evaluate¶
Evaluates the performance of a forecaster on a time series dataset.
To understand why this method is useful, it's important to understand the difference between nowcasting and forecasting. Nowcasting is about predicting a value at the next time step. This can be seen as a special case of regression, where the value to predict is the value at the next time step. In this case, the evaluate.progressive_val_score
function may be used to evaluate a model via progressive validation.
Forecasting models can also be evaluated via progressive validation. This is the purpose of this function. At each time step t
, the forecaster is asked to predict the values at t + 1
, t + 2
, ..., t + horizon
. The performance at each time step is measured and returned.
Parameters¶
-
dataset (Iterable[Tuple[dict, Any]])
A sequential time series.
-
model (river.time_series.base.Forecaster)
A forecaster.
-
metric (river.metrics.base.RegressionMetric)
A regression metric.
-
horizon (int)
-
grace_period (int) – defaults to
None
Initial period during which the metric is not updated. This is to fairly evaluate models which need a warming up period to start producing meaningful forecasts. The value of this parameter is equal to the horizon by default.