def test_zero_div(): a = torch.tensor([2.0, -1.0, -1.0, 2.0]) ground_truth = torch.tensor([0.0, 0.5, 0.2, 1.0]) m = MeanAbsoluteRelativeError() with raises(NotComputableError): m.update((a, ground_truth))
def test_wrong_input_shapes(): m = MeanAbsoluteRelativeError() with raises(ValueError): m.update((torch.rand(4, 1, 2), torch.rand(4, 1))) with raises(ValueError): m.update((torch.rand(4, 1), torch.rand(4, 1, 2))) with raises(ValueError): m.update((torch.rand(4, 1, 2), torch.rand(4,))) with raises(ValueError): m.update((torch.rand(4,), torch.rand(4, 1, 2)))
def test_wrong_input_shapes(): m = MeanAbsoluteRelativeError() with raises(ValueError, match=r"Input data shapes should be the same, but given"): m.update((torch.rand(4), torch.rand(4, 1))) with raises(ValueError, match=r"Input data shapes should be the same, but given"): m.update((torch.rand(4, 1), torch.rand(4,)))
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.rand(size=(offset * idist.get_world_size(), )).to(device) y_preds = torch.rand(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) m = MeanAbsoluteRelativeError(device=metric_device) m.attach(engine, "mare") data = list(range(n_iters)) engine.run(data=data, max_epochs=n_epochs) assert "mare" in engine.state.metrics mare = engine.state.metrics["mare"] np_y_true = y_true.cpu().numpy() np_y_preds = y_preds.cpu().numpy() abs_error = np.sum(abs(np_y_true - np_y_preds) / abs(np_y_true)) num_samples = len(y_preds) np_res = abs_error / num_samples assert approx(mare) == np_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 torch.from_numpy(y_pred_batch), torch.from_numpy(y_true_batch) engine = Engine(update_fn) m = MeanAbsoluteRelativeError() m.attach(engine, "mare") np_y = y.numpy().ravel() np_y_pred = y_pred.numpy().ravel() data = list(range(y_pred.shape[0] // batch_size)) mare = engine.run(data, max_epochs=1).metrics["mare"] abs_error = np.sum(abs(np_y - np_y_pred) / abs(np_y)) num_samples = len(y_pred) res = abs_error / num_samples assert res == approx(mare)
def _test(metric_device): metric_device = torch.device(metric_device) m = MeanAbsoluteRelativeError(device=metric_device) torch.manual_seed(10 + rank) y_pred = torch.randint(1, 11, size=(10,), device=device).float() y = torch.randint(1, 11, size=(10,), device=device).float() m.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() res = m.compute() abs_error = np.sum(abs(np_y - np_y_pred) / abs(np_y)) num_samples = len(y_pred) np_res = abs_error / num_samples assert np_res == approx(res)
def test_zero_sample(): m = MeanAbsoluteRelativeError() with raises(NotComputableError): m.compute()
def test_mean_absolute_relative_error(): a = torch.rand(4) b = torch.rand(4) c = torch.rand(4) d = torch.rand(4) ground_truth = torch.rand(4) m = MeanAbsoluteRelativeError() m.update((a, ground_truth)) abs_error_a = torch.sum( torch.abs(ground_truth - a) / torch.abs(ground_truth)) num_samples_a = a.size()[0] sum_error = abs_error_a sum_samples = num_samples_a MARE_a = sum_error / sum_samples assert m.compute() == approx(MARE_a.item()) m.update((b, ground_truth)) abs_error_b = torch.sum( torch.abs(ground_truth - b) / torch.abs(ground_truth)) num_samples_b = b.size()[0] sum_error += abs_error_b sum_samples += num_samples_b MARE_b = sum_error / sum_samples assert m.compute() == approx(MARE_b.item()) m.update((c, ground_truth)) abs_error_c = torch.sum( torch.abs(ground_truth - c) / torch.abs(ground_truth)) num_samples_c = c.size()[0] sum_error += abs_error_c sum_samples += num_samples_c MARE_c = sum_error / sum_samples assert m.compute() == approx(MARE_c.item()) m.update((d, ground_truth)) abs_error_d = torch.sum( torch.abs(ground_truth - d) / torch.abs(ground_truth)) num_samples_d = d.size()[0] sum_error += abs_error_d sum_samples += num_samples_d MARE_d = sum_error / sum_samples assert m.compute() == approx(MARE_d.item())
def test_zero_sample(): m = MeanAbsoluteRelativeError() with raises( NotComputableError, match=r"MeanAbsoluteRelativeError must have at least one sample"): m.compute()