def test_interface_variance():
interface_test(new_ext.Variance())
model = Sequential(Flatten(), Linear(784, 10),)
lossfunc = CrossEntropyLoss()
model = extend(model)
lossfunc = extend(lossfunc)
# %%
# We can now evaluate the loss and do a backward pass with Backpack
# -----------------------------------------------------------------
loss = lossfunc(model(X), y)
with backpack(
extensions.BatchGrad(),
extensions.Variance(),
extensions.SumGradSquared(),
extensions.BatchL2Grad(),
extensions.DiagGGNMC(mc_samples=1),
extensions.DiagGGNExact(),
extensions.DiagHessian(),
extensions.KFAC(mc_samples=1),
extensions.KFLR(),
extensions.KFRA(),
):
loss.backward()
# %%
# And here are the results
# -----------------------------------------------------------------
def variance(self):
with backpack(new_ext.Variance()):
_, _, loss = self.problem.forward_pass()
loss.backward()
variances = [p.variance for p in self.problem.model.parameters()]
return variances
def variance(self):
with backpack(new_ext.Variance()):
self.loss().backward()
variances = [p.variance for p in self.model.parameters()]
return variances
"""
from torch.nn import CrossEntropyLoss, Flatten, Linear, Sequential
from backpack import backpack, extend, extensions
from backpack.utils.examples import load_mnist_data
B = 4
X, y = load_mnist_data(B)
print("# Gradient with PyTorch, gradient variance with BackPACK | B =", B)
model = Sequential(
Flatten(),
Linear(784, 10),
)
lossfunc = CrossEntropyLoss()
model = extend(model)
lossfunc = extend(lossfunc)
loss = lossfunc(model(X), y)
with backpack(extensions.Variance()):
loss.backward()
for name, param in model.named_parameters():
print(name)
print(".grad.shape: ", param.grad.shape)
print(".variance.shape: ", param.variance.shape)
def variance(self) -> List[Tensor]: # noqa:D102
with backpack(new_ext.Variance()):
_, _, loss = self.problem.forward_pass()
loss.backward()
return self.problem.collect_data("variance")