sleepless.data.transforms#

Signal transformations for our pipelines.

Classes

EEGPowerBand([pick_chan])

Extract feature from a py:class:DelayedSample list.

FeatureExtractorChambon([pick_chan])

Extract feature from a py:class:DelayedSample list.

RawToEpochs(event_id, chunk_duration[, ...])

Transform Raw and Annotation objects to Epochs object, with some preprocessing options.

ResampleEpochs(sampling_freq)

Resample mne.Epochs object from a py:class:DelayedSample list.

ToTorchDataset([normalize, pick_chan, ...])

Build Torch dataset from a py:class:DelayedSample list.
class sleepless.data.transforms.RawToEpochs(event_id, chunk_duration, picks_chan=None, no_overlapping=True, crop_wake_time=0.0, wake_stage_name=None)[source]#

Bases: object

Transform Raw and Annotation objects to Epochs object, with some preprocessing options. Data are not loaded in memory until mne.Epochs.get_data() is call.

This class was inspired from version 1.4 of https://mne.tools/stable/auto_tutorials/clinical/60_sleep.html

Parameters:

  • raw_obj – A mne raw object

  • annot_obj – A mne object for annotating segments of raw data

  • event_id (dict[str, int]) – Map stage_name (str=keys) and interger event codes (int=values)

  • chunk_duration (float) – The window time (in seconde) that was used to annotated

  • picks_chan (UnionType[str, list[str], None]) – List of channels to keep (e.g. [‘eeg’,’eog’,’emg’]), if None (default)

  • no_overlapping (bool) – If True remove last sample from epoch to avoid overlapping, Default (False)

  • crop_wake_time (float) – Wake time (in minute) to keep at the begining and the end, in some case it is usefull to crop a part of the wake time if it is too long regarding the other stage or not usefull (e.g. walking time)

  • wake_stage_name (Optional[str]) – Only needed for crop_wake_time. (e.g. “Sleep stage W”)

transform(raw_obj, annot_obj)[source]#

Return the mne.Epochs object from raw data and label of a sample.

Return type:

Epochs

Returns:

A mne Epochs object

class sleepless.data.transforms.EEGPowerBand(pick_chan=None)[source]#

Bases: object

Extract feature from a py:class:DelayedSample list.

This class was copied and modified from https://mne.tools/stable/auto_tutorials/clinical/60_sleep.html v1.4

Modification: change to class, adding pick_chan attribute, management division by zero

Parameters:

pick_chan (list[str]) – the channel type (e.g. “eeg”,”eog”) or name (e.g. “Fpz-Cz”) whom extract the features, if None default compute features for all EEG channels.

Returns:

py:class:DelayedSample list where features have been extracted

class sleepless.data.transforms.ToTorchDataset(normalize=False, pick_chan=None, n_past_epochs=0)[source]#

Bases: object

Build Torch dataset from a py:class:DelayedSample list.

Parameters:

  • normalize (bool) – If True, normalized the sample

  • pick_chan (list[str]) – the channel type (e.g. “eeg”,”eog”) or name (e.g. “Fpz-Cz”) whom extract the features, if None default compute features for all EEG channels.

  • n_past_epochs (int) – number of precedent epochs to include in the ListSampleDataset object (by concatenation).

Returns:

torch.utils.data.dataset.ConcatDataset of all samples

class sleepless.data.transforms.ResampleEpochs(sampling_freq)[source]#

Bases: object

Resample mne.Epochs object from a py:class:DelayedSample list.

Parameters:

sampling_freq (int) – sampling frequency to which resample

Returns:

py:class:DelayedSample list with resampled data

class sleepless.data.transforms.FeatureExtractorChambon(pick_chan=None)[source]#

Bases: object

Extract feature from a py:class:DelayedSample list. 26 manually chosen features (total power (5), relative power (5), power ratio (10), spectral entropy, mean, variance, skewness, kurtosis, 75% quantile.)

This class was copied and modified from https://mne.tools/stable/auto_tutorials/clinical/60_sleep.html v1.4

Modification: change to class, adding pick_chan attribute, management division by zero, adding computation of more features

Reference: [Chambon-2018]

Parameters:

pick_chan (list[str]) – the channel type (e.g. “eeg”,”eog”) or name (e.g. “Fpz-Cz”) whom extract the features, if None default compute features for all EEG channels.

Returns:

py:class:DelayedSample list where features have been extracted