Log10#

class torch_uncertainty.metrics.regression.Log10(**kwargs)[source]#

Computes the LOG10 metric.

The Log10 metric computes the mean absolute error in the base-10 logarithmic space.

\[\text{Log10} = \frac{1}{N} \sum_{i=1}^{N} |\log_{10}(y_i) - \log_{10}(\hat{y_i})|\]

where: - \(N\) is the number of elements in the batch. - \(y_i\) represents the true target values. - \(\hat{y_i}\) represents the predicted values.

This metric is useful for scenarios where the data spans multiple orders of magnitude, and evaluating error in log-space provides a more meaningful comparison.

Inputs: - preds: \((N)\) - target: \((N)\)

Parameters:

kwargs – Additional keyword arguments, see Advanced metric settings.

Example:

from torch_uncertainty.metrics.regression import Log10
import torch

# Initialize the metric
log10_metric = Log10()

# Example predictions and targets
preds = torch.tensor([10.0, 100.0, 1000.0])
target = torch.tensor([12.0, 95.0, 1020.0])

# Update the metric state
log10_metric.update(preds, target)

# Compute the Log10 error
result = log10_metric.compute()
print(f"Log10 Error: {result.item()}")
# Output: Log10 Error: 0.03668594
update(pred, target)[source]#

Update state with predictions and targets.