model.train()
PeDurx = 50
batch_size = 50
pe = PoissonEncoder(duration=PeDurx)
optimizer = Adam(model.parameters(), lr=0.003)
train_loader = DataLoader(dataset=MnistDataset(training=True, flatten=False),
collate_fn=collate_fn,
shuffle=True,
batch_size=batch_size)
test_loader = DataLoader(dataset=MnistDataset(training=False, flatten=False),
collate_fn=collate_fn,
shuffle=False,
batch_size=batch_size)
for epoch in range(15):
for i, (data, target) in enumerate(train_loader):
target = Variable(target)
for t, q in enumerate(pe.Encoding(data)):
data = Variable(q, requires_grad=True)
output, time = model(data, t)
loss = F.Spikeloss(output, target, time_step=time)
loss.backward() # calc gradients
optimizer.step() # update gradients
print("Epoch:" + str(epoch) + " Time: " + str(t) + " loss: " +
str(loss.data))
model = SNN()
model.train()
PeDurx = 50
batch_size = 64
Pencoder = PoissonEncoder(duration=PeDurx)
optimizer = SGD(model.parameters(), lr=0.01)
train_loader = DataLoader(dataset=MnistDataset(training=True, flatten=True),
collate_fn=collate_fn,
shuffle=True,
batch_size=batch_size)
test_loader = DataLoader(dataset=MnistDataset(training=False, flatten=True),
collate_fn=collate_fn,
shuffle=False,
batch_size=batch_size)
for i in range(15):
for i, (image, label) in enumerate(train_loader):
label = Variable(label)
for t, j in enumerate(Pencoder.Encoding(image)):
image = Variable(j, requires_grad=True)
t = t + 1
output = model(image, t)
optimizer.zero_grad()
loss = F.Spikeloss(output, label, PeDurx)
print(loss.data)
loss.backward()
optimizer.step()
PeDurx = 45
batch_size = 64
optimizer = SGD(model.parameters(), lr=0.003)
encoder = PoissonEncoder(duration=PeDurx)
train_loader = DataLoader(dataset=MnistDataset(training=True, flatten=False),
collate_fn=collate_fn, shuffle=True,
batch_size=batch_size)
test_loader = DataLoader(dataset=MnistDataset(training=False, flatten=False),
collate_fn=collate_fn, shuffle=False,
batch_size=batch_size)
for epoch in range(15):
for i, (data, target) in enumerate(train_loader):
target = Variable(target)
for j, q in enumerate(encoder.Encoding(data)):
pdata = Variable(q, requires_grad=True)
output, v_current = model(pdata, j+1)
for k, v in enumerate(v_current):
os.makedirs(f"image/{str(k + 1)}", exist_ok=True)
if k < 4:
seaborn.heatmap(v[0][0], cbar=False).figure.savefig("image/" + str(k + 1) + "/" + str(j) + ".png")
else:
seaborn.heatmap(v, cbar=False).figure.savefig("image/" + str(k + 1) + "/" + str(j) + ".png")
optimizer.zero_grad()
loss = F.Spikeloss(output, target, time_step=PeDurx)
loss.backward() # calc gradients
optimizer.step() # update gradients
print("Epoch:" + str(epoch) + " Time: " + str(i) + " loss: " + str(loss.data))