SpectralState.from_result#

classmethod SpectralState.from_result(result, *, active_threshold=None)[source]#

Compute every coordinate from a fitted result.

The weights are formed from the result’s effective_amplitude squared, with a \(10^{-12}\) floor on the row sums so a row of silent components gives a uniform-in-nothing zero weight rather than a division by zero; the entropy floors each weight the same way inside the logarithm. The velocity loop is \(O(r K \log K)\) and the correlation \(O(r K^2)\).

Parameters:

  • result (NSSDNResult) – A fitted NSSDNResult.

  • active_threshold (float | None) – Weight floor in \([0, 1]\); the package default (\(10^{-2}\)) when None.

Return type:

SpectralState

Returns:

The populated state, with nobs equal to the result’s path rows.

Raises:

SpecificationError – If the threshold lies outside \([0, 1]\).

Example

>>> import numpy as np
>>> from ns_sdn.spectral_decomposition import NSSDN, Stage
>>> y = np.sin(0.3 * np.arange(160.0))
>>> res = NSSDN(y, k=2, state_dim=8, schedule=(Stage(1e-2, 20),), seed=0).fit()
>>> state = SpectralState.from_result(res, active_threshold=0.05)
>>> state.active_threshold, state.active.shape, state.active.dtype
(0.05, (160, 2), dtype('bool'))
>>> SpectralState.from_result(res, active_threshold=2.0)
Traceback (most recent call last):
    ...
ns_sdn.exceptions.SpecificationError: active_threshold must lie in [0, 1]; got 2.0.