def test_thalamus(Simulator, plt, seed):
model = spa.SPA(seed=seed)
with model:
model.vision = spa.Buffer(dimensions=16, neurons_per_dimension=80)
model.vision2 = spa.Buffer(dimensions=16, neurons_per_dimension=80)
model.motor = spa.Buffer(dimensions=16, neurons_per_dimension=80)
model.motor2 = spa.Buffer(dimensions=32, neurons_per_dimension=80)
actions = spa.Actions(
'dot(vision, A) --> motor=A, motor2=vision*vision2',
'dot(vision, B) --> motor=vision, motor2=vision*A*~B',
'dot(vision, ~A) --> motor=~vision, motor2=~vision*vision2')
model.bg = spa.BasalGanglia(actions)
model.thalamus = spa.Thalamus(model.bg)
def input_f(t):
if t < 0.1:
return 'A'
elif t < 0.3:
return 'B'
elif t < 0.5:
return '~A'
else:
return '0'
model.input = spa.Input(vision=input_f, vision2='B*~A')
input, vocab = model.get_module_input('motor')
input2, vocab2 = model.get_module_input('motor2')
p = nengo.Probe(input, 'output', synapse=0.03)
p2 = nengo.Probe(input2, 'output', synapse=0.03)
sim = Simulator(model)
sim.run(0.5)
t = sim.trange()
data = vocab.dot(sim.data[p].T)
data2 = vocab2.dot(sim.data[p2].T)
plt.subplot(2, 1, 1)
plt.plot(t, data.T)
plt.subplot(2, 1, 2)
plt.plot(t, data2.T)
# Action 1
assert data[0, t == 0.1] > 0.8
assert data[1, t == 0.1] < 0.2
assert data2[0, t == 0.1] < 0.35
assert data2[1, t == 0.1] > 0.4
# Action 2
assert data[0, t == 0.3] < 0.2
assert data[1, t == 0.3] > 0.8
assert data2[0, t == 0.3] > 0.5
assert data2[1, t == 0.3] < 0.3
# Action 3
assert data[0, t == 0.5] > 0.8
assert data[1, t == 0.5] < 0.2
assert data2[0, t == 0.5] < 0.5
assert data2[1, t == 0.5] > 0.4
def model(self, p):
model = spa.SPA()
with model:
model.vision = spa.Buffer(dimensions=p.D)
model.phrase = spa.Buffer(dimensions=p.D)
model.motor = spa.Buffer(dimensions=p.D)
model.noun = spa.Memory(dimensions=p.D, synapse=0.1)
model.verb = spa.Memory(dimensions=p.D, synapse=0.1)
model.bg = spa.BasalGanglia(
spa.Actions(
'dot(vision, WRITE) --> verb=vision',
'dot(vision, ONE+TWO+THREE) --> noun=vision',
'0.5*(dot(vision, NONE-WRITE-ONE-TWO-THREE) + '
'dot(phrase, WRITE*VERB))'
'--> motor=phrase*~NOUN',
))
model.thal = spa.Thalamus(model.bg)
model.cortical = spa.Cortical(
spa.Actions(
'phrase=noun*NOUN',
'phrase=verb*VERB',
))
def vision_input(t):
index = int(t / p.time_per_word) % 3
return ['WRITE', 'ONE', 'NONE'][index]
model.input = spa.Input(vision=vision_input)
self.motor_vocab = model.get_output_vocab('motor')
self.p_thal = nengo.Probe(model.thal.actions.output, synapse=0.03)
self.p_motor = nengo.Probe(model.motor.state.output, synapse=0.03)
return model
def thalamus_net(d=2, n=20, seed=None):
model = spa.SPA(seed=seed)
with model:
model.vision = spa.Buffer(dimensions=d, neurons_per_dimension=n)
model.vision2 = spa.Buffer(dimensions=d, neurons_per_dimension=n)
model.motor = spa.Buffer(dimensions=d, neurons_per_dimension=n)
model.motor2 = spa.Buffer(dimensions=d * 2, neurons_per_dimension=n)
actions = spa.Actions(
"dot(vision, A) --> motor=A, motor2=vision*vision2",
"dot(vision, B) --> motor=vision, motor2=vision*A*~B",
"dot(vision, ~A) --> motor=~vision, motor2=~vision*vision2",
)
model.bg = spa.BasalGanglia(actions)
model.thalamus = spa.Thalamus(model.bg)
def input_f(t):
if t < 0.1:
return "A"
elif t < 0.3:
return "B"
elif t < 0.5:
return "~A"
else:
return "0"
model.input = spa.Input(vision=input_f, vision2="B*~A")
return model
def test_routing(Simulator, seed, plt):
D = 3
model = spa.SPA(seed=seed)
with model:
model.ctrl = spa.Buffer(16, label="ctrl")
def input_func(t):
if t < 0.2:
return "A"
elif t < 0.4:
return "B"
else:
return "C"
model.input = spa.Input(ctrl=input_func)
model.buff1 = spa.Buffer(D, label="buff1")
model.buff2 = spa.Buffer(D, label="buff2")
model.buff3 = spa.Buffer(D, label="buff3")
node1 = nengo.Node([0, 1, 0])
node2 = nengo.Node([0, 0, 1])
nengo.Connection(node1, model.buff1.state.input)
nengo.Connection(node2, model.buff2.state.input)
actions = spa.Actions(
"dot(ctrl, A) --> buff3=buff1",
"dot(ctrl, B) --> buff3=buff2",
"dot(ctrl, C) --> buff3=buff1*buff2",
)
model.bg = spa.BasalGanglia(actions)
model.thal = spa.Thalamus(model.bg)
buff3_probe = nengo.Probe(model.buff3.state.output, synapse=0.03)
with Simulator(model) as sim:
sim.run(0.6)
data = sim.data[buff3_probe]
plt.plot(sim.trange(), data)
valueA = np.mean(data[150:200], axis=0) # should be [0, 1, 0]
valueB = np.mean(data[350:400], axis=0) # should be [0, 0, 1]
valueC = np.mean(data[550:600], axis=0) # should be [1, 0, 0]
assert valueA[0] < 0.2
assert valueA[1] > 0.8
assert valueA[2] < 0.2
assert valueB[0] < 0.2
assert valueB[1] < 0.2
assert valueB[2] > 0.8
assert valueC[0] > 0.8
assert valueC[1] < 0.2
assert valueC[2] < 0.2
def model(self, p):
model = spa.SPA()
with model:
model.vision = spa.Buffer(dimensions=p.D)
model.phrase = spa.Buffer(dimensions=p.D)
model.motor = spa.Buffer(dimensions=p.D)
model.noun = spa.Memory(dimensions=p.D, synapse=0.1)
model.verb = spa.Memory(dimensions=p.D, synapse=0.1)
model.bg = spa.BasalGanglia(
spa.Actions(
'dot(vision, WRITE) --> verb=vision',
'dot(vision, ONE+TWO+THREE) --> noun=vision',
'0.5*(dot(vision, NONE-WRITE-ONE-TWO-THREE) + '
'dot(phrase, WRITE*VERB))'
'--> motor=phrase*~NOUN',
))
model.thal = spa.Thalamus(model.bg)
model.cortical = spa.Cortical(
spa.Actions(
'phrase=noun*NOUN',
'phrase=verb*VERB',
))
def vision_input(t):
index = int(t / p.time_per_word) % 3
return ['WRITE', 'ONE', 'NONE'][index]
model.input = spa.Input(vision=vision_input)
self.motor_vocab = model.get_output_vocab('motor')
self.p_thal = nengo.Probe(model.thal.actions.output, synapse=0.03)
self.p_motor = nengo.Probe(model.motor.state.output, synapse=0.03)
split.split_input_nodes(model, max_dim=64)
self.replaced = split.split_passthrough(model, max_dim=p.split_max_dim)
if p.pass_ensembles > 0:
split.pass_ensembles(model, max_dim=p.pass_ensembles)
if p.backend == 'nengo_spinnaker':
import nengo_spinnaker
nengo_spinnaker.add_spinnaker_params(model.config)
for node in model.all_nodes:
if node.output is None:
if node.size_in > p.pf_max_dim:
print 'limiting', node
model.config[node].n_cores_per_chip = p.pf_cores
model.config[node].n_chips = p.pf_n_chips
model.config[nengo_spinnaker.Simulator].placer_kwargs = dict(
effort=0.1)
if p.fixed_seed:
for ens in model.all_ensembles:
ens.seed = 1
return model
def test_routing_recurrency_compilation(Simulator, seed):
D = 2
model = spa.SPA(seed=seed)
with model:
model.buff1 = spa.Buffer(D, label="buff1")
model.buff2 = spa.Buffer(D, label="buff2")
actions = spa.Actions("0.5 --> buff2=buff1, buff1=buff2")
model.bg = spa.BasalGanglia(actions)
model.thal = spa.Thalamus(model.bg)
with Simulator(model) as sim:
assert sim
def test_thalamus(Simulator):
model = spa.SPA(seed=30)
with model:
model.vision = spa.Buffer(dimensions=16, neurons_per_dimension=80)
model.vision2 = spa.Buffer(dimensions=16, neurons_per_dimension=80)
model.motor = spa.Buffer(dimensions=16, neurons_per_dimension=80)
model.motor2 = spa.Buffer(dimensions=32, neurons_per_dimension=80)
actions = spa.Actions(
'dot(vision, A) --> motor=A, motor2=vision*vision2',
'dot(vision, B) --> motor=vision, motor2=vision*A*~B',
'dot(vision, ~A) --> motor=~vision, motor2=~vision*vision2'
)
model.bg = spa.BasalGanglia(actions)
model.thalamus = spa.Thalamus(model.bg)
def input_f(t):
if t < 0.1:
return 'A'
elif t < 0.3:
return 'B'
elif t < 0.5:
return '~A'
else:
return '0'
model.input = spa.Input(vision=input_f, vision2='B*~A')
input, vocab = model.get_module_input('motor')
input2, vocab2 = model.get_module_input('motor2')
p = nengo.Probe(input, 'output', synapse=0.03)
p2 = nengo.Probe(input2, 'output', synapse=0.03)
sim = Simulator(model)
sim.run(0.5)
data = vocab.dot(sim.data[p].T)
data2 = vocab2.dot(sim.data[p2].T)
assert 0.8 < data[0, 100] < 1.1 # Action 1
assert -0.2 < data2[0, 100] < 0.35
assert -0.25 < data[1, 100] < 0.2
assert 0.4 < data2[1, 100] < 0.6
assert -0.2 < data[0, 299] < 0.2 # Action 2
assert 0.6 < data2[0, 299] < 0.95
assert 0.8 < data[1, 299] < 1.1
assert -0.2 < data2[1, 299] < 0.2
assert 0.8 < data[0, 499] < 1.0 # Action 3
assert 0.0 < data2[0, 499] < 0.5
assert -0.2 < data[1, 499] < 0.2
assert 0.4 < data2[1, 499] < 0.7
def test_errors():
# motor does not exist
with pytest.raises(SpaParseError):
with spa.SPA() as model:
model.vision = spa.Buffer(dimensions=16)
actions = spa.Actions("0.5 --> motor=A")
model.bg = spa.BasalGanglia(actions)
# dot products not implemented
with pytest.raises(NotImplementedError):
with spa.SPA() as model:
model.scalar = spa.Buffer(dimensions=16, subdimensions=1)
actions = spa.Actions("0.5 --> scalar=dot(scalar, FOO)")
model.bg = spa.BasalGanglia(actions)
model.thalamus = spa.Thalamus(model.bg)
def test_routing(Simulator):
D = 2
model = spa.SPA(seed=123)
with model:
model.ctrl = spa.Buffer(D, label='ctrl')
def input_func(t):
if t < 0.25:
return 'A'
else:
return 'B'
model.input = spa.Input(ctrl=input_func)
model.buff1 = spa.Buffer(D, label='buff1')
model.buff2 = spa.Buffer(D, label='buff2')
model.buff3 = spa.Buffer(D, label='buff3')
node1 = nengo.Node([1, 0])
node2 = nengo.Node([0, 1])
nengo.Connection(node1, model.buff1.state.input)
nengo.Connection(node2, model.buff2.state.input)
actions = spa.Actions('dot(ctrl, A) --> buff3=buff1',
'dot(ctrl, B) --> buff3=buff2')
model.bg = spa.BasalGanglia(actions)
model.thal = spa.Thalamus(model.bg)
buff1_probe = nengo.Probe(model.buff1.state.output, synapse=0.03)
buff2_probe = nengo.Probe(model.buff2.state.output, synapse=0.03)
buff3_probe = nengo.Probe(model.buff3.state.output, synapse=0.03)
thal_probe = nengo.Probe(model.thal.output, synapse=0.03)
sim = nengo.Simulator(model)
sim.run(0.5)
data1 = sim.data[buff1_probe]
data2 = sim.data[buff2_probe]
data3 = sim.data[buff3_probe]
data_thal = sim.data[thal_probe]
trange = sim.trange()
assert abs(np.mean(data1[trange < 0.25] - data3[trange < 0.25])) < 0.15
assert abs(np.mean(data2[trange > 0.25] - data3[trange > 0.25])) < 0.15
assert data_thal[249][1] < 0.01
assert data_thal[249][0] > 0.8
assert data_thal[499][1] > 0.8
assert data_thal[499][0] < 0.01
def test_nondefault_routing(Simulator, seed, plt):
D = 3
model = spa.SPA(seed=seed)
with model:
model.ctrl = spa.Buffer(16, label='ctrl')
def input_func(t):
if t < 0.2:
return 'A'
elif t < 0.4:
return 'B'
else:
return 'C'
model.input = spa.Input(ctrl=input_func)
model.buff1 = spa.Buffer(D, label='buff1')
model.buff2 = spa.Buffer(D, label='buff2')
model.cmp = spa.Compare(D)
node1 = nengo.Node([0, 1, 0])
node2 = nengo.Node([0, 0, 1])
nengo.Connection(node1, model.buff1.state.input)
nengo.Connection(node2, model.buff2.state.input)
actions = spa.Actions('dot(ctrl, A) --> cmp_A=buff1, cmp_B=buff1',
'dot(ctrl, B) --> cmp_A=buff1, cmp_B=buff2',
'dot(ctrl, C) --> cmp_A=buff2, cmp_B=buff2',
)
model.bg = spa.BasalGanglia(actions)
model.thal = spa.Thalamus(model.bg)
compare_probe = nengo.Probe(model.cmp.output, synapse=0.03)
sim = Simulator(model)
sim.run(0.6)
vocab = model.get_output_vocab('cmp')
data = sim.data[compare_probe]
similarity = np.dot(data, vocab.parse('YES').v)
valueA = np.mean(similarity[150:200], axis=0) # should be [1]
valueB = np.mean(similarity[350:400], axis=0) # should be [0]
valueC = np.mean(similarity[550:600], axis=0) # should be [1]
assert valueA > 0.6
assert valueB < 0.3
assert valueC > 0.6
def test_nondefault_routing(Simulator, seed):
D = 3
model = spa.SPA(seed=seed)
with model:
model.ctrl = spa.Buffer(16, label="ctrl")
def input_func(t):
if t < 0.2:
return "A"
elif t < 0.4:
return "B"
else:
return "C"
model.input = spa.Input(ctrl=input_func)
model.buff1 = spa.Buffer(D, label="buff1")
model.buff2 = spa.Buffer(D, label="buff2")
model.cmp = spa.Compare(D)
node1 = nengo.Node([0, 1, 0])
node2 = nengo.Node([0, 0, 1])
nengo.Connection(node1, model.buff1.state.input)
nengo.Connection(node2, model.buff2.state.input)
actions = spa.Actions(
"dot(ctrl, A) --> cmp_A=buff1, cmp_B=buff1",
"dot(ctrl, B) --> cmp_A=buff1, cmp_B=buff2",
"dot(ctrl, C) --> cmp_A=buff2, cmp_B=buff2",
)
model.bg = spa.BasalGanglia(actions)
model.thal = spa.Thalamus(model.bg)
compare_probe = nengo.Probe(model.cmp.output, synapse=0.03)
with Simulator(model) as sim:
sim.run(0.6)
similarity = sim.data[compare_probe]
valueA = np.mean(similarity[150:200], axis=0) # should be [1]
valueB = np.mean(similarity[350:400], axis=0) # should be [0]
valueC = np.mean(similarity[550:600], axis=0) # should be [1]
assert valueA > 0.6
assert valueB < 0.3
assert valueC > 0.6
def model(self, p):
model = spa.SPA()
with model:
model.vision = spa.Buffer(dimensions=p.D)
model.state = spa.Memory(dimensions=p.D)
actions = ['dot(state, S%d) --> state=S%d' % (i,(i+1))
for i in range(p.n_actions - 1)]
actions.append('dot(state, S%d) --> state=vision' %
(p.n_actions - 1))
model.bg = spa.BasalGanglia(actions=spa.Actions(*actions))
model.thal = spa.Thalamus(model.bg)
model.input = spa.Input(vision='S%d' % p.start,
state=lambda t: 'S%d' % p.start if
t < 0.1 else '0')
self.probe = nengo.Probe(model.thal.actions.output, synapse=0.03)
return model
def model(self, p):
model = spa.SPA()
with model:
model.vision = spa.Buffer(dimensions=p.D)
model.state = spa.Memory(dimensions=p.D)
actions = [
'dot(state, S%d) --> state=S%d' % (i, (i + 1))
for i in range(p.n_actions - 1)
]
actions.append('dot(state, S%d) --> state=vision' %
(p.n_actions - 1))
model.bg = spa.BasalGanglia(actions=spa.Actions(*actions))
model.thal = spa.Thalamus(model.bg)
model.input = spa.Input(vision='S%d' % p.start,
state=lambda t: 'S%d' % p.start
if t < 0.1 else '0')
self.probe = nengo.Probe(model.thal.actions.output, synapse=0.03)
split.split_passthrough(model, max_dim=p.split_max_dim)
if p.pass_ensembles > 0:
split.pass_ensembles(model, max_dim=p.pass_ensembles)
if p.backend == 'nengo_spinnaker':
import nengo_spinnaker
nengo_spinnaker.add_spinnaker_params(model.config)
for node in model.all_nodes:
if node.output is None:
if node.size_in > p.pf_max_dim:
print 'limiting', node
model.config[node].n_cores_per_chip = p.pf_cores
model.config[node].n_chips = p.pf_n_chips
model.config[nengo_spinnaker.Simulator].placer_kwargs = dict(
effort=0.1)
if p.fixed_seed:
for ens in model.all_ensembles:
ens.seed = 1
return model
def model(self, p):
model = spa.SPA()
with model:
model.state = spa.Memory(dimensions=p.D)
actions = [
'dot(state, S%d) --> state=S%d' % (i, (i + 1) % p.n_actions)
for i in range(p.n_actions)
]
model.bg = spa.BasalGanglia(actions=spa.Actions(*actions))
model.thal = spa.Thalamus(model.bg)
def state_input(t):
if t < 0.1:
return 'S0'
else:
return '0'
model.input = spa.Input(state=state_input)
self.probe = nengo.Probe(model.thal.actions.output, synapse=0.03)
return model
def __init__(self, num_actions):
spa.SPA.__init__(self)
#Specify the modules to be used
self.cortex = spa.Buffer(dimensions=dimensions)
# Generate names for actions
def int_to_ascii(i):
return chr(ord("A") + i)
# Generate sequential action names
self.action_names = [
int_to_ascii((i / 26) % 26) + int_to_ascii(i % 26)
for i in range(num_actions)
]
# From this build action mapping
action_mappings = [
"dot(cortex, %s) --> cortex = %s" % (a, b)
for (a, b) in zip(self.action_names, self.action_names[1:])
]
action_mappings.append("dot(cortex, %s) --> cortex = %s" %
(self.action_names[-1], self.action_names[0]))
# Build SPA actions
actions = spa.Actions(*action_mappings)
self.bg = spa.BasalGanglia(actions=actions)
self.thal = spa.Thalamus(self.bg)
#Specify the input
def start(t):
if t < 0.05:
return self.action_names[0]
else:
return '0'
self.input = spa.Input(cortex=start)
def create_model():
#print trial_info
print('---- INTIALIZING MODEL ----')
global model
model = spa.SPA()
with model:
#display current stimulus pair (not part of model)
if nengo_gui_on and True:
model.pair_input = nengo.Node(present_pair)
model.pair_display = nengo.Node(
display_func,
size_in=model.pair_input.size_out) # to show input
nengo.Connection(model.pair_input,
model.pair_display,
synapse=None)
# control
model.control_net = nengo.Network()
with model.control_net:
#assuming the model knows which hand to use (which was blocked)
model.hand_input = nengo.Node(get_hand)
model.target_hand = spa.State(Dmid, vocab=vocab_motor, feedback=1)
nengo.Connection(model.hand_input,
model.target_hand.input,
synapse=None)
model.attend = spa.State(D, vocab=vocab_attend,
feedback=.5) # vocab_attend
model.goal = spa.State(Dlow, vocab=vocab_goal,
feedback=.7) # current goal
### vision ###
# set up network parameters
n_vis = X_train.shape[1] # nr of pixels, dimensions of network
n_hid = 1000 # nr of gabor encoders/neurons
# random state to start
rng = np.random.RandomState(9)
encoders = Gabor().generate(
n_hid, (4, 4), rng=rng) # gabor encoders, 11x11 apparently, why?
encoders = Mask(
(14, 90)).populate(encoders, rng=rng,
flatten=True) # use them on part of the image
model.visual_net = nengo.Network()
with model.visual_net:
#represent currently attended item
model.attended_item = nengo.Node(present_item2, size_in=D)
nengo.Connection(model.attend.output, model.attended_item)
model.vision_gabor = nengo.Ensemble(
n_hid,
n_vis,
eval_points=X_train,
neuron_type=nengo.LIF(),
intercepts=nengo.dists.Uniform(-0.1, 0.1),
#intercepts=nengo.dists.Choice([-0.5]), #should we comment this out? not sure what's happening
#max_rates=nengo.dists.Choice([100]),
encoders=encoders)
#recurrent connection (time constant 500 ms)
# strength = 1 - (100/500) = .8
zeros = np.zeros_like(X_train)
nengo.Connection(
model.vision_gabor,
model.vision_gabor,
synapse=0.005, #.1
eval_points=np.vstack(
[X_train, zeros,
np.random.randn(*X_train.shape)]),
transform=.5)
model.visual_representation = nengo.Ensemble(n_hid,
dimensions=Dmid)
model.visconn = nengo.Connection(
model.vision_gabor,
model.visual_representation,
synapse=0.005, #was .005
eval_points=X_train,
function=train_targets,
solver=nengo.solvers.LstsqL2(reg=0.01))
nengo.Connection(model.attended_item,
model.vision_gabor,
synapse=.02) #.03) #synapse?
# display attended item, only in gui
if nengo_gui_on:
# show what's being looked at
model.display_attended = nengo.Node(
display_func,
size_in=model.attended_item.size_out) # to show input
nengo.Connection(model.attended_item,
model.display_attended,
synapse=None)
#add node to plot total visual activity
model.visual_activation = nengo.Node(None, size_in=1)
nengo.Connection(model.vision_gabor.neurons,
model.visual_activation,
transform=np.ones((1, n_hid)),
synapse=None)
### central cognition ###
# concepts
model.concepts = spa.AssociativeMemory(
vocab_all_words, #vocab_concepts,
wta_output=True,
wta_inhibit_scale=1, #was 1
#default_output_key='NONE', #what to say if input doesn't match
threshold=0.3
) # how strong does input need to be for it to recognize
nengo.Connection(
model.visual_representation,
model.concepts.input,
transform=.8 * vision_mapping
) #not too fast to concepts, might have to be increased to have model react faster to first word.
#concepts accumulator
model.concepts_evidence = spa.State(
1, feedback=1, feedback_synapse=0.005
) #the lower the synapse, the faster it accumulates (was .1)
concepts_evidence_scale = 2.5
nengo.Connection(model.concepts.am.elem_output,
model.concepts_evidence.input,
transform=concepts_evidence_scale * np.ones(
(1, model.concepts.am.elem_output.size_out)),
synapse=0.005)
#concepts switch
model.do_concepts = spa.AssociativeMemory(vocab_reset,
default_output_key='CLEAR',
threshold=.2)
nengo.Connection(
model.do_concepts.am.ensembles[-1],
model.concepts_evidence.all_ensembles[0].neurons,
transform=np.ones(
(model.concepts_evidence.all_ensembles[0].n_neurons, 1)) * -10,
synapse=0.005)
# pair representation
model.vis_pair = spa.State(
D, vocab=vocab_all_words, feedback=1.0, feedback_synapse=.05
) #was 2, 1.6 works ok, but everything gets activated.
#learned words
model.dm_learned_words = spa.AssociativeMemory(
vocab_learned_words, threshold=.2
) #default_output_key='NONE' familiarity should be continuous over all items, so no wta
nengo.Connection(model.dm_learned_words.output,
model.dm_learned_words.input,
transform=.4,
synapse=.02)
# this stores the accumulated evidence for or against familiarity
model.familiarity = spa.State(
1, feedback=.9,
feedback_synapse=0.1) #fb syn influences speed of acc
familiarity_scale = 0.2 #keep stable for negative fam
#nengo.Connection(model.dm_learned_words.am.ensembles[-1], model.familiarity.input, transform=-(familiarity_scale+0.8)) #accumulate to -1
nengo.Connection(
model.dm_learned_words.am.elem_output,
model.familiarity.input, #am.element_output == all outputs, we sum
transform=(familiarity_scale + .1) * np.ones(
(1, model.dm_learned_words.am.elem_output.size_out))
) #accumulate to 1
model.do_fam = spa.AssociativeMemory(vocab_reset,
default_output_key='CLEAR',
threshold=.2)
nengo.Connection(
model.do_fam.am.ensembles[-1],
model.familiarity.all_ensembles[0].neurons,
transform=np.ones(
(model.familiarity.all_ensembles[0].n_neurons, 1)) * -10,
synapse=0.005)
#negative accumulator
nengo.Connection(
model.do_fam.am.elem_output,
model.familiarity.input,
transform=-familiarity_scale * np.ones(
(1, model.do_fam.am.elem_output.size_out))) #accumulate to -1
#fam model.dm_pairs = spa.AssociativeMemory(vocab_learned_pairs, input_keys=list_of_pairs,wta_output=True)
#fam nengo.Connection(model.dm_pairs.output,model.dm_pairs.input,transform=.5)
#this works:
#fam model.representation = spa.AssociativeMemory(vocab_learned_pairs, input_keys=list_of_pairs, wta_output=True)
#fam nengo.Connection(model.representation.output, model.representation.input, transform=2)
#fam model.rep_filled = spa.State(1,feedback_synapse=.005) #no fb syn specified
#fam nengo.Connection(model.representation.am.elem_output,model.rep_filled.input, #am.element_output == all outputs, we sum
#fam transform=.8*np.ones((1,model.representation.am.elem_output.size_out)),synapse=0)
#this doesn't:
#model.representation = spa.State(D,feedback=1)
#model.rep_filled = spa.State(1,feedback_synapse=.005) #no fb syn specified
#nengo.Connection(model.representation.output,model.rep_filled.input, #am.element_output == all outputs, we sum
# transform=.8*np.ones((1,model.representation.output.size_out)),synapse=0)
# this shouldn't really be fixed I think
#fam model.comparison = spa.Compare(D, vocab=vocab_concepts)
#motor
model.motor_net = nengo.Network()
with model.motor_net:
#input multiplier
model.motor_input = spa.State(Dmid, vocab=vocab_motor)
#higher motor area (SMA?)
model.motor = spa.State(Dmid, vocab=vocab_motor, feedback=.7)
#connect input multiplier with higher motor area
nengo.Connection(model.motor_input.output,
model.motor.input,
synapse=.1,
transform=2)
#finger area
model.fingers = spa.AssociativeMemory(
vocab_fingers,
input_keys=['L1', 'L2', 'R1', 'R2'],
wta_output=True)
nengo.Connection(model.fingers.output,
model.fingers.input,
synapse=0.1,
transform=0.3) #feedback
#conncetion between higher order area (hand, finger), to lower area
nengo.Connection(model.motor.output,
model.fingers.input,
transform=.25 * motor_mapping) #was .2
#finger position (spinal?)
model.finger_pos = nengo.networks.EnsembleArray(n_neurons=50,
n_ensembles=4)
nengo.Connection(model.finger_pos.output,
model.finger_pos.input,
synapse=0.1,
transform=0.8) #feedback
#connection between finger area and finger position
nengo.Connection(model.fingers.am.elem_output,
model.finger_pos.input,
transform=1.0 *
np.diag([0.55, .54, .56, .55])) #fix these
model.bg = spa.BasalGanglia(
spa.Actions(
#wait & start
a_aa_wait='dot(goal,WAIT) - .9 --> goal=0',
a_attend_item1=
'dot(goal,DO_TASK) - .1 --> goal=RECOG, attend=ITEM1, do_concepts=GO',
#attend words
b_attending_item1=
'dot(goal,RECOG) + dot(attend,ITEM1) - concepts_evidence - .3 --> goal=RECOG, attend=ITEM1, do_concepts=GO', # vis_pair=2.5*(ITEM1*concepts)',
c_attend_item2=
'dot(goal,RECOG) + dot(attend,ITEM1) + concepts_evidence - 1.6 --> goal=RECOG2, attend=ITEM2, vis_pair=3*(ITEM1*concepts)',
d_attending_item2=
'dot(goal,RECOG2+RECOG) + dot(attend,ITEM2) - concepts_evidence - .4 --> goal=RECOG2, attend=ITEM2, do_concepts=GO, dm_learned_words=1.0*(~ITEM1*vis_pair)', #vis_pair=1.2*(ITEM2*concepts)
e_judge_familiarity=
'dot(goal,RECOG2) + dot(attend,ITEM2) + concepts_evidence - 1.8 --> goal=FAMILIARITY, do_fam=GO, vis_pair=1.9*(ITEM2*concepts), dm_learned_words=2.0*(~ITEM1*vis_pair+~ITEM2*vis_pair)',
#judge familiarity
f_judge_familiarity=
'dot(goal,FAMILIARITY) - .1 --> goal=FAMILIARITY, do_fam=GO, dm_learned_words=.8*(~ITEM1*vis_pair+~ITEM2*vis_pair)',
g_respond_unfamiliar=
'dot(goal,FAMILIARITY) - familiarity - .5*dot(fingers,L1+L2+R1+R2) - .6 --> goal=RESPOND, do_fam=GO, motor_input=1.6*(target_hand+MIDDLE)',
h_respond_familiar=
'dot(goal,FAMILIARITY) + familiarity - .5*dot(fingers,L1+L2+R1+R2) - .6 --> goal=RESPOND, do_fam=GO, motor_input=1.6*(target_hand+INDEX)',
x_response_done=
'1.1*dot(goal,RESPOND) + 1.5*dot(fingers,L1+L2+R1+R2) - .7--> goal=2*END',
y_end='dot(goal,END)-.1 --> goal=END',
z_threshold='.05 --> goal=0'
#possible to match complete buffer, ie is representation filled?
# motor_input=1.5*target_hand+MIDDLE,
))
#'dot(attention, W1) - evidence - 0.8 --> motor=NO, attention=W1',
#'dot(attention, W1) + evidence - 0.8 --> attention=W2, reset=EVIDENCE',
#'dot(attention, W1) --> attention=W1', # if we don't set attention it goes back to 0
#'dot(attention, W2) - evidence - 0.8 --> motor=NO, attention=W2',
#'dot(attention, W2) + evidence - 0.8 --> motor=YES, attention=W2',
#'dot(attention, W2) --> attention=W2', # option might be feedback on attention, then no rule 3/6 but default rule
model.thalamus = spa.Thalamus(model.bg)
model.cortical = spa.Cortical( # cortical connection: shorthand for doing everything with states and connections
spa.Actions(
# 'motor_input = .04*target_hand',
# 'dm_learned_words = .1*vis_pair',
#'dm_pairs = 2*stimulus'
#'vis_pair = 2*attend*concepts+concepts',
#fam 'comparison_A = 2*vis_pair',
#fam 'comparison_B = 2*representation*~attend',
))
#probes
model.pr_motor_pos = nengo.Probe(
model.finger_pos.output,
synapse=.01) #raw vector (dimensions x time)
model.pr_motor = nengo.Probe(model.fingers.output, synapse=.01)
#model.pr_motor1 = nengo.Probe(model.motor.output, synapse=.01)
if not nengo_gui_on:
model.pr_vision_gabor = nengo.Probe(
model.vision_gabor.neurons, synapse=.005
) #do we need synapse, or should we do something with the spikes
model.pr_familiarity = nengo.Probe(
model.dm_learned_words.am.elem_output,
synapse=.01) #element output, don't include default
model.pr_concepts = nengo.Probe(
model.concepts.am.elem_output,
synapse=.01) # element output, don't include default
#multiply spikes with the connection weights
#input
model.input = spa.Input(goal=goal_func)
#print(sum(ens.n_neurons for ens in model.all_ensembles))
#return model
#to show select BG rules
# get names rules
if nengo_gui_on and False:
vocab_actions = spa.Vocabulary(model.bg.output.size_out)
for i, action in enumerate(model.bg.actions.actions):
vocab_actions.add(action.name.upper(),
np.eye(model.bg.output.size_out)[i])
model.actions = spa.State(model.bg.output.size_out,
vocab=vocab_actions)
nengo.Connection(model.thalamus.output, model.actions.input)
for net in model.networks:
if net.label is not None and net.label.startswith('channel'):
net.label = ''
def main():
model = spa.SPA(label="Vector Storage")
with model:
# Dimensionality of each representation
num_dimensions = 2
sub_dimensions = 2
# Create the vocabulary
vocab = Vocabulary(num_dimensions, randomize=False)
# Form the inputs
stored_value_1 = [1] * num_dimensions
stored_value_1 = [
s / np.linalg.norm(stored_value_1) for s in stored_value_1
]
vocab.add("Stored_value_1", stored_value_1)
stored_value_2 = [(-1**i) for i in range(num_dimensions)]
stored_value_2 = [
s / np.linalg.norm(stored_value_2) for s in stored_value_2
]
vocab.add("Stored_value_2", stored_value_2)
def first_input(t):
if t < 10:
return "Stored_value_2"
else:
return "Stored_value_1"
def second_input(t):
if t < 5:
return "Stored_value_1"
else:
return "Stored_value_2"
# Buffers to store the input
model.buffer1 = spa.Buffer(dimensions=num_dimensions,
subdimensions=sub_dimensions,
neurons_per_dimension=200,
direct=True)
model.buffer2 = spa.Buffer(dimensions=num_dimensions,
subdimensions=sub_dimensions,
neurons_per_dimension=200,
direct=True)
# Probe to visualize the values stored in the buffers
buffer_1_probe = nengo.Probe(model.buffer1.state.output)
buffer_2_probe = nengo.Probe(model.buffer2.state.output)
# Connect up the inputs
model.input = spa.Input(buffer1=first_input)
model.input = spa.Input(buffer2=second_input)
# Buffer to store the output
model.buffer3 = spa.Buffer(dimensions=num_dimensions,
subdimensions=sub_dimensions,
neurons_per_dimension=200,
direct=True)
buffer_3_probe = nengo.Probe(model.buffer3.state.output)
# Control system
actions = spa.Actions(
'dot(buffer1, Stored_value_2) --> buffer3=Stored_value_2',
'dot(buffer1, Stored_value_1) --> buffer3=Stored_value_1+Stored_value_2'
)
model.bg = spa.BasalGanglia(actions)
model.thalamus = spa.Thalamus(model.bg)
# Start the simulator
sim = nengo.Simulator(model)
# Dynamic plotting
plt.ion() # Dynamic updating of plots
fig = plt.figure(figsize=(15, 8))
plt.show()
ax = fig.gca()
ax.set_title("Vector Storage")
while True:
sim.run(1) # Run for an additional 1 second
plt.clf() # Clear the figure
plt.plot(sim.trange(),
similarity(sim.data, buffer_1_probe, vocab),
label="Buffer 1 Value") # Plot the entire dataset so far
plt.plot(sim.trange(),
similarity(sim.data, buffer_2_probe, vocab),
label="Buffer 2 Value")
plt.plot(sim.trange(),
similarity(sim.data, buffer_3_probe, vocab),
label="Buffer 3 Value")
print sim.data[buffer_1_probe][-1]
print sim.data[buffer_2_probe][-1]
print sim.data[buffer_3_probe][-1]
plt.legend(vocab.keys * 3, loc=2)
plt.draw() # Re-draw
def Spaun():
model = spa.SPA(label='Spaun', seed=cfg.seed)
with model:
model.config[nengo.Ensemble].max_rates = cfg.max_rates
model.config[nengo.Ensemble].neuron_type = cfg.neuron_type
model.config[nengo.Ensemble].n_neurons = cfg.n_neurons_ens
model.config[nengo.Connection].synapse = cfg.pstc
model.stim = Stimulus()
model.vis = Vision()
model.ps = ProdSys()
model.reward = RewardEval()
model.enc = InfoEnc()
model.mem = Memory()
model.trfm = TrfmSys()
model.dec = InfoDec()
model.mtr = Motor()
model.monitor = Monitor()
model.learn_conns = []
if hasattr(model, 'vis') and hasattr(model, 'ps'):
copy_draw_action = \
['0.5 * (dot(ps_task, X) + dot(vis, ZER)) --> ps_task = W, ps_state = TRANS0, ps_dec = FWD', # noqa
'dot(ps_task, W-DEC) - dot(vis, QM) --> ps_state = ps_state'] # noqa
recog_action = \
['0.5 * (dot(ps_task, X) + dot(vis, ONE)) --> ps_task = R, ps_state = TRANS0, ps_dec = FWD', # noqa
'dot(ps_task, R-DEC) - dot(vis, QM) --> ps_state = ps_state'] # noqa
learn_state_action = ['0.5 * (dot(ps_task, X) + dot(vis, TWO)) - dot(vis, QM) --> ps_task = L, ps_state = LEARN, ps_dec = FWD'] # noqa
# '0.5 * (dot(ps_task, L) + dot(vis, {:s} + {:s})) --> ps_state = LEARN'.format(*vocab.reward_sp_strs)] # noqa
if hasattr(model, 'reward'):
learn_action = ['0.5 * dot(ps_task, L) - dot(vis, QM) --> ps_action = %s, ps_state = LEARN, ps_dec = NONE' % # noqa
s for s in vocab.ps_action_learn_sp_strs]
else:
learn_action = []
mem_action = \
['0.5 * (dot(ps_task, X) + dot(vis, THR)) --> ps_task = M, ps_state = TRANS0, ps_dec = FWD', # noqa
'dot(ps_task, M-DEC) - dot(vis, F + R + QM) --> ps_state = ps_state', # noqa
'0.5 * (dot(ps_task, M-DEC) + dot(vis, F)) - dot(vis, QM) --> ps_dec = FWD', # noqa
'0.5 * (dot(ps_task, M-DEC) + dot(vis, R)) - dot(vis, QM) --> ps_dec = REV'] # noqa
if hasattr(model, 'trfm'):
count_action = \
['0.5 * (dot(ps_task, X) + dot(vis, FOR)) --> ps_task = C, ps_state = TRANS0, ps_dec = FWD', # noqa
'0.5 * (dot(ps_task, C-DEC) + dot(ps_state, TRANS0)) - dot(vis, QM) --> ps_state = CNT0', # noqa
'0.5 * (dot(ps_task, C-DEC) + dot(ps_state, CNT0)) - dot(vis, QM) --> ps_state = CNT1', # noqa
'(0.25 * (dot(ps_task, DEC) + dot(ps_state, CNT1)) + 0.5 * dot(trfm_compare, NO_MATCH)) + (dot(ps_dec, CNT) - 1) - dot(vis, QM) --> ps_dec = CNT, ps_state = CNT1', # noqa
'(0.25 * (dot(ps_task, DEC) + dot(ps_state, CNT1)) + 0.5 * dot(trfm_compare, MATCH)) + (dot(ps_dec, CNT) - 1) - dot(vis, QM) --> ps_dec = FWD, ps_state = TRANS0'] # noqa
qa_action = \
['0.5 * (dot(ps_task, X) + dot(vis, FIV)) --> ps_task = A, ps_state = TRANS0, ps_dec = FWD', # noqa
'dot(ps_task, A-DEC) - dot(vis, K + P + QM) --> ps_state = ps_state', # noqa
'0.5 * (dot(ps_task, A-DEC) + dot(vis, K)) - dot(vis, QM) --> ps_state = QAK', # noqa
'0.5 * (dot(ps_task, A-DEC) + dot(vis, P)) - dot(vis, QM) --> ps_state = QAP'] # noqa
rvc_action = \
['0.5 * (dot(ps_task, X) + dot(vis, SIX)) --> ps_task = V, ps_state = TRANS0, ps_dec = FWD', # noqa
'0.5 * (dot(ps_task, V-DEC) + dot(ps_state, TRANS0)) - dot(vis, QM) --> ps_state = TRANS1', # noqa
'0.5 * (dot(ps_task, V-DEC) + dot(ps_state, TRANS1)) - dot(vis, QM) --> ps_state = TRANS0'] # noqa
fi_action = \
['0.5 * (dot(ps_task, X) + dot(vis, SEV)) --> ps_task = F, ps_state = TRANS0, ps_dec = FWD', # noqa
'0.5 * (dot(ps_task, F-DEC) + dot(ps_state, TRANS0)) - dot(vis, QM) --> ps_state = TRANS1', # noqa
'0.5 * (dot(ps_task, F-DEC) + dot(ps_state, TRANS1)) - dot(vis, QM) --> ps_state = TRANS2', # noqa
'0.5 * (dot(ps_task, F-DEC) + dot(ps_state, TRANS2)) - dot(vis, QM) --> ps_state = TRANS0'] # noqa
else:
count_action = []
qa_action = []
rvc_action = []
fi_action = []
decode_action = \
['dot(vis, QM) - 0.75 * dot(ps_task, W + C + V + F + L) --> ps_task = DEC, ps_state = ps_state, ps_dec = ps_dec', # noqa
'0.5 * (dot(vis, QM) + dot(ps_task, W)) --> ps_task = DEC, ps_state = ps_state, ps_dec = DECW', # noqa
'0.5 * (dot(vis, QM) + dot(ps_task, C)) --> ps_task = DEC, ps_state = ps_state, ps_dec = CNT', # noqa
'0.5 * (dot(vis, QM) + dot(ps_task, V + F)) --> ps_task = DEC, ps_state = ps_state, ps_dec = DECI', # noqa
'0.5 * (dot(vis, QM) + dot(ps_task, L)) --> ps_task = L, ps_state = LEARN, ps_dec = FWD', # noqa
'dot(ps_task, DEC) - dot(ps_dec, CNT) - dot(vis, QM + A) --> ps_task = ps_task, ps_state = ps_state, ps_dec = ps_dec'] # noqa
default_action = \
[]
# List learning task spa actions first, so we know the precise
# indicies of the learning task actions (i.e. the first N)
all_actions = (learn_action + learn_state_action +
copy_draw_action + recog_action + mem_action +
count_action + qa_action + rvc_action + fi_action +
decode_action + default_action)
actions = spa.Actions(*all_actions)
model.bg = spa.BasalGanglia(actions=actions, input_synapse=0.008,
label='Basal Ganglia')
model.thal = spa.Thalamus(model.bg, subdim_channel=1,
mutual_inhibit=1, route_inhibit=5.0,
label='Thalamus')
# ----- Set up connections (and save record of modules) -----
if hasattr(model, 'vis'):
model.vis.setup_connections(model)
if hasattr(model, 'ps'):
model.ps.setup_connections(model)
if hasattr(model, 'bg'):
if hasattr(model, 'reward'):
with model.bg:
# Generate random biases for each learn action, so that
# there is some randomness to the initial action choice
bias_node = nengo.Node(1)
bias_ens = nengo.Ensemble(cfg.n_neurons_ens, 1,
label='BG Bias Ensemble')
nengo.Connection(bias_node, bias_ens)
for i in range(len(learn_action)):
init_trfm = (np.random.random() *
cfg.learn_init_trfm_max)
trfm_val = cfg.learn_init_trfm_bias + init_trfm
model.learn_conns.append(
nengo.Connection(bias_ens, model.bg.input[i],
transform=trfm_val))
cfg.learn_init_transforms.append(trfm_val)
logger.write("# learn_init_trfms: %s\n" %
(str(cfg.learn_init_transforms)))
if hasattr(model, 'thal'):
pass
if hasattr(model, 'reward'):
model.reward.setup_connections(model, model.learn_conns)
if hasattr(model, 'enc'):
model.enc.setup_connections(model)
if hasattr(model, 'mem'):
model.mem.setup_connections(model)
if hasattr(model, 'trfm'):
model.trfm.setup_connections(model)
if hasattr(model, 'dec'):
model.dec.setup_connections(model)
if hasattr(model, 'mtr'):
model.mtr.setup_connections(model)
if hasattr(model, 'monitor'):
model.monitor.setup_connections(model)
return model
def _test_sequence(Simulator, plt, seed, outfile, prune_passthrough):
dimensions = 32
subdimensions = 16
T = 4.0
seq_length = 6
with spa.SPA(seed=seed) as model:
model.state = spa.Memory(dimensions=dimensions,
subdimensions=subdimensions)
seq_actions = [
'dot(state,A%d) --> state=A%d' % (i, (i + 1) % seq_length)
for i in range(seq_length)
]
model.bg = spa.BasalGanglia(spa.Actions(*seq_actions))
model.thal = spa.Thalamus(model.bg)
def stim_state(t):
if t < 0.1:
return 'A0'
else:
return '0'
model.input = spa.Input(state=stim_state)
p_state = nengo.Probe(model.state.state.output, synapse=0.01)
if 'spinnaker' in Simulator.__module__:
nengo_spinnaker.add_spinnaker_params(model.config)
model.config[
model.input.input_nodes['state']].function_of_time = True
vocab = model.get_input_vocab('state')
if prune_passthrough:
model = remove_passthrough_nodes(model)
sim = Simulator(model)
sim.run(T)
t = sim.trange()
data = sim.data[p_state]
ideal = np.array([vocab.parse('A%d' % i).v for i in range(seq_length)])
dotp = np.dot(data, ideal.T)
best = np.argmax(dotp, axis=1)
delta = np.diff(best)
indexes = np.where(delta != 0)
# [:, 1:] ignores the first transition, which is meaningless
delta_t = np.diff(indexes)[:, 1:] * 0.001
mean = np.mean(delta_t)
std = np.std(delta_t)
outfile.write('"n_neurons": %d,\n' % sum(e.n_neurons
for e in model.all_ensembles))
outfile.write('"simtime": %f,\n' % T)
outfile.write('"timing_mean": %f,\n' % mean)
outfile.write('"timing_std": %f,\n' % std)
figsize = (onecolumn, 4.0) if horizontal else (onecolumn * 2, 3.0)
setup(figsize=figsize,
palette_args={
'palette': "cubehelix",
'n_colors': 6
})
plt.plot(t[t < 1.0], dotp[t < 1.0])
for transition in t[indexes[0]]:
plt.axvline(transition, c='0.5', lw=1, ls=':')
plt.ylabel('Similarity to item')
plt.xlabel('Time (s)')
plt.xlim(right=1.0)
sns.despine()
if prune_passthrough:
plt.saveas = 'results-4.svg'
else:
plt.saveas = 'results-5.svg'
if hasattr(sim, 'close'):
sim.close()
def main():
model = spa.SPA(label="Vector Storage")
with model:
# Dimensionality of each representation
num_dimensions = 30
sub_dimensions = 1
# Create the vocabulary
vocab = Vocabulary(num_dimensions, randomize=False)
# Form the inputs manually by directly defining their vectors
stored_value_1 = np.random.rand(num_dimensions) - [0.5
] * num_dimensions
stored_value_1 = [
s / np.linalg.norm(stored_value_1) for s in stored_value_1
]
vocab.add("Stored_value_1", stored_value_1)
stored_value_2 = np.random.rand(num_dimensions) - [0.5
] * num_dimensions
stored_value_2 = [
s / np.linalg.norm(stored_value_2) for s in stored_value_2
]
vocab.add("Stored_value_2", stored_value_2)
stored_value_3 = np.random.rand(num_dimensions) - [0.5
] * num_dimensions
stored_value_3 = [
s / np.linalg.norm(stored_value_3) for s in stored_value_3
]
vocab.add("Stored_value_3", stored_value_3)
# Create a semantic pointer corresponding to the "correct" answer for the operation
sum_vector = np.subtract(np.add(stored_value_1, stored_value_2),
stored_value_3)
sum_vector = sum_vector / np.linalg.norm(sum_vector)
vocab.add("Sum", sum_vector)
# Define the control signal inputs as random vectors
r1 = [1] * num_dimensions
r1 = r1 / np.linalg.norm(r1)
r2 = [(-1)**k for k in range(num_dimensions)]
r2 = r2 / np.linalg.norm(r2)
vocab.add("Hold_signal", r1)
vocab.add("Start_signal", r2)
# Control when the vector operation takes place
def control_input(t):
if t < 1:
return "Hold_signal"
else:
return "Start_signal"
# inputs to the input word buffers
def first_input(t):
return "Stored_value_1"
def second_input(t):
return "Stored_value_2"
def third_input(t):
return "Stored_value_3"
# Control buffer
model.control = spa.Buffer(dimensions=num_dimensions,
subdimensions=sub_dimensions,
neurons_per_dimension=200,
direct=True,
vocab=vocab)
control_probe = nengo.Probe(model.control.state.output)
# Buffers to store the inputs: e.g., King, Man, Woman
model.word_buffer1 = spa.Buffer(dimensions=num_dimensions,
subdimensions=sub_dimensions,
neurons_per_dimension=200,
direct=True,
vocab=vocab)
model.word_buffer2 = spa.Buffer(dimensions=num_dimensions,
subdimensions=sub_dimensions,
neurons_per_dimension=200,
direct=True,
vocab=vocab)
model.word_buffer3 = spa.Buffer(dimensions=num_dimensions,
subdimensions=sub_dimensions,
neurons_per_dimension=200,
direct=True,
vocab=vocab)
# Probe to visualize the values stored in the buffers
buffer_1_probe = nengo.Probe(model.word_buffer1.state.output)
buffer_2_probe = nengo.Probe(model.word_buffer2.state.output)
buffer_3_probe = nengo.Probe(model.word_buffer3.state.output)
# Buffer to hold the result: e.g. Queen
model.result = spa.Buffer(dimensions=num_dimensions,
subdimensions=sub_dimensions,
neurons_per_dimension=200,
direct=True,
vocab=vocab)
result_probe = nengo.Probe(model.result.state.output)
# Control system
actions = spa.Actions(
'dot(control, Start_signal) --> result = word_buffer1 + word_buffer2 - word_buffer3'
)
model.bg = spa.BasalGanglia(actions)
model.thalamus = spa.Thalamus(model.bg, subdim_channel=sub_dimensions)
# Connect up the inputs
model.input = spa.Input(control=control_input,
word_buffer1=first_input,
word_buffer2=second_input,
word_buffer3=third_input)
# Start the simulator
sim = nengo.Simulator(model)
# Dynamic plotting
plt.ion() # Dynamic updating of plots
fig = plt.figure(figsize=(15, 8))
plt.show()
ax = fig.gca()
ax.set_title("Vector Storage")
while True:
sim.run(1) # Run for an additional 1 second
plt.clf() # Clear the figure
plt.plot(sim.trange(),
similarity(sim.data, result_probe, vocab),
label="Buffer 3 Value")
plt.legend(vocab.keys * 3, loc=2)
plt.draw() # Re-draw
print sim.data[buffer_1_probe][-1]
print sim.data[buffer_2_probe][-1]
print sim.data[buffer_3_probe][-1]
print sim.data[result_probe][-1]
print "\n"
plt.show()
e_store_result_and_update_count = 'dot(goal,COUNT) + letter_memory_status - .4 --> goal=COUNT, answer=%i*~NEXT*letter_memory, number_memory=6*IDENTITY*count' % store_weight,
f_store_count_and_compare = 'dot(goal,COUNT) + 1.5*number_memory_status - .4 --> goal=1.4*COMPARE_COUNT+.6*COUNT, count=%i*~NEXT*number_memory, comparison_cleanA=arg2, comparison_cleanB=~NEXT*number_memory' % store_weight,
g_update_result = 'dot(goal,COMPARE_COUNT) - 2*comparison_status -.9 --> goal=COUNT, letter_memory=3*IDENTITY*answer',
h_answer_counted = 'dot(goal,COMPARE_COUNT) + 2*comparison_status - .3 --> goal=1.4*COMPARE-COUNT',
#Compare and respond
v_compare = 'dot(goal,COMPARE) --> goal=(.8*COMPARE)+RESPOND, comparison2_cleanA=2*target, comparison2_cleanB=2*answer, do_learn=GO, in_layer=ITEM1*arg1+ITEM2*arg2+RESULT*answer, correct=ITEM1*arg1+ITEM2*arg2+RESULT*answer',
w_respond_match = 'dot(goal,RESPOND) + comparison2_status - 0.3 --> goal=DONE-COMPARE, motor=YES',
x_respond_mismatch = 'dot(goal,RESPOND) - comparison2_status - 0.3 --> goal=DONE-COMPARE, motor=NO',
y_done = 'dot(goal,DONE) + dot(motor,YES+NO) - .5 --> goal=2*DONE',
z_threshold = '0.1 -->'
)
model.bg = spa.BasalGanglia(actions)
model.thalamus = spa.Thalamus(model.bg, synapse_channel = .04)
#to show BG/thalamus rules
vocab_actions = spa.Vocabulary(model.bg.output.size_out)
for i, action in enumerate(model.bg.actions.actions):
vocab_actions.add(action.name.upper(), np.eye(model.bg.output.size_out)[i])
model.actions = spa.State(model.bg.output.size_out,subdimensions=model.bg.output.size_out,
vocab=vocab_actions)
nengo.Connection(model.thalamus.output, model.actions.input,synapse=None)
for net in model.networks:
if net.label is not None and net.label.startswith('channel'):
net.label = ''
def __init__(self):
self.scalar = spa.Buffer(dimensions=16, subdimensions=1)
actions = spa.Actions('0.5 --> scalar=dot(scalar, FOO)')
self.bg = spa.BasalGanglia(actions)
self.thalamus = spa.Thalamus(self.bg)
def create_model():
#print trial_info
print('---- INTIALIZING MODEL ----')
global model
model = spa.SPA()
with model:
#display current stimulus pair (not part of model)
if nengo_gui_on and True:
model.pair_input = nengo.Node(present_pair)
model.pair_display = nengo.Node(
display_func,
size_in=model.pair_input.size_out) # to show input
nengo.Connection(model.pair_input,
model.pair_display,
synapse=None)
# control
model.control_net = nengo.Network()
with model.control_net:
#assuming the model knows which hand to use (which was blocked)
model.hand_input = nengo.Node(get_hand)
model.target_hand = spa.State(Dmid, vocab=vocab_motor, feedback=1)
nengo.Connection(model.hand_input,
model.target_hand.input,
synapse=None)
model.attend = spa.State(D, vocab=vocab_attend,
feedback=.5) # vocab_attend
model.goal = spa.State(Dlow, vocab=vocab_goal,
feedback=.7) # current goal
### vision ###
# set up network parameters
n_vis = X_train.shape[1] # nr of pixels, dimensions of network
n_hid = 1000 # nr of gabor encoders/neurons
# random state to start
rng = np.random.RandomState(9)
encoders = Gabor().generate(
n_hid, (4, 4), rng=rng) # gabor encoders, 11x11 apparently, why?
encoders = Mask(
(14, 90)).populate(encoders, rng=rng,
flatten=True) # use them on part of the image
model.visual_net = nengo.Network()
with model.visual_net:
#represent currently attended item
model.attended_item = nengo.Node(present_item2, size_in=D)
nengo.Connection(model.attend.output, model.attended_item)
model.vision_gabor = nengo.Ensemble(
n_hid,
n_vis,
eval_points=X_train,
# neuron_type=nengo.LIF(),
neuron_type=nengo.AdaptiveLIF(
tau_n=.01, inc_n=.05
), #to get a better fit, use more realistic neurons that adapt to input
intercepts=nengo.dists.Uniform(-0.1, 0.1),
#intercepts=nengo.dists.Choice([-0.5]), #should we comment this out? not sure what's happening
#max_rates=nengo.dists.Choice([100]),
encoders=encoders)
#recurrent connection (time constant 500 ms)
# strength = 1 - (100/500) = .8
zeros = np.zeros_like(X_train)
nengo.Connection(
model.vision_gabor,
model.vision_gabor,
synapse=0.005, #.1
eval_points=np.vstack(
[X_train, zeros,
np.random.randn(*X_train.shape)]),
transform=.5)
model.visual_representation = nengo.Ensemble(n_hid,
dimensions=Dmid)
model.visconn = nengo.Connection(
model.vision_gabor,
model.visual_representation,
synapse=0.005, #was .005
eval_points=X_train,
function=train_targets,
solver=nengo.solvers.LstsqL2(reg=0.01))
nengo.Connection(model.attended_item,
model.vision_gabor,
synapse=.02) #.03) #synapse?
# display attended item, only in gui
if nengo_gui_on:
# show what's being looked at
model.display_attended = nengo.Node(
display_func,
size_in=model.attended_item.size_out) # to show input
nengo.Connection(model.attended_item,
model.display_attended,
synapse=None)
#add node to plot total visual activity
model.visual_activation = nengo.Node(None, size_in=1)
nengo.Connection(model.vision_gabor.neurons,
model.visual_activation,
transform=np.ones((1, n_hid)),
synapse=None)
### central cognition ###
##### Concepts #####
model.concepts = spa.AssociativeMemory(
vocab_all_words, #vocab_concepts,
wta_output=True,
wta_inhibit_scale=1, #was 1
#default_output_key='NONE', #what to say if input doesn't match
threshold=0.3
) # how strong does input need to be for it to recognize
nengo.Connection(
model.visual_representation,
model.concepts.input,
transform=.8 * vision_mapping
) #not too fast to concepts, might have to be increased to have model react faster to first word.
#concepts accumulator
model.concepts_evidence = spa.State(
1, feedback=1, feedback_synapse=0.005
) #the lower the synapse, the faster it accumulates (was .1)
concepts_evidence_scale = 2.5
nengo.Connection(model.concepts.am.elem_output,
model.concepts_evidence.input,
transform=concepts_evidence_scale * np.ones(
(1, model.concepts.am.elem_output.size_out)),
synapse=0.005)
#concepts switch
model.do_concepts = spa.AssociativeMemory(vocab_reset,
default_output_key='CLEAR',
threshold=.2)
nengo.Connection(
model.do_concepts.am.ensembles[-1],
model.concepts_evidence.all_ensembles[0].neurons,
transform=np.ones(
(model.concepts_evidence.all_ensembles[0].n_neurons, 1)) * -10,
synapse=0.005)
###### Visual Representation ######
model.vis_pair = spa.State(
D, vocab=vocab_all_words, feedback=1.0, feedback_synapse=.05
) #was 2, 1.6 works ok, but everything gets activated.
##### Familiarity #####
# Assoc Mem with Learned Words
# - familiarity signal should be continuous over all items, so no wta
model.dm_learned_words = spa.AssociativeMemory(vocab_learned_words,
threshold=.2)
nengo.Connection(model.dm_learned_words.output,
model.dm_learned_words.input,
transform=.4,
synapse=.02)
# Familiarity Accumulator
model.familiarity = spa.State(
1, feedback=.9,
feedback_synapse=0.1) #fb syn influences speed of acc
#familiarity_scale = 0.2 #keep stable for negative fam
# familiarity accumulator switch
model.do_fam = spa.AssociativeMemory(vocab_reset,
default_output_key='CLEAR',
threshold=.2)
# reset
nengo.Connection(
model.do_fam.am.ensembles[-1],
model.familiarity.all_ensembles[0].neurons,
transform=np.ones(
(model.familiarity.all_ensembles[0].n_neurons, 1)) * -10,
synapse=0.005)
#first a sum to represent summed similarity
model.summed_similarity = nengo.Ensemble(n_neurons=100, dimensions=1)
nengo.Connection(
model.dm_learned_words.am.elem_output,
model.summed_similarity,
transform=np.ones(
(1,
model.dm_learned_words.am.elem_output.size_out))) #take sum
#then a connection to accumulate this summed sim
def familiarity_acc_transform(summed_sim):
fam_scale = .5
fam_threshold = 0 #actually, kind of bias
fam_max = 1
return fam_scale * (2 * ((summed_sim - fam_threshold) /
(fam_max - fam_threshold)) - 1)
nengo.Connection(model.summed_similarity,
model.familiarity.input,
function=familiarity_acc_transform)
##### Recollection & Representation #####
model.dm_pairs = spa.AssociativeMemory(
vocab_learned_pairs, wta_output=True) #input_keys=list_of_pairs
nengo.Connection(model.dm_pairs.output,
model.dm_pairs.input,
transform=.5,
synapse=.05)
#representation
rep_scale = 0.5
model.representation = spa.State(D,
vocab=vocab_all_words,
feedback=1.0)
model.rep_filled = spa.State(
1, feedback=.9,
feedback_synapse=.1) #fb syn influences speed of acc
model.do_rep = spa.AssociativeMemory(vocab_reset,
default_output_key='CLEAR',
threshold=.2)
nengo.Connection(
model.do_rep.am.ensembles[-1],
model.rep_filled.all_ensembles[0].neurons,
transform=np.ones(
(model.rep_filled.all_ensembles[0].n_neurons, 1)) * -10,
synapse=0.005)
nengo.Connection(model.representation.output,
model.rep_filled.input,
transform=rep_scale *
np.reshape(sum(vocab_learned_pairs.vectors),
((1, D))))
###### Comparison #####
model.comparison = spa.Compare(D,
vocab=vocab_all_words,
neurons_per_multiply=500,
input_magnitude=.3)
#turns out comparison is not an accumulator - we also need one of those.
model.comparison_accumulator = spa.State(
1, feedback=.9,
feedback_synapse=0.05) #fb syn influences speed of acc
model.do_compare = spa.AssociativeMemory(vocab_reset,
default_output_key='CLEAR',
threshold=.2)
#reset
nengo.Connection(
model.do_compare.am.ensembles[-1],
model.comparison_accumulator.all_ensembles[0].neurons,
transform=np.ones(
(model.comparison_accumulator.all_ensembles[0].n_neurons, 1)) *
-10,
synapse=0.005)
#error because we apply a function to a 'passthrough' node, inbetween ensemble as a solution:
model.comparison_result = nengo.Ensemble(n_neurons=100, dimensions=1)
nengo.Connection(model.comparison.output, model.comparison_result)
def comparison_acc_transform(comparison):
comparison_scale = .6
comparison_threshold = 0 #actually, kind of bias
comparison_max = .6
return comparison_scale * (2 * (
(comparison - comparison_threshold) /
(comparison_max - comparison_threshold)) - 1)
nengo.Connection(model.comparison_result,
model.comparison_accumulator.input,
function=comparison_acc_transform)
#motor
model.motor_net = nengo.Network()
with model.motor_net:
#input multiplier
model.motor_input = spa.State(Dmid, vocab=vocab_motor)
#higher motor area (SMA?)
model.motor = spa.State(Dmid, vocab=vocab_motor, feedback=.7)
#connect input multiplier with higher motor area
nengo.Connection(model.motor_input.output,
model.motor.input,
synapse=.1,
transform=2)
#finger area
model.fingers = spa.AssociativeMemory(
vocab_fingers,
input_keys=['L1', 'L2', 'R1', 'R2'],
wta_output=True)
nengo.Connection(model.fingers.output,
model.fingers.input,
synapse=0.1,
transform=0.3) #feedback
#conncetion between higher order area (hand, finger), to lower area
nengo.Connection(model.motor.output,
model.fingers.input,
transform=.25 * motor_mapping) #was .2
#finger position (spinal?)
model.finger_pos = nengo.networks.EnsembleArray(n_neurons=50,
n_ensembles=4)
nengo.Connection(model.finger_pos.output,
model.finger_pos.input,
synapse=0.1,
transform=0.8) #feedback
#connection between finger area and finger position
nengo.Connection(model.fingers.am.elem_output,
model.finger_pos.input,
transform=1.0 *
np.diag([0.55, .54, .56, .55])) #fix these
model.bg = spa.BasalGanglia(
spa.Actions(
#wait & start
a_aa_wait='dot(goal,WAIT) - .9 --> goal=0',
a_attend_item1=
'dot(goal,DO_TASK) - .0 --> goal=RECOG, attend=ITEM1, do_concepts=GO',
#attend words
b_attending_item1=
'dot(goal,RECOG) + dot(attend,ITEM1) - concepts_evidence - .3 --> goal=RECOG, attend=ITEM1, do_concepts=GO', # vis_pair=2.5*(ITEM1*concepts)',
c_attend_item2=
'dot(goal,RECOG) + dot(attend,ITEM1) + concepts_evidence - 1.6 --> goal=RECOG2, attend=ITEM2, vis_pair=3*(ITEM1*concepts)',
d_attending_item2=
'dot(goal,RECOG2+RECOG) + dot(attend,ITEM2) - concepts_evidence - .4 --> goal=RECOG2, attend=ITEM2, do_concepts=GO, dm_learned_words=1.0*(~ITEM1*vis_pair)', #vis_pair=1.2*(ITEM2*concepts)
e_start_familiarity=
'dot(goal,RECOG2) + dot(attend,ITEM2) + concepts_evidence - 1.8 --> goal=FAMILIARITY, do_fam=GO, vis_pair=1.9*(ITEM2*concepts), dm_learned_words=2.0*(~ITEM1*vis_pair+~ITEM2*vis_pair)',
#judge familiarity
f_accumulate_familiarity=
'1.1*dot(goal,FAMILIARITY) - 0.2 --> goal=FAMILIARITY-RECOG2, do_fam=GO, dm_learned_words=.8*(~ITEM1*vis_pair+~ITEM2*vis_pair)',
g_respond_unfamiliar=
'dot(goal,FAMILIARITY) - familiarity - .5*dot(fingers,L1+L2+R1+R2) - .6 --> goal=RESPOND_MISMATCH-FAMILIARITY, do_fam=GO, motor_input=1.6*(target_hand+MIDDLE)',
#g2_respond_familiar = 'dot(goal,FAMILIARITY) + familiarity - .5*dot(fingers,L1+L2+R1+R2) - .6 --> goal=RESPOND, do_fam=GO, motor_input=1.6*(target_hand+INDEX)',
#recollection & representation
h_recollection=
'dot(goal,FAMILIARITY) + familiarity - .5*dot(fingers,L1+L2+R1+R2) - .6 --> goal=RECOLLECTION-FAMILIARITY, dm_pairs = vis_pair',
i_representation=
'dot(goal,RECOLLECTION) - rep_filled - .1 --> goal=RECOLLECTION, dm_pairs = vis_pair, representation=3*dm_pairs, do_rep=GO',
#comparison & respond
j_10_compare_word1=
'dot(goal,RECOLLECTION+1.4*COMPARE_ITEM1) + rep_filled - .9 --> goal=COMPARE_ITEM1-RECOLLECTION, do_rep=GO, do_compare=GO, comparison_A = ~ITEM1*vis_pair, comparison_B = ~ITEM1*representation',
k_11_match_word1=
'dot(goal,COMPARE_ITEM1) + comparison_accumulator - .7 --> goal=COMPARE_ITEM2-COMPARE_ITEM1, do_rep=GO, comparison_A = ~ITEM1*vis_pair, comparison_B = ~ITEM1*representation',
l_12_mismatch_word1=
'dot(goal,COMPARE_ITEM1) + .4 * dot(goal,RESPOND_MISMATCH) - comparison_accumulator - .7 --> goal=RESPOND_MISMATCH-COMPARE_ITEM1, do_rep=GO, motor_input=1.6*(target_hand+MIDDLE), do_compare=GO, comparison_A = ~ITEM1*vis_pair, comparison_B = ~ITEM1*representation',
compare_word2=
'dot(goal,COMPARE_ITEM2) - .5 --> goal=COMPARE_ITEM2, do_compare=GO, comparison_A = ~ITEM2*vis_pair, comparison_B = ~ITEM2*representation',
m_match_word2=
'dot(goal,COMPARE_ITEM2) + comparison_accumulator - .7 --> goal=RESPOND_MATCH-COMPARE_ITEM2, motor_input=1.6*(target_hand+INDEX), do_compare=GO, comparison_A = ~ITEM2*vis_pair, comparison_B = ~ITEM2*representation',
n_mismatch_word2=
'dot(goal,COMPARE_ITEM2) - comparison_accumulator - dot(fingers,L1+L2+R1+R2)- .7 --> goal=RESPOND_MISMATCH-COMPARE_ITEM2, motor_input=1.6*(target_hand+MIDDLE),do_compare=GO, comparison_A = ~ITEM2*vis_pair, comparison_B = ~ITEM2*representation',
#respond
o_respond_match=
'dot(goal,RESPOND_MATCH) - .1 --> goal=RESPOND_MATCH, motor_input=1.6*(target_hand+INDEX)',
p_respond_mismatch=
'dot(goal,RESPOND_MISMATCH) - .1 --> goal=RESPOND_MISMATCH, motor_input=1.6*(target_hand+MIDDLE)',
#finish
x_response_done=
'dot(goal,RESPOND_MATCH) + dot(goal,RESPOND_MISMATCH) + 2*dot(fingers,L1+L2+R1+R2) - .7 --> goal=2*END',
y_end=
'dot(goal,END)-.1 --> goal=END-RESPOND_MATCH-RESPOND_MISMATCH',
z_threshold='.05 --> goal=0'
#possible to match complete buffer, ie is representation filled?
# motor_input=1.5*target_hand+MIDDLE,
))
print(model.bg.actions.count)
#print(model.bg.dimensions)
model.thalamus = spa.Thalamus(model.bg)
model.cortical = spa.Cortical( # cortical connection: shorthand for doing everything with states and connections
spa.Actions(
# 'motor_input = .04*target_hand',
# 'dm_learned_words = .1*vis_pair',
#'dm_pairs = 2*stimulus'
#'vis_pair = 2*attend*concepts+concepts',
#fam 'comparison_A = 2*vis_pair',
#fam 'comparison_B = 2*representation*~attend',
))
#probes
model.pr_motor_pos = nengo.Probe(
model.finger_pos.output,
synapse=.01) #raw vector (dimensions x time)
model.pr_motor = nengo.Probe(model.fingers.output, synapse=.01)
#model.pr_motor1 = nengo.Probe(model.motor.output, synapse=.01)
if not nengo_gui_on:
model.pr_vision_gabor = nengo.Probe(
model.vision_gabor.neurons, synapse=.005
) #do we need synapse, or should we do something with the spikes
model.pr_familiarity = nengo.Probe(
model.dm_learned_words.am.elem_output,
synapse=.01) #element output, don't include default
model.pr_concepts = nengo.Probe(
model.concepts.am.elem_output,
synapse=.01) # element output, don't include default
#multiply spikes with the connection weights
#input
model.input = spa.Input(goal=goal_func)
#print(sum(ens.n_neurons for ens in model.all_ensembles))
#return model
#to show select BG rules
# get names rules
if nengo_gui_on:
vocab_actions = spa.Vocabulary(model.bg.output.size_out)
for i, action in enumerate(model.bg.actions.actions):
vocab_actions.add(action.name.upper(),
np.eye(model.bg.output.size_out)[i])
model.actions = spa.State(model.bg.output.size_out,
subdimensions=model.bg.output.size_out,
vocab=vocab_actions)
nengo.Connection(model.thalamus.output, model.actions.input)
for net in model.networks:
if net.label is not None and net.label.startswith('channel'):
net.label = ''
def run(self):
n_neurons = 40
subdim = 16
direct = False
output = self.stimuli.output # vector output from coherence score
threshold = self.stimuli.threshold # for correct answer on inf task
g_transform = np.transpose(np.ones((1, self.dimensions))) # mem gate
model = spa.SPA(label='ConceptModel', seed=self.seed)
with model:
# Vision Component
model.vision = spa.Buffer(dimensions=self.dimensions,
subdimensions=subdim,
neurons_per_dimension=n_neurons,
vocab=self.main_voc,
direct=direct)
# Memory Component
model.sp_mem = spa.Buffer(dimensions=self.dimensions,
subdimensions=subdim,
neurons_per_dimension=n_neurons,
vocab=self.main_voc,
direct=direct)
model.context_mem = spa.Memory(dimensions=self.dimensions,
subdimensions=subdim,
neurons_per_dimension=n_neurons,
vocab=self.main_voc,
tau=0.01 / 0.3,
direct=direct)
# Inferential Evaluation Subsystem
model.inference = spa.Buffer(dimensions=self.dimensions,
subdimensions=subdim,
neurons_per_dimension=n_neurons,
vocab=self.feat_voc,
direct=direct)
model.score = spa.Memory(dimensions=1,
neurons_per_dimension=3000,
synapse=0.05,
direct=direct)
model.gatesignal = spa.Memory(dimensions=1,
neurons_per_dimension=500,
synapse=0.05,
direct=direct)
model.cleanup1 = spa.AssociativeMemory(
input_vocab=self.feat_voc,
output_vocab=self.weight_voc,
n_neurons_per_ensemble=250,
threshold=0.25)
# Perceptual Evaluation Subsystem
model.cleanup2 = spa.AssociativeMemory(input_vocab=self.label_voc,
output_vocab=self.label_voc,
n_neurons_per_ensemble=1000,
threshold=threshold)
# Shared Component
model.decision = spa.Memory(dimensions=self.dimensions,
subdimensions=subdim,
neurons_per_dimension=n_neurons,
vocab=self.label_voc,
tau=0.01 / 0.3,
synapse=0.05,
direct=direct)
# Motor Component
model.motor = spa.Memory(dimensions=self.dimensions,
subdimensions=subdim,
neurons_per_dimension=n_neurons,
vocab=self.label_voc,
synapse=0.1,
direct=direct)
# Convenient probing of rule applications
if self.raster:
model.apps = spa.Buffer(dimensions=self.dimensions,
subdimensions=subdim,
neurons_per_dimension=n_neurons,
vocab=self.feat_voc,
direct=direct)
self.pApps = nengo.Probe(model.apps.state.output, synapse=0.03)
self.appSpikes = nengo.Probe(model.apps.state.ea_ensembles[3],
'spikes')
nengo.Connection(model.cleanup1.output, model.apps.state.input)
# Action definitions
actions = spa.Actions(
'dot(vision, POSNER) --> context_mem=VIS',
'dot(vision, BROOKS) --> context_mem=VIS',
'dot(vision, MURPHY) --> context_mem=INFER',
'dot(context_mem, VIS) --> gatesignal=2, context_mem=R5',
'dot(context_mem, INFER) --> context_mem=R1',
'dot(context_mem, R1) --> inference=sp_mem*~R1, context_mem=R2',
'dot(context_mem, R2) --> inference=sp_mem*~R2, context_mem=R3',
'dot(context_mem, R3) --> inference=sp_mem*~R3, context_mem=R4',
'dot(context_mem, R4) --> inference=sp_mem*~R4, context_mem=R5',
'dot(context_mem, R5) --> motor=decision')
# Basal ganglia and thalamus
model.bg = spa.BasalGanglia(actions)
model.thal = spa.Thalamus(model.bg)
# Subnetworks defined outside of SPA
model.product = Product(600, self.dimensions)
model.vis_gate = Product(300, self.dimensions)
model.mem_gate = Product(300, self.dimensions)
model.conv = CircularConvolution(250,
self.dimensions,
invert_a=True)
# Connections for gate with memory for perceptual evaluation tasks
nengo.Connection(model.vision.state.output, model.vis_gate.B)
nengo.Connection(model.gatesignal.state.output,
model.vis_gate.A,
transform=g_transform)
nengo.Connection(model.vis_gate.output, model.conv.A)
nengo.Connection(model.sp_mem.state.output, model.mem_gate.B)
nengo.Connection(model.gatesignal.state.output,
model.mem_gate.A,
transform=g_transform)
nengo.Connection(model.mem_gate.output, model.conv.B)
nengo.Connection(model.conv.output, model.decision.state.input)
# Connections for inferential evaluation tasks
nengo.Connection(model.inference.state.output,
model.cleanup1.input)
nengo.Connection(model.cleanup1.output, model.product.A)
nengo.Connection(model.vision.state.output, model.product.B)
nengo.Connection(model.cleanup2.output, model.decision.state.input)
nengo.Connection(model.product.output,
model.score.state.input,
transform=model.product.dot_product_transform())
nengo.Connection(model.score.state.output,
model.cleanup2.input,
transform=np.transpose(output))
# Input to visual buffer
def vision(t):
if t < 0.08: return self.stimuli.task
if 0.0801 <= t < 1: return self.probe
else: return '0'
model.input = spa.Input(vision=vision, sp_mem=self.stimuli.memory)
# Inhibit the gate with inference actions
actions = [2, 4, 5, 6, 7, 8]
target = model.gatesignal
for action in actions:
for e in target.all_ensembles:
nengo.Connection(model.thal.actions.ensembles[action],
e.neurons,
transform=[[-2]] * e.n_neurons)
# Set radius for ensemble computing scalar coherence score.
for ens in model.score.state.ensembles:
ens.radius = 2
# Define probes for semantic pointer plots
self.pVision = nengo.Probe(model.vision.state.output, synapse=0.03)
self.pMotor = nengo.Probe(model.motor.state.output, synapse=0.03)
self.pMemory = nengo.Probe(model.sp_mem.state.output, synapse=0.03)
self.pContext = nengo.Probe(model.context_mem.state.output,
synapse=0.03)
self.pScore = nengo.Probe(model.score.state.output, synapse=0.03)
self.pDecision = nengo.Probe(model.decision.state.output,
synapse=0.03)
self.pActions = nengo.Probe(model.thal.actions.output,
synapse=0.01)
self.pUtility = nengo.Probe(model.bg.input, synapse=0.01)
# Define probes for spike rasters
self.visSpikes = nengo.Probe(model.vision.state.ea_ensembles[3],
'spikes')
self.conSpikes = nengo.Probe(
model.context_mem.state.ea_ensembles[5], 'spikes')
self.memSpikes = nengo.Probe(model.sp_mem.state.ea_ensembles[7],
'spikes')
self.motSpikes = nengo.Probe(model.motor.state.ea_ensembles[1],
'spikes')
self.scoSpikes = nengo.Probe(model.score.state.ea_ensembles[0],
'spikes')
# Run the model
sim = nengo.Simulator(model)
sim.run(0.45)
# Save graph if plotting chosen
if self.raster:
graph = Plotter(self, sim)
if self.stimuli.task == 'MURPHY':
graph.plot_spikes_inf()
else:
graph.plot_spikes_vis()
# Assess correctness of output
self.measure_output(sim)
def create_model(
trial_info=('Target', 1, 'Short', 'CARGO', 'HOOD'), hand='LEFT',
seedin=1):
initialize_vocabs()
print trial_info
global global_item1
global global_item2
global_item1 = trial_info[3]
global_item2 = trial_info[4]
global model
model = spa.SPA(seed=seedin)
with model:
#display current stimulus pair
model.pair_input = nengo.Node(present_pair)
model.pair_display = nengo.Node(
display_func, size_in=model.pair_input.size_out) # to show input
nengo.Connection(model.pair_input, model.pair_display, synapse=None)
#visual
model.visual_net = nengo.Network()
with model.visual_net:
if not extended_visual:
model.stimulus = spa.State(D, vocab=vocab_concepts, feedback=1)
model.stim = spa.Input(stimulus=present_item_simple)
else:
#represent currently attended item
model.attended_item = nengo.Node(present_item)
model.vision_process = nengo.Ensemble(
n_hid,
n_vis,
eval_points=X_train,
neuron_type=nengo.LIFRate(),
intercepts=nengo.dists.Choice([-0.5
]), #can switch these off
max_rates=nengo.dists.Choice([100]), # why?
encoders=encoders)
# 1000 neurons, nrofpix = dimensions
# visual_representation = nengo.Node(size_in=Dmid) #output, in this case 466 outputs
model.visual_representation = nengo.Ensemble(
n_hid, dimensions=Dmid) # output, in this case 466 outputs
model.visconn = nengo.Connection(
model.vision_process,
model.visual_representation,
synapse=0.005,
eval_points=X_train,
function=train_targets,
solver=nengo.solvers.LstsqL2(reg=0.01))
nengo.Connection(model.attended_item,
model.vision_process,
synapse=None)
# display attended item
model.display_node = nengo.Node(
display_func,
size_in=model.attended_item.size_out) # to show input
nengo.Connection(model.attended_item,
model.display_node,
synapse=None)
#control
model.control_net = nengo.Network()
with model.control_net:
model.attend = spa.State(
D, vocab=vocab_concepts,
feedback=.5) #if attend item, goes to concepts
model.goal = spa.State(D, vocab_goal, feedback=1) #current goal
model.target_hand = spa.State(Dmid, vocab=vocab_motor, feedback=1)
# concepts
model.concepts = spa.AssociativeMemory(vocab_concepts,
wta_output=True,
wta_inhibit_scale=1)
if not extended_visual:
nengo.Connection(model.stimulus.output, model.concepts.input)
else:
nengo.Connection(model.visual_representation,
model.concepts.input,
transform=vision_mapping)
# pair representation
model.vis_pair = spa.State(D, vocab=vocab_concepts, feedback=1)
model.dm_items = spa.AssociativeMemory(
vocab_items
) #familiarity should be continuous over all items, so no wta
nengo.Connection(model.dm_items.output,
model.dm_items.input,
transform=.5,
synapse=.01)
model.familiarity = spa.State(
1, feedback_synapse=.01) #no fb syn specified
nengo.Connection(
model.dm_items.am.elem_output,
model.familiarity.input, #am.element_output == all outputs, we sum
transform=.8 * np.ones(
(1, model.dm_items.am.elem_output.size_out)))
#model.dm_pairs = spa.AssociativeMemory(vocab_concepts, input_keys=list_of_pairs,wta_output=True)
#nengo.Connection(model.dm_items.output,model.dm_pairs.input)
#motor
model.motor_net = nengo.Network()
with model.motor_net:
#input multiplier
model.motor_input = spa.State(Dmid, vocab=vocab_motor)
#higher motor area (SMA?)
model.motor = spa.State(Dmid, vocab=vocab_motor, feedback=1)
#connect input multiplier with higher motor area
nengo.Connection(model.motor_input.output,
model.motor.input,
synapse=.1,
transform=10)
#finger area
model.fingers = spa.AssociativeMemory(
vocab_fingers,
input_keys=['L1', 'L2', 'R1', 'R2'],
wta_output=True)
#conncetion between higher order area (hand, finger), to lower area
nengo.Connection(model.motor.output,
model.fingers.input,
transform=.4 * motor_mapping)
#finger position (spinal?)
model.finger_pos = nengo.networks.EnsembleArray(n_neurons=50,
n_ensembles=4)
nengo.Connection(model.finger_pos.output,
model.finger_pos.input,
synapse=0.1,
transform=0.3) #feedback
#connection between finger area and finger position
nengo.Connection(model.fingers.am.elem_output,
model.finger_pos.input,
transform=np.diag([0.55, .53, .57,
.55])) #fix these
model.bg = spa.BasalGanglia(
spa.Actions(
'dot(goal,DO_TASK)-.5 --> dm_items=vis_pair, goal=RECOG, attend=ITEM1',
'dot(goal,RECOG)+dot(attend,ITEM1)+familiarity-2 --> goal=RECOG2, dm_items=vis_pair, attend=ITEM2',
'dot(goal,RECOG)+dot(attend,ITEM1)+(1-familiarity)-2 --> goal=RECOG2, attend=ITEM2', #motor_input=1.5*target_hand+MIDDLE,
'dot(goal,RECOG2)+dot(attend,ITEM2)+familiarity-1.3 --> goal=RESPOND, dm_items=vis_pair,motor_input=1.2*target_hand+INDEX, attend=ITEM2',
'dot(goal,RECOG2)+dot(attend,ITEM2)+(1-familiarity)-1.3 --> goal=RESPOND, motor_input=1.5*target_hand+MIDDLE, attend=ITEM2',
'dot(goal,RESPOND)+dot(motor,MIDDLE+INDEX)-1.4 --> goal=END',
'dot(goal,END) --> goal=END',
#'.6 -->',
))
model.thalamus = spa.Thalamus(model.bg)
model.cortical = spa.Cortical( # cortical connection: shorthand for doing everything with states and connections
spa.Actions(
# 'motor_input = .04*target_hand',
#'dm_items = .8*concepts', #.5
#'dm_pairs = 2*stimulus'
'vis_pair = attend*concepts+concepts', ))
#probes
#model.pr_goal = nengo.Probe(model.goal.output,synapse=.01) #sample_every=.01 seconds, etc...
model.pr_motor_pos = nengo.Probe(
model.finger_pos.output,
synapse=.01) #raw vector (dimensions x time)
model.pr_motor = nengo.Probe(model.fingers.output, synapse=.01)
model.pr_motor1 = nengo.Probe(model.motor.output, synapse=.01)
#model.pr_target = nengo.Probe(model.target_hand.output, synapse=.01)
#input
model.input = spa.Input(
goal=lambda t: 'DO_TASK' if t < 0.05 else '0',
target_hand=hand,
#attend=lambda t: 'ITEM1' if t < 0.1 else 'ITEM2',
)
def main():
print "Loading Word2Vec model..."
word2vec_model = gensim.models.Word2Vec.load("word2vec_model_1_cleaned")
word2vec_model.init_sims(replace=True)
word2vec_vocab = word2vec_model.index2word
import readline
readline.parse_and_bind("tab: complete")
def complete(text, state):
results = [x for x in word2vec_vocab if x.startswith(text)] + [None]
return results[state]
readline.set_completer(complete)
print "This program uses an SPA network in Nengo to perform vector operations on a semantically structured word-vector space *learned* from a sentence corpus."
print "When trained on a large corpus of English sentences, for example, it should produce: Vector[king] - Vector[man] + Vector[woman] = Vector[king]"
print "For now, it just does subtraction..."
print "\nPress <tab> twice to see all your autocomplete options."
print "_______________________________________________________"
line1 = raw_input('\nFirst word:> ')
line2 = raw_input('\nSecond word:> ')
if line1 and line2 in word2vec_vocab:
val1 = word2vec_model[line1]
val2 = word2vec_model[line2]
diff = val1 - val2
dot_products = [
np.dot(word2vec_model[word2vec_model.index2word[i]], diff)
for i in range(len(word2vec_model.index2word))
]
closest_word = word2vec_model.index2word[dot_products.index(
max(dot_products))]
print "\nWhat the Nengo model SHOULD return is something like: %s" % closest_word
print "\nDefining SPA network..."
model = spa.SPA(label="Vector Storage")
with model:
# Dimensionality of each representation
num_dimensions = 100
sub_dimensions = 1
# Create the vocabulary
vocab = Vocabulary(num_dimensions, randomize=False)
stored_value_1 = val1
vocab.add("Stored_value_1", stored_value_1)
stored_value_2 = val2
vocab.add("Stored_value_2", stored_value_2)
# Create a semantic pointer corresponding to the "correct" answer for the operation
sum_vector = np.subtract(stored_value_1, stored_value_2)
sum_vector = sum_vector / np.linalg.norm(sum_vector)
vocab.add("Correct_target", sum_vector)
# Define the control signal inputs as random vectors
r1 = [1] * num_dimensions
r1 = r1 / np.linalg.norm(r1)
r2 = [(-1)**k for k in range(num_dimensions)]
r2 = r2 / np.linalg.norm(r2)
vocab.add("Hold_signal", r1)
vocab.add("Start_signal", r2)
# Control when the vector operation takes place
def control_input(t):
if t < 1:
return "Hold_signal"
else:
return "Start_signal"
# Control buffer
model.control = spa.Buffer(dimensions=num_dimensions,
subdimensions=sub_dimensions,
neurons_per_dimension=200,
direct=True,
vocab=vocab)
control_probe = nengo.Probe(model.control.state.output)
# Inputs to the word input buffers
def first_input(t):
return "Stored_value_1"
def second_input(t):
return "Stored_value_2"
# Buffers to store the inputs:
model.word_buffer1 = spa.Buffer(dimensions=num_dimensions,
subdimensions=sub_dimensions,
neurons_per_dimension=200,
direct=True,
vocab=vocab)
model.word_buffer2 = spa.Buffer(dimensions=num_dimensions,
subdimensions=sub_dimensions,
neurons_per_dimension=200,
direct=True,
vocab=vocab)
# Probe to visualize the values stored in the buffers
buffer_1_probe = nengo.Probe(model.word_buffer1.state.output)
buffer_2_probe = nengo.Probe(model.word_buffer2.state.output)
# Buffer to hold the result:
model.result = spa.Buffer(dimensions=num_dimensions,
subdimensions=sub_dimensions,
neurons_per_dimension=200,
direct=True,
vocab=vocab)
result_probe = nengo.Probe(model.result.state.output)
# Control system
actions = spa.Actions(
'dot(control, Start_signal) --> result = word_buffer1 - word_buffer2',
'dot(control, Hold_signal) --> result = Hold_signal')
model.bg = spa.BasalGanglia(actions)
model.thalamus = spa.Thalamus(model.bg, subdim_channel=sub_dimensions)
# Connect up the inputs
model.input = spa.Input(control=control_input,
word_buffer1=first_input,
word_buffer2=second_input)
# Start the simulator
sim = nengo.Simulator(model)
# Dynamic plotting
plt.ion() # Dynamic updating of plots
fig = plt.figure(figsize=(15, 8))
plt.show()
ax = fig.gca()
ax.set_title("Vector Storage")
while True:
sim.run(0.5) # Run for an additional 1 second
plt.clf() # Clear the figure
plt.plot(sim.trange(),
similarity(sim.data, result_probe, vocab),
label="Buffer 3 Value")
legend_symbols = vocab.keys
plt.legend(legend_symbols, loc=2)
plt.draw() # Re-draw
# Go back to our manually-stored vocabulary and see how well it did
diff = sim.data[result_probe][-1]
dot_products = [
np.dot(word2vec_model[word2vec_model.index2word[i]], diff)
for i in range(len(word2vec_model.index2word))
]
closest_word = word2vec_model.index2word[dot_products.index(
max(dot_products))]
print "Time: %f" % sim.trange()[-1]
print "\nWhat the Nengo model DID return is something like: %s" % closest_word
print "\n"
plt.show()
def simulate(lg=.2, le=.2, save_dir=None, sim_time=6):
nengo.rc.set('progress', 'progress_bar',
'nengo.utils.progress.TerminalProgressBar')
d = 64
n = 50 * d
# class for delayed connection
delay_t = 0.2 # 200 ms pro Silbe
dt = 0.001
delay = Delay(dimensions=d, timesteps=int(delay_t / dt))
# SPA-Model:
model = spa.SPA()
with model:
# --------------------- SPA representations ----------------
# High level SPs
model.visual = spa.Buffer(dimensions=d)
model.phonemic = spa.Buffer(dimensions=d)
model.premotor = spa.Buffer(dimensions=d)
model.audiexpect = spa.Buffer(dimensions=d)
model.motor = spa.Buffer(dimensions=d)
model.motor1 = nengo.Ensemble(n, dimensions=d)
model.motor2 = nengo.Ensemble(n, dimensions=d)
model.somato = spa.Buffer(dimensions=d)
model.delaynode = nengo.Node(delay.step, size_in=d, size_out=d)
# High level routing
actions = spa.Actions(
'dot(visual, BA) --> phonemic=BA',
'0.2 --> phonemic=NEUTRAL, premotor=NEUTRAL, audiexpect=NEUTRAL, motor=NEUTRAL',
'dot(phonemic, BA) --> premotor=BA, motor=BA_EXEC, audiexpect=BA',
'dot(phonemic, DA) --> premotor=DA, motor=DA_EXEC, audiexpect=DA',
'dot(phonemic, GA) --> premotor=GA, motor=GA_EXEC, audiexpect=GA',
'dot(phonemic, PA) --> premotor=PA, motor=PA_EXEC, audiexpect=PA',
'dot(phonemic, TA) --> premotor=TA, motor=TA_EXEC, audiexpect=TA',
'dot(phonemic, KA) --> premotor=KA, motor=KA_EXEC, audiexpect=KA',
'dot(somato, BA_EXEC) --> phonemic=DA',
'dot(somato, DA_EXEC) --> phonemic=GA',
'dot(somato, GA_EXEC) --> phonemic=PA',
'dot(somato, PA_EXEC) --> phonemic=TA',
'dot(somato, TA_EXEC) --> phonemic=KA',
'dot(somato, KA_EXEC) --> phonemic=BA',
)
nengo.networks.actionselection.Weights.lg = lg
nengo.networks.actionselection.Weights.le = le
nengo.networks.actionselection.Weights.wt = 1.0
nengo.networks.actionselection.Weights.wp = 0.9
nengo.networks.actionselection.Weights.wp_snr = 0.9 # not used yet
model.bg = spa.BasalGanglia(actions)
model.thalamus = spa.Thalamus(model.bg)
# model delay:
nengo.Connection(model.motor.state.output, model.motor1, synapse=0.01)
# no delay:
#nengo.Connection(model.motor1, model.motor2, synapse=0.01)
# delay:
nengo.Connection(model.motor1, model.delaynode)
nengo.Connection(model.delaynode, model.motor2)
nengo.Connection(model.motor2, model.somato.state.input, synapse=0.01)
# Provide input to high-level
def start(t):
if t < 0.25:
return 'ZERO'
elif t < 0.45:
return 'BA'
else:
return 'ZERO'
model.input = spa.Input(visual=start)
# set up probes
visual = nengo.Probe(model.visual.state.output, synapse=0.03)
phonemic = nengo.Probe(model.phonemic.state.output, synapse=0.03)
premotor = nengo.Probe(model.premotor.state.output, synapse=0.03)
audiexpect = nengo.Probe(model.audiexpect.state.output, synapse=0.03)
motor = nengo.Probe(model.motor.state.output, synapse=0.03)
somato = nengo.Probe(model.somato.state.output, synapse=0.03)
with nengo.Simulator(model) as sim:
sim.run(sim_time)
plt.figure(figsize=(18, 16))
plt.subplot(6, 1, 1)
plt.plot(sim.trange(), model.similarity(sim.data, visual))
plt.legend(model.get_input_vocab('visual').keys, fontsize='x-small')
plt.ylabel('Visual')
plt.subplot(6, 1, 2)
plt.plot(sim.trange(), model.similarity(sim.data, phonemic))
plt.legend(model.get_input_vocab('phonemic').keys, fontsize='x-small')
plt.ylabel('Phonemic')
plt.subplot(6, 1, 3)
plt.plot(sim.trange(), model.similarity(sim.data, audiexpect))
plt.legend(model.get_input_vocab('audiexpect').keys, fontsize='x-small')
plt.ylabel('Audi_expect')
plt.subplot(6, 1, 4)
plt.plot(sim.trange(), model.similarity(sim.data, premotor))
plt.legend(model.get_input_vocab('premotor').keys, fontsize='x-small')
plt.ylabel('Premotor')
plt.subplot(6, 1, 5)
plt.plot(sim.trange(), model.similarity(sim.data, motor))
plt.legend(model.get_input_vocab('motor').keys, fontsize='x-small')
plt.ylabel('Motor')
plt.subplot(6, 1, 6)
plt.plot(sim.trange(), model.similarity(sim.data, somato))
plt.legend(model.get_input_vocab('somato').keys, fontsize='x-small')
plt.ylabel('Somato')
if save_dir is not None:
plt.savefig(save_dir)
plt.close()
def Spaun():
model = spa.SPA(label='Spaun', seed=cfg.seed)
with model:
model.config[nengo.Ensemble].max_rates = cfg.max_rates
model.config[nengo.Ensemble].neuron_type = cfg.neuron_type
model.config[nengo.Connection].synapse = cfg.pstc
if 'S' in cfg.spaun_modules:
model.stim = Stimulus()
model.instr_stim = InstrStimulus()
model.monitor = Monitor()
if 'V' in cfg.spaun_modules:
model.vis = Vision()
if 'P' in cfg.spaun_modules:
model.ps = ProdSys()
if 'R' in cfg.spaun_modules:
model.reward = RewardEval()
if 'E' in cfg.spaun_modules:
model.enc = InfoEnc()
if 'W' in cfg.spaun_modules:
model.mem = Memory()
if 'T' in cfg.spaun_modules:
model.trfm = TrfmSys()
if 'D' in cfg.spaun_modules:
model.dec = InfoDec()
if 'M' in cfg.spaun_modules:
model.mtr = Motor()
if 'I' in cfg.spaun_modules:
model.instr = InstrProcess()
model.learn_conns = []
if hasattr(model, 'vis') and hasattr(model, 'ps'):
copy_draw_action = \
['0.5 * (dot(ps_task, X) + dot(vis, ZER)) --> ps_task = W, ps_state = TRANS0, ps_dec = FWD', # noqa
'dot(ps_task, W-DEC) - dot(vis, QM) --> ps_state = ps_state'] # noqa
# Copy drawing task format: A0[r]?X
recog_action = \
['0.5 * (dot(ps_task, X) + dot(vis, ONE)) --> ps_task = R, ps_state = TRANS0, ps_dec = FWD', # noqa
'dot(ps_task, R-DEC) - dot(vis, QM) --> ps_state = ps_state'] # noqa
# Digit recognition task format: A1[r]?X
learn_state_action = ['0.5 * (dot(ps_task, X) + dot(vis, TWO)) - dot(vis, QM) --> ps_task = L, ps_state = LEARN, ps_dec = FWD'] # noqa
# Learning task format: A2?X<REWARD>?X<REWARD>?X<REWARD>?X<REWARD>?X<REWARD> # noqa
if hasattr(model, 'reward'):
learn_action = ['0.5 * (dot(ps_task, 2*L) - 1) - dot(vis, QM) --> ps_action = %s, ps_state = LEARN, ps_dec = NONE' % # noqa
s for s in vocab.ps_action_learn_sp_strs]
else:
learn_action = []
mem_action = \
['0.5 * (dot(ps_task, X) + dot(vis, THR)) --> ps_task = M, ps_state = TRANS0, ps_dec = FWD', # noqa
'dot(ps_task, M-DEC) - dot(vis, F + R + QM) --> ps_state = ps_state', # noqa
'0.5 * (dot(ps_task, M) + dot(vis, F)) - dot(vis, QM) - dot(ps_task, DEC) --> ps_dec = FWD', # noqa
'0.5 * (dot(ps_task, M) + dot(vis, R)) - dot(vis, QM) - dot(ps_task, DEC) --> ps_dec = REV'] # noqa
# Working memory task format: A3[rr..rr]?XXX
# Reverse recall task format: A3[rr..rr]R?XXX
if hasattr(model, 'trfm'):
count_action = \
['0.5 * (dot(ps_task, X) + dot(vis, FOR)) --> ps_task = C, ps_state = TRANS0, ps_dec = FWD', # noqa
'0.5 * (dot(ps_task, C) + dot(ps_state, TRANS0)) - dot(vis, QM) - dot(ps_task, DEC) --> ps_state = CNT0', # noqa
'0.5 * (dot(ps_task, C) + dot(ps_state, CNT0)) - dot(vis, QM) - dot(ps_task, DEC) --> ps_state = CNT1', # noqa
'(0.25 * (dot(ps_task, DEC) + dot(ps_state, CNT1)) + 0.5 * dot(trfm_compare, NO_MATCH)) + (dot(ps_dec, CNT) - 1) - dot(vis, QM) --> ps_dec = CNT, ps_state = CNT1', # noqa
'(0.25 * (dot(ps_task, DEC) + dot(ps_state, CNT1)) + 0.5 * dot(trfm_compare, MATCH)) + (dot(ps_dec, CNT) - 1) - dot(vis, QM) --> ps_dec = FWD, ps_state = TRANS0'] # noqa
# Counting task format: A4[START_NUM][NUM_COUNT]?X..X
qa_action = \
['0.5 * (dot(ps_task, X) + dot(vis, FIV)) --> ps_task = A, ps_state = TRANS0, ps_dec = FWD', # noqa
'dot(ps_task, A-DEC) - dot(vis, K + P + QM) --> ps_state = ps_state', # noqa
'0.5 * (dot(ps_task, A) + dot(vis, K)) - dot(vis, QM) - dot(ps_task, DEC) --> ps_state = QAK', # noqa
'0.5 * (dot(ps_task, A) + dot(vis, P)) - dot(vis, QM) - dot(ps_task, DEC) --> ps_state = QAP'] # noqa
# Question answering task format: A5[rr..rr]P[r]?X (probing item in position) # noqa
# A5[rr..rr]K[r]?X (probing position of item (kind)) # noqa
rvc_action = \
['0.5 * (dot(ps_task, X) + dot(vis, SIX)) --> ps_task = V, ps_state = TRANS0, ps_dec = FWD', # noqa
'0.5 * (dot(ps_task, V) + dot(ps_state, TRANS0)) - dot(vis, QM) - dot(ps_task, DEC) --> ps_state = TRANS1', # noqa
'0.5 * (dot(ps_task, V) + dot(ps_state, TRANS1)) - dot(vis, QM) - dot(ps_task, DEC) --> ps_state = TRANS0'] # noqa
# Rapid variable creation task format: A6{[rr..rr][rr..rr]:NUM_EXAMPLES}?XX..XX # noqa
fi_action = \
['0.5 * (dot(ps_task, X) + dot(vis, SEV)) --> ps_task = F, ps_state = TRANS0, ps_dec = FWD', # noqa
'0.5 * (dot(ps_task, F) + dot(ps_state, TRANS0)) - dot(vis, QM) - dot(ps_task, DEC) --> ps_state = TRANS1', # noqa
'0.5 * (dot(ps_task, F) + dot(ps_state, TRANS1)) - dot(vis, QM) - dot(ps_task, DEC) --> ps_state = TRANS2', # noqa
'0.5 * (dot(ps_task, F) + dot(ps_state, TRANS2)) - dot(vis, QM) - dot(ps_task, DEC) --> ps_state = TRANS0'] # noqa
# Fluid intelligence task format: A7[CELL1_1][CELL1_2][CELL1_3][CELL2_1][CELL2_2][CELL2_3][CELL3_1][CELL3_2]?XX..XX # noqa
# Reaction task
react_action = \
['0.5 * (dot(ps_task, X) + dot(vis, EIG)) --> ps_task = REACT, ps_state = DIRECT, ps_dec = FWD', # noqa
'0.5 * (dot(ps_task, REACT) + dot(vis_mem, ONE)) --> trfm_input = POS1*THR', # noqa
'0.5 * (dot(ps_task, REACT) + dot(vis_mem, TWO)) --> trfm_input = POS1*FOR'] # noqa
# Stimulus response (hardcoded reaction) task format: A8?1X<expected 3>?2X<expected 4> # noqa
# Compare task -- See two items, check if their class matches each other # noqa
match_action = \
['0.5 * (dot(ps_task, X) + dot(vis, C)) --> ps_task = CMP, ps_state = TRANS1, ps_dec = FWD', # noqa
'0.5 * (dot(ps_task, CMP) + dot(ps_state, TRANS1)) - dot(vis, QM) - dot(ps_task, DEC) --> ps_state = TRANS2', # noqa
'0.5 * (dot(ps_task, CMP) + dot(ps_state, TRANS2)) - dot(vis, QM) - dot(ps_task, DEC) --> ps_state = TRANSC'] # noqa
# List / item matching task format: AC[r][r]?X<expected 1 if match, 0 if not> # noqa
# AC[rr..rr][rr..rr]?X<expected 1 if any item in first list appears in second list, 0 if not> # noqa
else:
count_action = []
qa_action = []
rvc_action = []
fi_action = []
react_action = []
match_action = []
if hasattr(model, 'trfm') and hasattr(model, 'instr'):
instr_action = \
['0.5 * (dot(ps_task, X) + dot(vis, NIN)) --> ps_task = INSTR, ps_state = DIRECT, ps_dec = FWD', # noqa
# 'dot(ps_task, INSTR) - dot(vis, QM + A) --> trfm_input = instr_data', # noqa - Note: this is the bg 'instr' rule in it's simplified form
'dot(ps_task, INSTR) - dot(vis, QM + A + M + P + CLOSE) - dot(ps_state, INSTRP) --> instr_en = ENABLE, ps_task = instr_task, ps_state = instr_state, ps_dec = instr_dec, trfm_input = instr_data', # noqa
'1.5 * dot(vis, M + V) --> ps_task = INSTR, ps_state = TRANS0, ps_dec = FWD', # noqa - Note: V no longer keeps state, dec information from before. Need to set in instruction
'0.5 * (dot(ps_task, INSTR) + dot(vis, P)) --> ps_task = INSTR, ps_state = INSTRP', # noqa
'0.5 * (dot(ps_task, INSTR) + dot(ps_state, INSTRP)) --> ps_task = INSTR, ps_state = TRANS0', # noqa
# 'dot(instr_util, INSTR) - dot(ps_task, INSTR) --> instr_en = ENABLE, ps_task = instr_task, ps_state = instr_state, ps_dec = instr_dec, trfm_input = instr_data', # noqa - Note: this is the untested 'use output of IPS to set PS states' implementation
] # noqa
# Instructed tasks task formats:
# Instructed stimulus response task format: A9?rX<expected answer from instruction>?rX<expected answer from instruction> # noqa
# Instructed custom task format: M<0-9>[INSTRUCTED TASK FORMAT]?XX..XX # noqa
# Instructed positional task formats: MP<0-9>[INSTRUCTED TASK FORMAT]?XX..XX V[INSTRUCTED TASK FORMAT]?XX..XX # noqa
# - P<0-9> selects appropriate instruction from list of instructions # noqa
# - V increments instruction position by 1
else:
instr_action = []
decode_action = \
['dot(vis, QM) - 0.6 * dot(ps_task, W+C+V+F+L+REACT) --> ps_task = ps_task + DEC, ps_state = ps_state + 0.5 * TRANS0, ps_dec = ps_dec + 0.5 * FWD', # noqa
'0.5 * (dot(vis, QM) + dot(ps_task, W-DEC)) --> ps_task = W + DEC, ps_state = ps_state, ps_dec = DECW', # noqa
'0.5 * (dot(vis, QM) + dot(ps_task, C-DEC)) --> ps_task = C + DEC, ps_state = ps_state, ps_dec = CNT', # noqa
'0.5 * (dot(vis, QM) + dot(ps_task, V+F-DEC)) --> ps_task = ps_task + DEC, ps_state = ps_state, ps_dec = DECI', # noqa
'0.7 * dot(vis, QM) + 0.3 * dot(ps_task, L) --> ps_task = L + DEC, ps_state = LEARN, ps_dec = FWD', # noqa
'0.5 * (dot(vis, QM) + dot(ps_task, REACT)) --> ps_task = REACT + DEC, ps_state = DIRECT, ps_dec = FWD', # noqa
'0.75 * dot(ps_task, DEC-REACT-INSTR) - dot(ps_state, LEARN) - dot(ps_dec, CNT) - dot(vis, QM + A + M) --> ps_task = ps_task, ps_state = ps_state, ps_dec = ps_dec'] # noqa
default_action = \
[]
# List learning task spa actions first, so we know the precise
# indicies of the learning task actions (i.e. the first N)
all_actions = (learn_action + learn_state_action +
copy_draw_action + recog_action + mem_action +
count_action + qa_action + rvc_action + fi_action +
decode_action + default_action + react_action +
instr_action + match_action)
actions = spa.Actions(*all_actions)
model.bg = spa.BasalGanglia(actions=actions, input_synapse=0.008,
label='Basal Ganglia')
model.thal = spa.Thalamus(model.bg, subdim_channel=1,
mutual_inhibit=1, route_inhibit=5.0,
label='Thalamus')
# ----- Set up connections (and save record of modules) -----
if hasattr(model, 'vis'):
model.vis.setup_connections(model)
if hasattr(model, 'ps'):
model.ps.setup_connections(model)
# Modify any 'channel' ensemble arrays to have
# get_optimal_sp_radius radius sizes
for net in model.ps.all_networks:
if net.label is not None and net.label[:7] == 'channel':
for ens in net.all_ensembles:
ens.radius = cfg.get_optimal_sp_radius()
if hasattr(model, 'bg'):
if hasattr(model, 'reward'):
# Clear learning transforms
del cfg.learn_init_transforms[:]
with model.bg:
# Generate random biases for each learn action, so that
# there is some randomness to the initial action choice
bias_node = nengo.Node(1)
bias_ens = nengo.Ensemble(cfg.n_neurons_ens, 1,
label='BG Bias Ensemble')
nengo.Connection(bias_node, bias_ens)
for i in range(len(learn_action)):
init_trfm = (np.random.random() *
cfg.learn_init_trfm_max)
trfm_val = cfg.learn_init_trfm_bias + init_trfm
model.learn_conns.append(
nengo.Connection(bias_ens, model.bg.input[i],
transform=trfm_val))
cfg.learn_init_transforms.append(trfm_val)
logger.write("# learn_init_trfms: %s\n" %
(str(cfg.learn_init_transforms)))
if hasattr(model, 'thal'):
pass
if hasattr(model, 'reward'):
model.reward.setup_connections(model, model.learn_conns)
if hasattr(model, 'enc'):
model.enc.setup_connections(model)
if hasattr(model, 'mem'):
model.mem.setup_connections(model)
if hasattr(model, 'trfm'):
model.trfm.setup_connections(model)
# Modify any 'channel' ensemble arrays to have
# get_optimal_sp_radius radius sizes
for net in model.trfm.all_networks:
if net.label is not None and net.label[:7] == 'channel':
for ens in net.all_ensembles:
ens.radius = cfg.get_optimal_sp_radius()
if hasattr(model, 'dec'):
model.dec.setup_connections(model)
if hasattr(model, 'mtr'):
model.mtr.setup_connections(model)
if hasattr(model, 'instr'):
model.instr.setup_connections(model)
if hasattr(model, 'monitor'):
model.monitor.setup_connections(model)
return model
import nengo
import nengo.spa as spa
D = 32 # the dimensionality of the vectors
model = spa.SPA()
with model:
model.vision = spa.Buffer(D)
model.speech = spa.Buffer(D)
actions = spa.Actions(
'dot(vision, DOG) --> speech=BARK',
'dot(vision, CAT) --> speech=MEOW',
'dot(vision, RAT) --> speech=SQUEAK',
'dot(vision, COW) --> speech=MOO',
)
model.bg = spa.BasalGanglia(actions)
model.thalamus = spa.Thalamus(model.bg)
feedback_synapse=0.01)
model.vision = spa.State(dimensions=dimensions)
# Specify the action mapping
actions = spa.Actions(
'dot(vision, START) --> state = vision',
'dot(state, A) --> state = B',
'dot(state, B) --> state = C',
'dot(state, C) --> state = D',
'dot(state, D) --> state = E',
'dot(state, E) --> state = A'
)
#Creating the BG and Thalamus components that confirm to the specified rules
model.BG = spa.BasalGanglia(actions=actions)
model.thal = spa.Thalamus(model.BG)
#Change the seed of this RNG to change the vocabulary
rng = np.random.RandomState(0)
vocab = Vocabulary(dimensions=dimensions)
#Create the transformation matrix (pd) and the cleanup ensemble (cleanupA)
pd = [vocab['A'].v.tolist()]
model.cleanup = spa.State(neurons_per_dimension=100, dimensions=1)
#Function that provides the model with an initial input semantic pointer.
def start(t):
if t < 0.4:
return '0.8*START+D'
else:
return '0'