Regression
Mean squared logarithmic error
Implemented
regression.msleDefinition
Mean squared difference of log(1 + y); emphasises relative error and penalises under-prediction.
Formula
(1/n) Σ (log(1 + ŷᵢ) − log(1 + yᵢ))²
Range: [0, ∞)
Inputs and outputs
- y_true: array of real targets
- y_pred: array of real predictions
- sample_weight: optional
Returns: MetricResult
Assumptions
No assumptions beyond valid, aligned inputs of the documented types.
Limitations
No metric-specific limitations are documented yet. Interpret the value alongside the task, data, and other metrics.
Python API
import evalsuite as es
es.msle(yr, pr)References
- Hyndman RJ, Koehler AB. Another look at measures of forecast accuracy. Int J Forecast. 2006;22(4):679-688.