MeanSquaredLogError#

class torch_uncertainty.metrics.regression.MeanSquaredLogError(squared=True, **kwargs)[source]#

Compute the Mean Squared Logarithmic Error (MSLE).

This metric is commonly used in regression problems where the relative difference between predictions and targets is of greater importance than the absolute difference. It is particularly effective for datasets with wide-ranging magnitudes, as it penalizes underestimation more than overestimation.

\[\text{MSELog} = \frac{1}{N}\sum_i^N (\log \hat{y_i} - \log y_i)^2\]

where \(y\) is a tensor of target values, and \(\hat{y}\) is a tensor of predictions.

As input to forward and update the metric accepts the following input:

  • preds (Tensor): Predictions from model

  • target (Tensor): Ground truth values

As output of forward and compute the metric returns the following output:

  • mse_log (Tensor): A tensor with the

    mean squared logarithmic error over the state

Parameters:

  • squared (bool) – If True, returns MSLE. If False, returns RMSLE.

  • kwargs – Additional keyword arguments, see Advanced metric settings.

Reference:

[1] From big to small: Multi-scale local planar guidance for monocular depth estimation.

Example:

from torch_uncertainty.metrics.regression import MeanSquaredLogError
import torch

# Initialize the metric
msle_metric = MeanSquaredLogError(squared=True)

# Example predictions and targets (must be non-negative)
preds = torch.tensor([2.5, 1.0, 2.0, 8.0])
target = torch.tensor([3.0, 1.5, 2.0, 7.0])

# Update the metric state
msle_metric.update(preds, target)

# Compute the Mean Squared Logarithmic Error
result = msle_metric.compute()
print(f"Mean Squared Logarithmic Error: {result.item()}")
# Output: Mean Squared Logarithmic Error: 0.05386843904852867
update(preds, target)[source]#

Update state with predictions and targets.

Return type:

None