osl_dynamics.analysis.static#

Functions to calculate static properties of time series data.

Functions#

`functional_connectivity`(data[, conn_type, demean, ...])	Calculate functional connectivity networks.
`welch_spectra`(data, sampling_frequency[, ...])	Calculate spectra using Welch's method.
`multitaper_spectra`(data, sampling_frequency[, ...])	Calculate multitaper spectra.

Module Contents#

osl_dynamics.analysis.static.functional_connectivity(data, conn_type='corr', demean=True, window_length=None, n_jobs=1)[source]#

Calculate functional connectivity networks.

This function uses the LedoitWolf estimator to calculate covariance. If another measure is requested it is calculated from this covariance matrix.

Parameters:

data (np.ndarray) – Time series data. Shape must be (n_sessions, n_samples, n_channels) or (n_samples, n_channels).
conn_type (str or func, optional) –
What measure should we use? Must be one of:
- "cov": covariance.
- "pcov": partial covariance.
- "corr": Pearson correlation.
- "pcorr": partial correlation.
- "prec": precision (inverse of the covariance matrix).
Or can be a function that accepts a (n_samples, n_channels) array and returns a (n_channels, n_channels) array.
demean (bool, optional) – Should we demean the data before calculating the functional connectivity?
window_length (int, optional) – Window length (in samples) to split the data and average the functional connectivity network for. If None, no averaging is done. Windows are non-overlapping.
n_jobs (int, optional) – Number of jobs.

Returns:

fc – Functional connectivity network. Shape is (n_sessions, n_channels, n_channels) or (n_channels, n_channels).

Return type:

np.ndarray

osl_dynamics.analysis.static.welch_spectra(data, sampling_frequency, window_length=None, step_size=None, frequency_range=None, standardize=True, averaging='mean', calc_cpsd=False, calc_coh=False, return_weights=False, keepdims=False, n_jobs=1)[source]#

Calculate spectra using Welch’s method.

Wrapper for spectral.welch_spectra assuming only one state is active for all time points.

Parameters:

data (np.ndarray or list) – Time series data. Must have shape (n_sessions, n_samples, n_channels) or (n_samples, n_channels).
sampling_frequency (float) – Sampling frequency in Hz.
window_length (int, optional) – Length of the data segment to use to calculate spectra. If None, we use 2 * sampling_frequency.
step_size (int, optional) – Step size for shifting the window. If None, we use window_length // 2.
frequency_range (list, optional) – Minimum and maximum frequency to keep.
standardize (bool, optional) – Should we standardize the data before calculating the spectra?
averaging (str, optional) – Method used to average periodograms. Must be 'mean' or 'median'.
calc_cpsd (bool, optional) – Should we return the cross spectra for psd? If True, we force calc_coh to False.
calc_coh (bool, optional) – Should we also return the coherence spectra?
return_weights (bool, optional) – Should we return the weights for session-specific spectra? This is useful for calculating a group average.
keepdims (bool, optional) – Should we enforce a (n_sessions, …) array is returned for psd and coh? If False, we remove any dimension of length 1.
n_jobs (int, optional) – Number of parallel jobs.

Returns:

f (np.ndarray) – Frequencies of the power spectra and coherences. Shape is (n_freq,).
psd (np.ndarray) – Power spectra for each session. Shape is (n_sessions, n_channels, n_freq) if calc_cpsd=False, otherwise the shape is (n_sessions, n_channels, n_channels, n_freq). Any axis of length 1 is removed if keepdims=False.
coh (np.ndarray) – Coherences. Shape is (n_sessions, n_channels, n_channels, n_freq). Any axis of length 1 is removed if keepdims=False. Only returned is calc_coh=True.
w (np.ndarray) – Weighting for session-specific spectra. Only returned if return_weights=True. Shape is (n_sessions,).

osl_dynamics.analysis.static.multitaper_spectra(data, sampling_frequency, window_length=None, time_half_bandwidth=4, n_tapers=7, frequency_range=None, standardize=True, averaging='mean', calc_cpsd=False, calc_coh=False, return_weights=False, keepdims=False, n_jobs=1)[source]#

Calculate multitaper spectra.

Wrapper for spectral.multitaper_spectra assuming only one state is active for all time points.

Parameters:

data (np.ndarray or list) – Time series data. Must have shape (n_sessions, n_samples, n_channels) or (n_samples, n_channels).
sampling_frequency (float) – Sampling frequency in Hz.
window_length (int, optional) – Length of the data segment to use to calculate the multitaper.
time_half_bandwidth (float, optional) – Parameter to control the resolution of the spectra.
n_tapers (int, optional) – Number of tapers.
frequency_range (list, optional) – Minimum and maximum frequency to keep.
standardize (bool, optional) – Should we standardize the data before calculating the multitaper?
averaging (str, optional) – Method used to average periodograms. Must be 'mean' or 'median'.
calc_cpsd (bool, optional) – Should we return the cross spectra for psd? If True, we force calc_coh to False.
calc_coh (bool, optional) – Should we also return the coherence spectra?
return_weights (bool, optional) – Should we return the weights for session-specific PSDs? Useful for calculating the group average PSD.
keepdims (bool, optional) – Should we enforce a (n_sessions, …) array is returned for psd and coh? If False, we remove any dimension of length 1.
n_jobs (int, optional) – Number of parallel jobs.

Returns:

f (np.ndarray) – Frequencies of the power spectra and coherences. Shape is (n_freq,).
psd (np.ndarray) – Power spectra for each session. Shape is (n_sessions, n_channels, n_freq) if calc_cpsd=False, otherwise the shape is (n_sessions, n_channels, n_channels, n_freq). Any axis of length 1 is removed if keepdims=False.
coh (np.ndarray) – Coherences. Shape is (n_sessions, n_channels, n_channels, n_freq). Any axis of length 1 is removed if keepdims=False. Only returned is calc_coh=True.
w (np.ndarray) – Weighting for session-specific spectra. Only returned if return_weights=True. Shape is (n_sessions,).