def test_compute(self):
auc_metric = ROCAUC()
act = Activations(softmax=True)
to_onehot = AsDiscrete(to_onehot=2)
device = f"cuda:{dist.get_rank()}" if torch.cuda.is_available(
) else "cpu"
if dist.get_rank() == 0:
y_pred = [
torch.tensor([0.1, 0.9], device=device),
torch.tensor([0.3, 1.4], device=device)
]
y = [
torch.tensor([0], device=device),
torch.tensor([1], device=device)
]
if dist.get_rank() == 1:
y_pred = [
torch.tensor([0.2, 0.1], device=device),
torch.tensor([0.1, 0.5], device=device),
torch.tensor([0.3, 0.4], device=device),
]
y = [
torch.tensor([0], device=device),
torch.tensor([1], device=device),
torch.tensor([1], device=device)
]
y_pred = [act(p) for p in y_pred]
y = [to_onehot(y_) for y_ in y]
auc_metric.update([y_pred, y])
result = auc_metric.compute()
np.testing.assert_allclose(0.66667, result, rtol=1e-4)
def test_compute(self):
auc_metric = ROCAUC(to_onehot_y=True, softmax=True)
y_pred = torch.Tensor([[0.1, 0.9], [0.3, 1.4]])
y = torch.Tensor([[0], [1]])
auc_metric.update([y_pred, y])
y_pred = torch.Tensor([[0.2, 0.1], [0.1, 0.5]])
y = torch.Tensor([[0], [1]])
auc_metric.update([y_pred, y])
auc = auc_metric.compute()
np.testing.assert_allclose(0.75, auc)
def test_compute(self):
auc_metric = ROCAUC(to_onehot_y=True, softmax=True)
device = f"cuda:{dist.get_rank()}" if torch.cuda.is_available(
) else "cpu"
if dist.get_rank() == 0:
y_pred = torch.tensor([[0.1, 0.9], [0.3, 1.4]], device=device)
y = torch.tensor([[0], [1]], device=device)
auc_metric.update([y_pred, y])
if dist.get_rank() == 1:
y_pred = torch.tensor([[0.2, 0.1], [0.1, 0.5]], device=device)
y = torch.tensor([[0], [1]], device=device)
auc_metric.update([y_pred, y])
result = auc_metric.compute()
np.testing.assert_allclose(0.75, result)
def test_compute(self):
auc_metric = ROCAUC()
act = Activations(softmax=True)
to_onehot = AsDiscrete(to_onehot=True, n_classes=2)
y_pred = torch.Tensor([[0.1, 0.9], [0.3, 1.4]])
y = torch.Tensor([[0], [1]])
y_pred = act(y_pred)
y = to_onehot(y)
auc_metric.update([y_pred, y])
y_pred = torch.Tensor([[0.2, 0.1], [0.1, 0.5]])
y = torch.Tensor([[0], [1]])
y_pred = act(y_pred)
y = to_onehot(y)
auc_metric.update([y_pred, y])
auc = auc_metric.compute()
np.testing.assert_allclose(0.75, auc)
def test_compute(self):
auc_metric = ROCAUC()
act = Activations(softmax=True)
to_onehot = AsDiscrete(to_onehot=True, num_classes=2)
y_pred = [torch.Tensor([0.1, 0.9]), torch.Tensor([0.3, 1.4])]
y = [torch.Tensor([0]), torch.Tensor([1])]
y_pred = [act(p) for p in y_pred]
y = [to_onehot(y_) for y_ in y]
auc_metric.update([y_pred, y])
y_pred = [torch.Tensor([0.2, 0.1]), torch.Tensor([0.1, 0.5])]
y = [torch.Tensor([0]), torch.Tensor([1])]
y_pred = [act(p) for p in y_pred]
y = [to_onehot(y_) for y_ in y]
auc_metric.update([y_pred, y])
auc = auc_metric.compute()
np.testing.assert_allclose(0.75, auc)
def main():
dist.init_process_group(backend="nccl", init_method="env://")
auc_metric = ROCAUC(to_onehot_y=True, softmax=True)
if dist.get_rank() == 0:
y_pred = torch.tensor([[0.1, 0.9], [0.3, 1.4]],
device=torch.device("cuda:0"))
y = torch.tensor([[0], [1]], device=torch.device("cuda:0"))
auc_metric.update([y_pred, y])
if dist.get_rank() == 1:
y_pred = torch.tensor([[0.2, 0.1], [0.1, 0.5]],
device=torch.device("cuda:1"))
y = torch.tensor([[0], [1]], device=torch.device("cuda:1"))
auc_metric.update([y_pred, y])
result = auc_metric.compute()
np.testing.assert_allclose(0.75, result)
dist.destroy_process_group()
def test_compute(self):
auc_metric = ROCAUC()
act = Activations(softmax=True)
to_onehot = AsDiscrete(to_onehot=True, n_classes=2)
device = f"cuda:{dist.get_rank()}" if torch.cuda.is_available(
) else "cpu"
if dist.get_rank() == 0:
y_pred = torch.tensor([[0.1, 0.9], [0.3, 1.4]], device=device)
y = torch.tensor([[0], [1]], device=device)
if dist.get_rank() == 1:
y_pred = torch.tensor([[0.2, 0.1], [0.1, 0.5], [0.3, 0.4]],
device=device)
y = torch.tensor([[0], [1], [1]], device=device)
y_pred = act(y_pred)
y = to_onehot(y)
auc_metric.update([y_pred, y])
result = auc_metric.compute()
np.testing.assert_allclose(0.66667, result, rtol=1e-4)
