def BindsNET_cpu(n_neurons, time):
t0 = t()
torch.set_default_tensor_type("torch.FloatTensor")
t1 = t()
network = Network()
network.add_layer(Input(n=n_neurons), name="X")
network.add_layer(LIFNodes(n=n_neurons), name="Y")
network.add_connection(
Connection(source=network.layers["X"], target=network.layers["Y"]),
source="X",
target="Y",
)
data = {"X": poisson(datum=torch.rand(n_neurons), time=time)}
network.run(inputs=data, time=time)
return t() - t0, t() - t1
# Spike recordings for all layers.
spikes = {}
for layer in layers:
spikes[layer] = Monitor(layers[layer], ["s"], time=plot_interval)
# Voltage recordings for excitatory and readout layers.
voltages = {}
for layer in set(layers.keys()) - {"X"}:
voltages[layer] = Monitor(layers[layer], ["v"], time=plot_interval)
# Add all layers and connections to the network.
for layer in layers:
network.add_layer(layers[layer], name=layer)
network.add_connection(input_exc_conn, source="X", target="E")
network.add_connection(exc_readout_conn, source="E", target="R")
# Add all monitors to the network.
for layer in layers:
network.add_monitor(spikes[layer], name="%s_spikes" % layer)
if layer in voltages:
network.add_monitor(voltages[layer], name="%s_voltages" % layer)
# Load the Breakout environment.
environment = GymEnvironment("BreakoutDeterministic-v4")
environment.reset()
pipeline = EnvironmentPipeline(
network,
conv_conn = Conv2dConnection(
input_layer,
conv_layer,
kernel_size=kernel_size,
stride=stride,
update_rule=PostPre,
norm=0.4 * kernel_size ** 2,
nu=[1e-4, 1e-2],
wmax=1.0,
)
network.add_layer(input_layer, name="X")
network.add_layer(conv_layer, name="Y")
network.add_connection(conv_conn, source="X", target="Y")
# Train the network.
print("Begin training.\n")
if args.tensorboard:
analyzer = TensorboardAnalyzer("logs/conv")
else:
analyzer = MatplotlibAnalyzer()
for step, batch in enumerate(tqdm(train_dataloader)):
# batch contains image, label, encoded_image since an image_encoder
# was provided
# batch["encoded_image"] is in BxTxCxHxW format
inputs = {"X": batch["encoded_image"]}
# Layers of neurons.
inpt = Input(n=80 * 80, shape=[80, 80], traces=True)
middle = LIFNodes(n=100, traces=True)
out = LIFNodes(n=4, refrac=0, traces=True)
# Connections between layers.
inpt_middle = Connection(source=inpt, target=middle, wmin=0, wmax=1e-1)
middle_out = Connection(source=middle, target=out, wmin=0, wmax=1)
# Add all layers and connections to the network.
network.add_layer(inpt, name="Input Layer")
network.add_layer(middle, name="Hidden Layer")
network.add_layer(out, name="Output Layer")
network.add_connection(inpt_middle,
source="Input Layer",
target="Hidden Layer")
network.add_connection(middle_out,
source="Hidden Layer",
target="Output Layer")
# Load the Breakout environment.
environment = GymEnvironment("BreakoutDeterministic-v4")
environment.reset()
# Build pipeline from specified components.
pipeline = EnvironmentPipeline(
network,
environment,
encoding=bernoulli,
action_function=select_softmax,
def ann_to_snn(
ann: Union[nn.Module, str],
input_shape: Sequence[int],
data: Optional[torch.Tensor] = None,
percentile: float = 99.9,
node_type: Optional[nodes.Nodes] = SubtractiveResetIFNodes,
**kwargs,
) -> Network:
# language=rst
"""
Converts an artificial neural network (ANN) written as a ``torch.nn.Module`` into a
near-equivalent spiking neural network.
:param ann: Artificial neural network implemented in PyTorch. Accepts either
``torch.nn.Module`` or path to network saved using ``torch.save()``.
:param input_shape: Shape of input data.
:param data: Data to use to perform data-based weight normalization of shape
``[n_examples, ...]``.
:param percentile: Percentile (in ``[0, 100]``) of activations to scale by in
data-based normalization scheme.
:param node_type: Class of ``Nodes`` to use in replacing ``torch.nn.Linear`` layers
in original ANN.
:return: Spiking neural network implemented in PyTorch.
"""
if isinstance(ann, str):
ann = torch.load(ann)
else:
ann = deepcopy(ann)
assert isinstance(ann, nn.Module)
if data is None:
import warnings
warnings.warn("Data is None. Weights will not be scaled.", RuntimeWarning)
else:
ann = data_based_normalization(
ann=ann, data=data.detach(), percentile=percentile
)
snn = Network()
input_layer = nodes.Input(shape=input_shape)
snn.add_layer(input_layer, name="Input")
children = []
for c in ann.children():
if isinstance(c, nn.Sequential):
for c2 in list(c.children()):
children.append(c2)
else:
children.append(c)
i = 0
prev = input_layer
while i < len(children) - 1:
current, nxt = children[i : i + 2]
layer, connection = _ann_to_snn_helper(prev, current, node_type, **kwargs)
i += 1
if layer is None or connection is None:
continue
snn.add_layer(layer, name=str(i))
snn.add_connection(connection, source=str(i - 1), target=str(i))
prev = layer
current = children[-1]
layer, connection = _ann_to_snn_helper(
prev, current, node_type, last=True, **kwargs
)
i += 1
if layer is not None or connection is not None:
snn.add_layer(layer, name=str(i))
snn.add_connection(connection, source=str(i - 1), target=str(i))
return snn
torch.manual_seed(seed)
network = Network(dt=dt)
inpt = Input(784, shape=(1, 28, 28))
network.add_layer(inpt, name="I")
output = LIFNodes(n_neurons,
thresh=-52 + np.random.randn(n_neurons).astype(float))
network.add_layer(output, name="O")
C1 = Connection(source=inpt,
target=output,
w=0.5 * torch.randn(inpt.n, output.n))
C2 = Connection(source=output,
target=output,
w=0.5 * torch.randn(output.n, output.n))
network.add_connection(C1, source="I", target="O")
network.add_connection(C2, source="O", target="O")
spikes = {}
for l in network.layers:
spikes[l] = Monitor(network.layers[l], ["s"], time=time)
network.add_monitor(spikes[l], name="%s_spikes" % l)
voltages = {"O": Monitor(network.layers["O"], ["v"], time=time)}
network.add_monitor(voltages["O"], name="O_voltages")
# Directs network to GPU
if gpu:
network.to("cuda")
# Get MNIST training images and labels.