection import MappingProjection
T1 = TransferMechanism(
size=4,
function=Linear)
proj = MappingProjection(matrix=[[1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1]])
T2 = TransferMechanism(
size=4,
function=Linear)
p = Process(size=4, pathway=[T1, proj, T2])
p.run(inputs={T1: [[1, 2, 3, 4]]})
proj.matrix = [[2, 2, 2, 2], [2, 2, 2, 2], [2, 2, 2, 2], [2, 2, 2, 2]]
assert np.allclose(proj.matrix, [[2, 2, 2, 2], [2, 2, 2, 2], [2, 2, 2, 2], [2, 2, 2, 2]])
# p.run(inputs={T1: [[1, 2, 3, 4]]})
T1.execute([[1, 2, 3, 4]])
proj.execute(context="EXECUTING testing projection")
assert np.allclose(proj.matrix, np.array([[2, 2, 2, 2], [2, 2, 2, 2], [2, 2, 2, 2], [2, 2, 2, 2]]))
def test_reset_to_previously_assigned_val(self):
T = TransferMechanism()
assert T.function.slope.base == 1.0
assert T.parameter_ports['slope'].value == 1.0
# set slope directly
T.function.slope.base = 2.0
assert T.function.slope.base == 2.0
# runtime param used for slope
T.execute(runtime_params={"slope": 10.0}, input=2.0)
assert T.value == 20.0
# slope restored to previously assigned value
assert T.function.slope.base == 2.0
assert T.parameter_ports['slope'].value == 2.0
T.execute(input=2.0)
assert T.value == 4.0
assert T.function.slope.base == 2.0
def test_reset_state_transfer_mechanism(self):
A = TransferMechanism(name='A', integrator_mode=True)
# Execute A twice
original_output = [A.execute(1.0), A.execute(1.0)]
# SAVING STATE - - - - - - - - - - - - - - - - - - - - - - - - -
reinitialize_values = []
for attr in A.integrator_function.stateful_attributes:
reinitialize_values.append(getattr(A.integrator_function, attr))
# Execute A twice AFTER saving the state so that it continues accumulating.
# We expect the next two outputs to repeat once we reset the state b/c we will return it to the current state
output_after_saving_state = [A.execute(1.0), A.execute(1.0)]
# RESETTING STATE - - - - - - - - - - - - - - - - - - - - - - - -
A.reinitialize(*reinitialize_values)
# We expect these results to match the results from immediately after saving the state
output_after_reinitialization = [A.execute(1.0), A.execute(1.0)]
assert np.allclose(output_after_saving_state,
output_after_reinitialization)
assert np.allclose(
original_output,
[np.array([[0.5]]), np.array([[0.75]])])
assert np.allclose(
output_after_reinitialization,
[np.array([[0.875]]), np.array([[0.9375]])])
def test_specification_dict(self):
R1 = TransferMechanism(output_ports=['FIRST', 'SECOND'])
T = TransferMechanism(
default_variable=[[0], [0]],
input_ports=[
{
NAME: 'FROM DECISION',
PROJECTIONS: [R1.output_ports['FIRST']]
},
{
NAME: 'FROM RESPONSE_TIME',
PROJECTIONS: R1.output_ports['SECOND']
}
])
np.testing.assert_array_equal(T.defaults.variable, np.array([[0], [0]]))
assert len(T.input_ports) == 2
assert T.input_ports.names[0] == 'FROM DECISION'
assert T.input_ports.names[1] == 'FROM RESPONSE_TIME'
for input_port in T.input_ports:
for projection in input_port.path_afferents:
assert projection.sender.owner is R1
T.execute()
def test_runtime_params_reset_to_most_recent_val(self):
# NOT instance defaults
# Construction
T = TransferMechanism()
assert T.function.slope == 1.0
assert T.parameter_states['slope'].value == 1.0
# Set slope attribute value directly
T.function.slope = 2.0
assert T.function.slope == 2.0
# Runtime param used for slope
T.execute(runtime_params={"slope": 10.0}, input=2.0)
assert T.function.slope == 10.0
assert T.parameter_states['slope'].value == 10.0
assert T.value == 20.0
# Runtime param NOT used for slope - reset to most recent slope value (2.0)
T.execute(input=2.0)
assert T.function.slope == 2.0
assert T.value == 4.0
def test_use_and_reset_not_affect_other_assigned_vals(self):
T = TransferMechanism()
# Intercept attr assigned
T.function.intercept.base = 2.0
assert T.function.intercept.base == 2.0
# runtime param used for slope
T.execute(runtime_params={"slope": 10.0}, input=2.0)
# Assigned intercept and runtime_param for slope are used:
assert T.value == 22.0
# slope restored to default, but intercept retains assigned value
assert T.function.slope.base == 1.0
assert T.parameter_ports['slope'].value == 1.0
assert T.function.intercept.base == 2.0
# previous runtime_param for slope not used again
T.execute(input=2.0)
assert T.value == 4.0
assert T.function.slope.base == 1.0
assert T.parameter_ports['slope'].value == 1.0
def test_configurable_params(self):
old_value = 0.2
new_value = 0.7
T = TransferMechanism(function=Linear(slope=old_value,
intercept=old_value),
noise=old_value,
integration_rate=old_value)
# SLOPE - - - - - - - -
assert np.allclose(T.function.slope.base, old_value)
assert np.allclose(T.function.slope.modulated, old_value)
T.function.slope.base = new_value
assert np.allclose(T.function.slope.base, new_value)
assert np.allclose(T.function.slope.modulated, old_value)
# INTERCEPT - - - - - - - -
assert np.allclose(T.function.intercept.base, old_value)
assert np.allclose(T.function.intercept.modulated, old_value)
T.function.intercept.base = new_value
assert np.allclose(T.function.intercept.base, new_value)
assert np.allclose(T.function.intercept.modulated, old_value)
# SMOOTHING FACTOR - - - - - - - -
assert np.allclose(T.integration_rate.base, old_value)
assert np.allclose(T.integration_rate.modulated, old_value)
T.integration_rate.base = new_value
# KAM changed 3/2/18 --
# function_params looks at ParameterPort value, so this will not update until next execution
assert np.allclose(T.integration_rate.base, new_value)
assert np.allclose(T.integration_rate.modulated, old_value)
# NOISE - - - - - - - -
assert np.allclose(T.noise.base, old_value)
assert np.allclose(T.noise.modulated, old_value)
T.noise.base = new_value
# KAM changed 3/2/18 --
# function_params looks at ParameterPort value, so this will not update until next execution
assert np.allclose(T.noise.base, new_value)
assert np.allclose(T.noise.modulated, old_value)
T.execute(1.0)
assert np.allclose(T.function.slope.base, new_value)
assert np.allclose(T.function.slope.modulated, new_value)
assert np.allclose(T.function.intercept.base, new_value)
assert np.allclose(T.function.intercept.modulated, new_value)
assert np.allclose(T.integration_rate.base, new_value)
assert np.allclose(T.integration_rate.modulated, new_value)
assert np.allclose(T.noise.base, new_value)
assert np.allclose(T.noise.modulated, new_value)
def test_runtime_param_error(self):
T = TransferMechanism()
with pytest.raises(ComponentError) as error_text:
T.execute(runtime_params={"glunfump": 10.0}, input=2.0)
assert ("Invalid specification in runtime_params arg for TransferMechanism" in str(error_text.value) and
"'glunfump'" in str(error_text.value))
def test_configurable_params(self):
old_value = 0.2
new_value = 0.7
T = TransferMechanism(function=Linear(slope=old_value,
intercept=old_value),
noise=old_value,
integration_rate=old_value)
# SLOPE - - - - - - - -
assert np.allclose(T.user_params["function_params"]["slope"],
old_value)
assert np.allclose(T.function.slope, old_value)
assert np.allclose(T.mod_slope, old_value)
T.function.slope = new_value
# KAM changed 3/2/18 --
# function_params looks at parameter state value, so this will not update until next execution
assert np.allclose(T.user_params["function_params"]["slope"],
old_value)
assert np.allclose(T.function.slope, new_value)
assert np.allclose(T.mod_slope, old_value)
# INTERCEPT - - - - - - - -
assert np.allclose(T.user_params["function_params"]["intercept"],
old_value)
assert np.allclose(T.function.intercept, old_value)
assert np.allclose(T.mod_intercept, old_value)
T.function.intercept = new_value
# KAM changed 3/2/18 --
# function_params looks at parameter state value, so this will not update until next execution
assert np.allclose(T.user_params["function_params"]["intercept"],
old_value)
assert np.allclose(T.function.intercept, new_value)
assert np.allclose(T.mod_intercept, old_value)
# SMOOTHING FACTOR - - - - - - - -
assert np.allclose(T.user_params["integration_rate"], old_value)
assert np.allclose(T.integration_rate, old_value)
assert np.allclose(T.mod_integration_rate, old_value)
T.integration_rate = new_value
# KAM changed 3/2/18 --
# function_params looks at parameter state value, so this will not update until next execution
assert np.allclose(T.user_params["integration_rate"], old_value)
assert np.allclose(T.integration_rate, new_value)
assert np.allclose(T.mod_integration_rate, old_value)
# NOISE - - - - - - - -
assert np.allclose(T.user_params["noise"], old_value)
assert np.allclose(T.noise, old_value)
assert np.allclose(T.mod_noise, old_value)
T.noise = new_value
# KAM changed 3/2/18 --
# function_params looks at parameter state value, so this will not update until next execution
assert np.allclose(T.user_params["noise"], old_value)
assert np.allclose(T.noise, new_value)
assert np.allclose(T.mod_noise, old_value)
T.execute(1.0)
assert np.allclose(T.user_params["function_params"]["slope"],
new_value)
assert np.allclose(T.function.slope, new_value)
assert np.allclose(T.mod_slope, new_value)
assert np.allclose(T.user_params["function_params"]["intercept"],
new_value)
assert np.allclose(T.function.intercept, new_value)
assert np.allclose(T.mod_intercept, new_value)
assert np.allclose(T.user_params["integration_rate"], new_value)
assert np.allclose(T.integration_rate, new_value)
assert np.allclose(T.mod_integration_rate, new_value)
assert np.allclose(T.user_params["noise"], new_value)
assert np.allclose(T.noise, new_value)
assert np.allclose(T.mod_noise, new_value)
