def _test(n_epochs, metric_device):
metric_device = torch.device(metric_device)
n_iters = 80
s = 16
n_classes = 2
offset = n_iters * s
y_true = torch.randint(0, n_classes, size=(offset * idist.get_world_size(),)).to(device)
y_preds = torch.randint(0, n_classes, size=(offset * idist.get_world_size(),)).to(device)
def update(engine, i):
return (
y_preds[i * s + rank * offset : (i + 1) * s + rank * offset],
y_true[i * s + rank * offset : (i + 1) * s + rank * offset],
)
engine = Engine(update)
ck = CohenKappa(device=metric_device)
ck.attach(engine, "ck")
data = list(range(n_iters))
engine.run(data=data, max_epochs=n_epochs)
assert "ck" in engine.state.metrics
res = engine.state.metrics["ck"]
if isinstance(res, torch.Tensor):
res = res.cpu().numpy()
true_res = cohen_kappa_score(y_true.cpu().numpy(), y_preds.cpu().numpy())
assert pytest.approx(res) == true_res
def test_cohen_kappa_all_weights_with_output_transform(weights):
np.random.seed(1)
size = 100
np_y_pred = np.random.randint(0, 2, size=(size, 1), dtype=np.long)
np_y = np.zeros((size, ), dtype=np.long)
np_y[size // 2:] = 1
np.random.shuffle(np_y)
ck_value_sk = cohen_kappa_score(np_y, np_y_pred)
batch_size = 10
def update_fn(engine, batch):
idx = (engine.state.iteration - 1) * batch_size
y_true_batch = np_y[idx:idx + batch_size]
y_pred_batch = np_y_pred[idx:idx + batch_size]
return idx, torch.from_numpy(y_pred_batch), torch.from_numpy(
y_true_batch)
engine = Engine(update_fn)
ck_metric = CohenKappa(output_transform=lambda x: (x[1], x[2]),
weights=weights)
ck_metric.attach(engine, "cohen_kappa")
data = list(range(size // batch_size))
ck_value = engine.run(data, max_epochs=1).metrics["cohen_kappa"]
assert ck_value == pytest.approx(ck_value_sk)
def _test(y_pred, y, batch_size):
def update_fn(engine, batch):
idx = (engine.state.iteration - 1) * batch_size
y_true_batch = np_y[idx : idx + batch_size]
y_pred_batch = np_y_pred[idx : idx + batch_size]
return idx, torch.from_numpy(y_pred_batch), torch.from_numpy(y_true_batch)
engine = Engine(update_fn)
ck_metric = CohenKappa(output_transform=lambda x: (x[1], x[2]), weights=weights)
ck_metric.attach(engine, "ck")
np_y = y.numpy()
np_y_pred = y_pred.numpy()
np_ck = cohen_kappa_score(np_y, np_y_pred, weights=weights)
data = list(range(y_pred.shape[0] // batch_size))
ck = engine.run(data, max_epochs=1).metrics["ck"]
assert isinstance(ck, float)
assert np_ck == pytest.approx(ck)