def test_multilayer():
Input_Layer = TransferMechanism(
name='Input Layer',
function=Logistic,
default_variable=np.zeros((2, )),
)
Hidden_Layer_1 = TransferMechanism(
name='Hidden Layer_1',
function=Logistic(),
# default_variable=np.zeros((5,)),
size=5)
Hidden_Layer_2 = TransferMechanism(
name='Hidden Layer_2',
function=Logistic(),
default_variable=[0, 0, 0, 0],
)
Output_Layer = TransferMechanism(
name='Output Layer',
function=Logistic,
default_variable=[0, 0, 0],
)
Input_Weights_matrix = (np.arange(2 * 5).reshape((2, 5)) + 1) / (2 * 5)
Middle_Weights_matrix = (np.arange(5 * 4).reshape((5, 4)) + 1) / (5 * 4)
Output_Weights_matrix = (np.arange(4 * 3).reshape((4, 3)) + 1) / (4 * 3)
# TEST PROCESS.LEARNING WITH:
# CREATION OF FREE STANDING PROJECTIONS THAT HAVE NO LEARNING (Input_Weights, Middle_Weights and Output_Weights)
# INLINE CREATION OF PROJECTIONS (Input_Weights, Middle_Weights and Output_Weights)
# NO EXPLICIT CREATION OF PROJECTIONS (Input_Weights, Middle_Weights and Output_Weights)
# This projection will be used by the process below by referencing it in the process' pathway;
# note: sender and receiver args don't need to be specified
Input_Weights = MappingProjection(
name='Input Weights',
matrix=Input_Weights_matrix,
)
# This projection will be used by the process below by assigning its sender and receiver args
# to mechanismss in the pathway
Middle_Weights = MappingProjection(
name='Middle Weights',
sender=Hidden_Layer_1,
receiver=Hidden_Layer_2,
matrix=Middle_Weights_matrix,
)
# Commented lines in this projection illustrate variety of ways in which matrix and learning signals can be specified
Output_Weights = MappingProjection(
name='Output Weights',
sender=Hidden_Layer_2,
receiver=Output_Layer,
matrix=Output_Weights_matrix,
)
p = Process(
# default_variable=[0, 0],
size=2,
pathway=[
Input_Layer,
# The following reference to Input_Weights is needed to use it in the pathway
# since it's sender and receiver args are not specified in its declaration above
Input_Weights,
Hidden_Layer_1,
# No projection specification is needed here since the sender arg for Middle_Weights
# is Hidden_Layer_1 and its receiver arg is Hidden_Layer_2
# Middle_Weights,
Hidden_Layer_2,
# Output_Weights does not need to be listed for the same reason as Middle_Weights
# If Middle_Weights and/or Output_Weights is not declared above, then the process
# will assign a default for missing projection
# Output_Weights,
Output_Layer
],
clamp_input=SOFT_CLAMP,
learning=LEARNING,
learning_rate=1.0,
target=[0, 0, 1],
prefs={
VERBOSE_PREF: False,
REPORT_OUTPUT_PREF: False
},
)
stim_list = {Input_Layer: [[-1, 30]]}
target_list = {Output_Layer: [[0, 0, 1]]}
def show_target():
i = s.input
t = s.target_input_states[0].parameters.value.get(s)
print('\nOLD WEIGHTS: \n')
print('- Input Weights: \n', Input_Weights.get_mod_matrix(s))
print('- Middle Weights: \n', Middle_Weights.get_mod_matrix(s))
print('- Output Weights: \n', Output_Weights.get_mod_matrix(s))
print('\nSTIMULI:\n\n- Input: {}\n- Target: {}\n'.format(i, t))
print('ACTIVITY FROM OLD WEIGHTS: \n')
print('- Middle 1: \n', Hidden_Layer_1.parameters.value.get(s))
print('- Middle 2: \n', Hidden_Layer_2.parameters.value.get(s))
print('- Output:\n', Output_Layer.parameters.value.get(s))
s = System(
processes=[p],
targets=[0, 0, 1],
learning_rate=1.0,
)
# s.reportOutputPref = True
results = s.run(
num_trials=10,
inputs=stim_list,
targets=target_list,
call_after_trial=show_target,
)
objective_output_layer = s.mechanisms[4]
results_list = []
for elem in s.results:
for nested_elem in elem:
nested_elem = nested_elem.tolist()
try:
iter(nested_elem)
except TypeError:
nested_elem = [nested_elem]
results_list.extend(nested_elem)
expected_output = [
(Output_Layer.get_output_values(s),
[np.array([0.22686074, 0.25270212, 0.91542149])]),
(objective_output_layer.output_states[MSE].parameters.value.get(s),
np.array(0.04082589331852094)),
(Input_Weights.get_mod_matrix(s),
np.array([
[0.09900247, 0.19839653, 0.29785764, 0.39739191, 0.49700232],
[0.59629092, 0.69403786, 0.79203411, 0.89030237, 0.98885379],
])),
(Middle_Weights.get_mod_matrix(s),
np.array([
[0.09490249, 0.10488719, 0.12074013, 0.1428774],
[0.29677354, 0.30507726, 0.31949676, 0.3404652],
[0.49857336, 0.50526254, 0.51830509, 0.53815062],
[0.70029406, 0.70544225, 0.71717037, 0.73594383],
[0.90192903, 0.90561554, 0.91609668, 0.93385292],
])),
(Output_Weights.get_mod_matrix(s),
np.array([
[-0.74447522, -0.71016859, 0.31575293],
[-0.50885177, -0.47444784, 0.56676582],
[-0.27333719, -0.23912033, 0.8178167],
[-0.03767547, -0.00389039, 1.06888608],
])),
(results, [[np.array([0.8344837, 0.87072018, 0.89997433])],
[np.array([0.77970193, 0.83263138, 0.90159627])],
[np.array([0.70218502, 0.7773823, 0.90307765])],
[np.array([0.60279149, 0.69958079, 0.90453143])],
[np.array([0.4967927, 0.60030321, 0.90610082])],
[np.array([0.4056202, 0.49472391, 0.90786617])],
[np.array([0.33763025, 0.40397637, 0.90977675])],
[np.array([0.28892812, 0.33633532, 0.9117193])],
[np.array([0.25348771, 0.28791896, 0.9136125])],
[np.array([0.22686074, 0.25270212, 0.91542149])]]),
]
# Test nparray output of log for Middle_Weights
for i in range(len(expected_output)):
val, expected = expected_output[i]
# setting absolute tolerance to be in accordance with reference_output precision
# if you do not specify, assert_allcose will use a relative tolerance of 1e-07,
# which WILL FAIL unless you gather higher precision values to use as reference
np.testing.assert_allclose(
val,
expected,
atol=1e-08,
err_msg='Failed on expected_output[{0}]'.format(i))
def test_stroop_model_learning(self):
process_prefs = {
REPORT_OUTPUT_PREF: True,
VERBOSE_PREF: False,
}
system_prefs = {
REPORT_OUTPUT_PREF: True,
VERBOSE_PREF: False,
}
colors = TransferMechanism(
default_variable=[0, 0],
function=Linear,
name="Colors",
)
words = TransferMechanism(
default_variable=[0, 0],
function=Linear,
name="Words",
)
hidden = TransferMechanism(
default_variable=[0, 0],
function=Logistic,
name="Hidden",
)
response = TransferMechanism(
default_variable=[0, 0],
function=Logistic(),
name="Response",
)
TransferMechanism(
default_variable=[0, 0],
function=Logistic,
name="Output",
)
CH_Weights_matrix = np.arange(4).reshape((2, 2))
WH_Weights_matrix = np.arange(4).reshape((2, 2))
HO_Weights_matrix = np.arange(4).reshape((2, 2))
CH_Weights = MappingProjection(
name='Color-Hidden Weights',
matrix=CH_Weights_matrix,
)
WH_Weights = MappingProjection(
name='Word-Hidden Weights',
matrix=WH_Weights_matrix,
)
HO_Weights = MappingProjection(
name='Hidden-Output Weights',
matrix=HO_Weights_matrix,
)
color_naming_process = Process(
default_variable=[1, 2.5],
pathway=[colors, CH_Weights, hidden, HO_Weights, response],
learning=LEARNING,
target=[2, 2],
name='Color Naming',
prefs=process_prefs,
)
word_reading_process = Process(
default_variable=[.5, 3],
pathway=[words, WH_Weights, hidden],
name='Word Reading',
learning=LEARNING,
target=[3, 3],
prefs=process_prefs,
)
s = System(
processes=[color_naming_process, word_reading_process],
targets=[20, 20],
name='Stroop Model',
prefs=system_prefs,
)
def show_target():
print('\nColor Naming\n\tInput: {}\n\tTarget: {}'.format([
np.ndarray.tolist(item.parameters.value.get(s))
for item in colors.input_states
], s.targets))
print('Wording Reading:\n\tInput: {}\n\tTarget: {}\n'.format([
np.ndarray.tolist(item.parameters.value.get(s))
for item in words.input_states
], s.targets))
print('Response: \n',
response.output_state.parameters.value.get(s))
print('Hidden-Output:')
print(HO_Weights.get_mod_matrix(s))
print('Color-Hidden:')
print(CH_Weights.get_mod_matrix(s))
print('Word-Hidden:')
print(WH_Weights.get_mod_matrix(s))
stim_list_dict = {colors: [[1, 1]], words: [[-2, -2]]}
target_list_dict = {response: [[1, 1]]}
results = s.run(
num_trials=2,
inputs=stim_list_dict,
targets=target_list_dict,
call_after_trial=show_target,
)
results_list = []
for elem in s.results:
for nested_elem in elem:
nested_elem = nested_elem.tolist()
try:
iter(nested_elem)
except TypeError:
nested_elem = [nested_elem]
results_list.extend(nested_elem)
objective_response = s.mechanisms[3]
objective_hidden = s.mechanisms[7]
from pprint import pprint
pprint(CH_Weights.__dict__)
print(CH_Weights._parameter_states["matrix"].value)
print(CH_Weights.get_mod_matrix(s))
expected_output = [
(colors.output_states[0].parameters.value.get(s),
np.array([1., 1.])),
(words.output_states[0].parameters.value.get(s),
np.array([-2., -2.])),
(hidden.output_states[0].parameters.value.get(s),
np.array([0.13227553, 0.01990677])),
(response.output_states[0].parameters.value.get(s),
np.array([0.51044657, 0.5483048])),
(objective_response.output_states[0].parameters.value.get(s),
np.array([0.48955343, 0.4516952])),
(objective_response.output_states[MSE].parameters.value.get(s),
np.array(0.22184555903789838)),
(CH_Weights.get_mod_matrix(s),
np.array([
[0.02512045, 1.02167245],
[2.02512045, 3.02167245],
])),
(WH_Weights.get_mod_matrix(s),
np.array([
[-0.05024091, 0.9566551],
[1.94975909, 2.9566551],
])),
(HO_Weights.get_mod_matrix(s),
np.array([
[0.03080958, 1.02830959],
[2.00464242, 3.00426575],
])),
(results, [[np.array([0.50899214, 0.54318254])],
[np.array([0.51044657, 0.5483048])]]),
]
for i in range(len(expected_output)):
val, expected = expected_output[i]
# setting absolute tolerance to be in accordance with reference_output precision
# if you do not specify, assert_allcose will use a relative tolerance of 1e-07,
# which WILL FAIL unless you gather higher precision values to use as reference
np.testing.assert_allclose(
val,
expected,
atol=1e-08,
err_msg='Failed on expected_output[{0}]'.format(i))
# KDM 10/16/18: Comparator Mechanism for Hidden is not executed by the system, because it's not associated with
# an output mechanism. So it actually should be None instead of previously [0, 0] which was likely
# a side effect with of conflation of different execution contexts
assert objective_hidden.output_states[0].parameters.value.get(
s) is None
