get_forecaster

pymgrid.forecast.get_forecaster(forecaster, observation_space, forecast_shape, time_series=None, increase_uncertainty=False, relative_noise=False)

Get the forecasting function for the time series module.

Parameters

forecastercallable, float, “oracle”, or None, default None.

Function that gives a forecast n-steps ahead.

• If callable, must take as arguments (val_c: float, val_{c+n}: float, n: int), where

  • val_c is the current value in the time series: self.time_series[self.current_step]

  • val_{c+n} is the value in the time series n steps in the future

  • n is the number of steps in the future at which we are forecasting.

  The output forecast = forecaster(val_c, val_{c+n}, n) must have the same sign as the inputs val_c and val_{c+n}.

• If float, serves as a standard deviation for a mean-zero gaussian noise function that is added to the true value.

• If "oracle", gives a perfect forecast.

• If None, no forecast.

forecast_shapeint or tuple of int

Expected shape of forecasts. If an integer, will return forecasts of shape (shape, 1).

observation_spaceModuleSpace

Observation space; used to determine values to pad missing forecasts when we are forecasting past the end of the time series.

time_series: ndarray[float] or None, default None.

The underlying time series.

• If callable(forecaster), used to validate UserDefinedForecaster.

• If forecaster is numeric and relative_noise is true, the noise standard deviation will be defined relative to the mean of the time series.

increase_uncertaintybool, default False

Whether to increase uncertainty for farther-out dates if using GaussianNoiseForecaster. Ignored otherwise.

relative_noisebool, default False

Whether to define noise standard deviation relative to mean of time series if using GaussianNoiseForecaster. Ignored otherwise.

Returns

forecasterForecaster

The forecasting function.