def loss_fun(x: ep.Tensor, k: int) -> ep.Tensor:
logits = self.model.forward(x)
ik = classes[:, k]
l0 = -ep.crossentropy(logits, i0)
lk = -ep.crossentropy(logits, ik)
loss = lk - l0
return loss.sum(), (loss, logits)
def loss_fun(
x: ep.Tensor, k: int
) -> Tuple[ep.Tensor, Tuple[ep.Tensor, ep.Tensor]]:
logits = model(x)
ik = classes[:, k]
l0 = -ep.crossentropy(logits, i0)
lk = -ep.crossentropy(logits, ik)
loss = lk - l0
return loss.sum(), (loss, logits)
def loss_fn(inputs: ep.Tensor,
labels: ep.Tensor) -> Tuple[ep.Tensor, ep.Tensor]:
logits = model(inputs)
sign = -1.0 if targeted else 1.0
loss = sign * ep.crossentropy(logits, labels).sum()
return loss, logits
def test_crossentropy_raises(dummy: Tensor) -> None:
t = ep.arange(dummy, 50).reshape((10, 5)).float32()
t = t / t.max()
ep.crossentropy(t, t.argmax(axis=-1))
t = ep.arange(dummy, 150).reshape((10, 5, 3)).float32()
t = t / t.max()
with pytest.raises(ValueError):
ep.crossentropy(t, t.argmax(axis=-1))
t = ep.arange(dummy, 50).reshape((10, 5)).float32()
t = t / t.max()
with pytest.raises(ValueError):
ep.crossentropy(t, t.argmax(axis=-1)[:8])
def loss_fn(inputs: ep.Tensor) -> ep.Tensor:
logits = model(inputs)
return ep.crossentropy(logits, labels).sum()
def test_crossentropy(dummy: Tensor) -> Tensor:
t = ep.arange(dummy, 50).reshape((10, 5)).float32()
t = t / t.max()
return ep.crossentropy(t, t.argmax(axis=-1))
def loss_fn(inputs):
logits = model.forward(inputs)
return ep.crossentropy(logits, labels).sum()
def loss_function(input_data):
logits = model(input_data)
return ep.crossentropy(logits, labels).sum()
def loss_fn(inputs: ep.Tensor, labels: ep.Tensor) -> ep.Tensor:
logits = ep.astensor(self.model.forward(inputs.tensor))
return ep.crossentropy(logits, labels).sum()
