def test_no_update():
ck = CohenKappa()
with pytest.raises(
NotComputableError, match=r"EpochMetric must have at least one example before it can be computed"
):
ck.compute()
def _test(y_pred, y, n_iters, metric_device):
metric_device = torch.device(metric_device)
ck = CohenKappa(device=metric_device)
torch.manual_seed(10 + rank)
ck.reset()
ck.update((y_pred, y))
if n_iters > 1:
batch_size = y.shape[0] // n_iters + 1
for i in range(n_iters):
idx = i * batch_size
ck.update(
(y_pred[idx:idx + batch_size], y[idx:idx + batch_size]))
# gather y_pred, y
y_pred = idist.all_gather(y_pred)
y = idist.all_gather(y)
np_y = y.cpu().numpy()
np_y_pred = y_pred.cpu().numpy()
res = ck.compute()
assert isinstance(res, float)
assert cohen_kappa_score(np_y, np_y_pred) == pytest.approx(res)
def _test(metric_device):
metric_device = torch.device(metric_device)
ck_metric = CohenKappa(device=metric_device)
torch.manual_seed(10 + rank)
y_pred = torch.randint(0, 2, size=(100, 1), device=device)
y = torch.randint(0, 2, size=(100, 1), device=device)
ck_metric.update((y_pred, y))
# gather y_pred, y
y_pred = idist.all_gather(y_pred)
y = idist.all_gather(y)
np_y_pred = y_pred.cpu().numpy()
np_y = y.cpu().numpy()
np_ck = cohen_kappa_score(np_y, np_y_pred)
res = ck_metric.compute()
assert res == pytest.approx(np_ck)
def test_multilabel_inputs():
ck = CohenKappa()
with pytest.raises(ValueError, match=r"multilabel-indicator is not supported"):
ck.reset()
ck.update((torch.randint(0, 2, size=(10, 4)).long(), torch.randint(0, 2, size=(10, 4)).long()))
ck.compute()
with pytest.raises(ValueError, match=r"multilabel-indicator is not supported"):
ck.reset()
ck.update((torch.randint(0, 2, size=(10, 6)).long(), torch.randint(0, 2, size=(10, 6)).long()))
ck.compute()
with pytest.raises(ValueError, match=r"multilabel-indicator is not supported"):
ck.reset()
ck.update((torch.randint(0, 2, size=(10, 8)).long(), torch.randint(0, 2, size=(10, 8)).long()))
ck.compute()
