DensityScore.compute

DensityScore.compute()

Evaluate every score.

The CRPS is computed on every (horizon, series) column at once by flattening the sample to (n_draws, h * k); the energy score once per horizon on the (n_draws, k) joint sample; the log score through _kernel_log_score(); and the point losses from the sample mean (squared error) and median (absolute error). Deterministic: nothing here draws.

Return type:

DensityScoreResult

Returns:

The DensityScoreResult.

Example

>>> import numpy as np
>>> rng = np.random.default_rng(0)
>>> paths, outcome = rng.standard_normal((2000, 4, 2)), rng.standard_normal((4, 2))
>>> record = DensityScore(paths, outcome, names=("gdp", "cpi")).compute()
>>> record.names, record.crps.shape, record.energy.shape, record.n_draws
(('gdp', 'cpi'), (4, 2), (4,), 2000)
>>> bool(np.allclose(record.squared_error, (paths.mean(axis=0) - outcome) ** 2))
True