def test_integration_ap_score_with_activated_output_transform(): np.random.seed(1) size = 100 np_y_pred = np.random.rand(size, 1) np_y_pred_softmax = torch.softmax(torch.from_numpy(np_y_pred), dim=1).numpy() np_y = np.zeros((size, ), dtype=np.long) np_y[size // 2:] = 1 np.random.shuffle(np_y) np_ap = average_precision_score(np_y, np_y_pred_softmax) 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) ap_metric = AveragePrecision( output_transform=lambda x: (torch.softmax(x[1], dim=1), x[2])) ap_metric.attach(engine, 'ap') data = list(range(size // batch_size)) ap = engine.run(data, max_epochs=1).metrics['ap'] assert ap == np_ap
def _test(y_preds, y_true, n_epochs, metric_device, update_fn): metric_device = torch.device(metric_device) engine = Engine(update_fn) ap = AveragePrecision(device=metric_device) ap.attach(engine, "ap") data = list(range(n_iters)) engine.run(data=data, max_epochs=n_epochs) assert "ap" in engine.state.metrics res = engine.state.metrics["ap"] true_res = average_precision_score(y_true.cpu().numpy(), y_preds.cpu().numpy()) assert pytest.approx(res) == true_res
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(), 10)).to(device) y_preds = torch.randint(0, n_classes, size=(offset * idist.get_world_size(), 10)).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) ap = AveragePrecision(device=metric_device) ap.attach(engine, "ap") data = list(range(n_iters)) engine.run(data=data, max_epochs=n_epochs) assert "ap" in engine.state.metrics res = engine.state.metrics["ap"] if isinstance(res, torch.Tensor): res = res.cpu().numpy() true_res = average_precision_score(y_true.cpu().numpy(), y_preds.cpu().numpy()) assert pytest.approx(res) == true_res
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) ap_metric = AveragePrecision(output_transform=lambda x: (x[1], x[2])) ap_metric.attach(engine, "ap") np_y = y.numpy() np_y_pred = y_pred.numpy() np_ap = average_precision_score(np_y, np_y_pred) data = list(range(y_pred.shape[0] // batch_size)) ap = engine.run(data, max_epochs=1).metrics["ap"] assert isinstance(ap, float) assert np_ap == pytest.approx(ap)