metrics

Implementation of metrics for (mulit-horizon) timeseries forecasting.

Classes

AggregationMetric(metric, **kwargs)	Calculate metric on mean prediction and actuals.
BetaDistributionLoss(name, quantiles, 0.1, …)	Beta distribution loss for unit interval data.
CompositeMetric(metrics, weights)	Metric that combines multiple metrics.
CrossEntropy(reduction, **kwargs)	Cross entropy loss for classification.
DistributionLoss(name, quantiles, 0.1, 0.25, …)	DistributionLoss base class.
LogNormalDistributionLoss(name, quantiles, …)	Log-normal loss.
MAE(reduction, **kwargs)	Mean average absolute error.
MAPE(reduction, **kwargs)	Mean absolute percentage.
MASE(reduction, **kwargs)	Mean absolute scaled error
Metric(name, quantiles[, reduction])	Base metric class that has basic functions that can handle predicting quantiles and operate in log space.
MultiHorizonMetric(reduction, **kwargs)	Abstract class for defining metric for a multihorizon forecast
MultiLoss(metrics, weights)	Metric that can be used with muliple metrics.
NegativeBinomialDistributionLoss(name, …)	Negative binomial loss, e.g.
NormalDistributionLoss(name, quantiles, 0.1, …)	Normal distribution loss.
PoissonLoss(reduction, **kwargs)	Poisson loss for count data
QuantileLoss(quantiles, 0.1, 0.25, 0.5, …)	Quantile loss, i.e. a quantile of q=0.5 will give half of the mean absolute error as it is calcualted as.
RMSE([reduction])	Root mean square error
SMAPE(reduction, **kwargs)	Symmetric mean absolute percentage.

Metrics AggregationMetric