From 4fb978b71fcd1dda8fe34b5b1e0621f68e1f4dfc Mon Sep 17 00:00:00 2001 From: Akshat Sinha Date: Thu, 8 Jan 2026 20:40:25 +0530 Subject: [PATCH 1/3] feat(metrics): Add MAPEMetric for regression evaluation. Signed-off-by: Akshat Sinha --- monai/metrics/__init__.py | 1 + monai/metrics/regression.py | 49 +++++++++++++++++++++++++++++++++++++ 2 files changed, 50 insertions(+) diff --git a/monai/metrics/__init__.py b/monai/metrics/__init__.py index ae20903cfd..78117d9db6 100644 --- a/monai/metrics/__init__.py +++ b/monai/metrics/__init__.py @@ -28,6 +28,7 @@ from .panoptic_quality import PanopticQualityMetric, compute_panoptic_quality from .regression import ( MAEMetric, + MAPEMetric, MSEMetric, MultiScaleSSIMMetric, PSNRMetric, diff --git a/monai/metrics/regression.py b/monai/metrics/regression.py index 317d624e73..83609bfb7a 100644 --- a/monai/metrics/regression.py +++ b/monai/metrics/regression.py @@ -143,6 +143,38 @@ def _compute_metric(self, y_pred: torch.Tensor, y: torch.Tensor) -> torch.Tensor return compute_mean_error_metrics(y_pred, y, func=self.abs_func) +class MAPEMetric(RegressionMetric): + r"""Compute Mean Absolute Percentage Error between two tensors using function: + + .. math:: + \operatorname {MAPE}\left(Y, \hat{Y}\right) =\frac {100}{n}\sum _{i=1}^{n}\left|\frac{y_i-\hat{y_i}}{y_i}\right|. + + More info: https://en.wikipedia.org/wiki/Mean_absolute_percentage_error + + Input `y_pred` is compared with ground truth `y`. + Both `y_pred` and `y` are expected to be real-valued, where `y_pred` is output from a regression model. + Note: Tackling the undefined error, a tiny epsilon value is added to the denominator part. + + Example of the typical execution steps of this metric class follows :py:class:`monai.metrics.metric.Cumulative`. + Args: + reduction: define the mode to reduce metrics, will only execute reduction on `not-nan` values, + available reduction modes: {``"none"``, ``"mean"``, ``"sum"``, ``"mean_batch"``, ``"sum_batch"``, + ``"mean_channel"``, ``"sum_channel"``}, default to ``"mean"``. if "none", will not do reduction. + get_not_nans: whether to return the `not_nans` count, if True, aggregate() returns (metric, not_nans). + epsilonDefaults to 1e-7. + + """ + + def __init__( + self, reduction: MetricReduction | str = MetricReduction.MEAN, get_not_nans: bool = False, epsilon: float = 1e-7 + ) -> None: + super().__init__(reduction=reduction, get_not_nans=get_not_nans) + self.epsilon = epsilon + + def _compute_metric(self, y_pred: torch.Tensor, y: torch.Tensor) -> torch.Tensor: + return compute_mape_metric(y_pred, y, epsilon=self.epsilon) + + class RMSEMetric(RegressionMetric): r"""Compute Root Mean Squared Error between two tensors using function: @@ -220,6 +252,23 @@ def compute_mean_error_metrics(y_pred: torch.Tensor, y: torch.Tensor, func: Call return torch.mean(flt(func(y - y_pred)), dim=-1, keepdim=True) +def compute_mape_metric(y_pred: torch.Tensor, y: torch.Tensor, epsilon: float = 1e-7) -> torch.Tensor: + """ + Compute Mean Absolute Percentage Error. + + Args: + y_pred: predicted values + y: ground truth values + epsilon: small value to avoid division by zero + + Returns: + MAPE value as percentage + """ + flt = partial(torch.flatten, start_dim=1) + percentage_error = torch.abs((y - y_pred) / (torch.abs(y) + epsilon)) * 100.0 + return torch.mean(flt(percentage_error), dim=-1, keepdim=True) + + class KernelType(StrEnum): GAUSSIAN = "gaussian" UNIFORM = "uniform" From 808350a443f482c57692ba75d108e25d08270b69 Mon Sep 17 00:00:00 2001 From: Akshat Sinha Date: Thu, 8 Jan 2026 20:46:53 +0530 Subject: [PATCH 2/3] calc improvements Signed-off-by: Akshat Sinha --- monai/metrics/regression.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/monai/metrics/regression.py b/monai/metrics/regression.py index 83609bfb7a..1a51068413 100644 --- a/monai/metrics/regression.py +++ b/monai/metrics/regression.py @@ -265,7 +265,7 @@ def compute_mape_metric(y_pred: torch.Tensor, y: torch.Tensor, epsilon: float = MAPE value as percentage """ flt = partial(torch.flatten, start_dim=1) - percentage_error = torch.abs((y - y_pred) / (torch.abs(y) + epsilon)) * 100.0 + percentage_error = torch.abs(y - y_pred) / torch.clamp(torch.abs(y), min=epsilon) * 100.0 return torch.mean(flt(percentage_error), dim=-1, keepdim=True) From f324869edf1d938fdb2141ac5d680d8ef3f4d9f4 Mon Sep 17 00:00:00 2001 From: "pre-commit-ci[bot]" <66853113+pre-commit-ci[bot]@users.noreply.github.com> Date: Thu, 8 Jan 2026 15:17:27 +0000 Subject: [PATCH 3/3] [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci --- monai/metrics/regression.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/monai/metrics/regression.py b/monai/metrics/regression.py index 1a51068413..b3e55279a6 100644 --- a/monai/metrics/regression.py +++ b/monai/metrics/regression.py @@ -255,12 +255,12 @@ def compute_mean_error_metrics(y_pred: torch.Tensor, y: torch.Tensor, func: Call def compute_mape_metric(y_pred: torch.Tensor, y: torch.Tensor, epsilon: float = 1e-7) -> torch.Tensor: """ Compute Mean Absolute Percentage Error. - + Args: y_pred: predicted values y: ground truth values epsilon: small value to avoid division by zero - + Returns: MAPE value as percentage """