coco_pipe.descriptors.extractors._psd

Shared PSD computation for PSD-consuming descriptor paths.

This module holds the reusable PSD step used by the descriptors planner and by PSD-consuming extractors when they need a standalone spectral input. It does not define descriptor semantics. It only:

  • prepare a writable runtime environment for MNE-backed PSD helpers

  • lazily import the MNE PSD functions used by descriptors

  • compute Welch or multitaper PSD batches on explicit NumPy inputs

Author: Hamza Abdelhedi (hamza.abdelhedi@umontreal.ca)

Functions

load_mne_psd_functions()

Lazily import MNE PSD helpers with writable runtime cache locations.

compute_psd(→ tuple[numpy.ndarray, numpy.ndarray])

Compute PSD values for one batch of segmented signals.

Module Contents

coco_pipe.descriptors.extractors._psd.load_mne_psd_functions()[source]

Lazily import MNE PSD helpers with writable runtime cache locations.

Returns:

(psd_array_welch, psd_array_multitaper) imported from mne.time_frequency.

Return type:

tuple

Notes

MNE may write cache or config files during import/use. The descriptors module keeps those paths inside the system temp directory so PSD computation remains sandbox-friendly.

coco_pipe.descriptors.extractors._psd.compute_psd(X: numpy.ndarray, sfreq: float, method: str, fmin: float, fmax: float, n_jobs: int | None = None) → tuple[numpy.ndarray, numpy.ndarray][source]

Compute PSD values for one batch of segmented signals.

Parameters:

  • X (np.ndarray) – Input array with shape (n_obs, n_channels, n_times).

  • sfreq (float) – Sampling frequency in Hertz.

  • method ({"welch", "multitaper"}) – PSD estimator to use.

  • fmin (float) – Lower frequency bound passed to the PSD backend.

  • fmax (float) – Upper frequency bound passed to the PSD backend.

  • n_jobs (int, optional) – Parallel worker count forwarded to the MNE PSD backend when the caller enables PSD-level parallelism. None leaves the backend default in place.

Returns:

PSD values with shape (n_obs, n_channels, n_freqs) and the aligned frequency grid with shape (n_freqs,).

Return type:

tuple[np.ndarray, np.ndarray]

Notes

For Welch PSDs, the descriptors module uses:

  • n_fft = min(n_times, 256)

  • n_per_seg = n_fft

while enforcing a minimum of 8 for both values. This keeps Welch behavior bounded and deterministic across the current descriptor tests and examples.