import numpy as np
import pandas as pd
import scipy
import mne
from typing import List, Optional, Dict, Any
[docs]
def bciciv2b(subject: int=1,
session_list: Optional[List[str]] = None,
labels: List[str] = ['left-hand', 'right-hand'],
path: str = 'data/BCICIV2b/') -> Dict[str, Any]:
"""
Description
-----------
This function loads EEG data for a specific subject and session from the bciciv2b dataset.
It processes the data to fit the structure of the `eegdata` dictionary, which is used
for further processing and analysis.
The dataset can be found at:
- https://www.bbci.de/competition/iv/#download
- https://www.bbci.de/competition/iv/results/index.html#labels
Parameters
----------
subject : int
index of the subject to retrieve the data from
session_list : list, optional
list of session codes
labels : dict
dictionary mapping event names to event codes
path :
path to the directory tha contains the datasets files.
Returns
-------
dict
A dictionary containing the following keys:
- X: EEG data as a numpy array [trials, 1, channels, time].
- y: Labels corresponding to the EEG data.
- sfreq: Sampling frequency of the EEG data.
- y_dict: Mapping of labels to integers.
- events: Dictionary describing event markers.
- ch_names: List of channel names.
- tmin: Start time of the EEG data.
- data_type: Type of the data ('epochs').
Examples
--------
Load EEG data for subject 1, all sessions, and default labels:
>>> from bciflow.datasets import bciciv2b
>>> eeg_data = bciciv2b(subject=1)
>>> print(eeg_data['X'].shape) # Shape of the EEG data
>>> print(eeg_data['y']) # Labels
'''
"""
if type(subject) != int:
raise ValueError("Has to be a int type value")
if subject > 9 or subject < 1:
raise ValueError("Has to be an existing subject")
if type(labels) != list:
raise ValueError("labels has to be a list type value")
for i in labels:
if i not in ['left-hand', 'right-hand']:
raise ValueError("labels has to be a sublist of ['left-hand', 'right-hand']")
if type(session_list) != list and session_list != None:
raise ValueError("Has to be an List or None type")
if path[-1] != '/':
path += '/'
sfreq = 250.
events = {'get_start': [0, 3],
'beep_sound': [2],
'cue': [3, 4],
'task_exec': [4, 7],
'break': [7, 8.5]}
ch_names = ['C3', 'Cz', 'C4']
ch_names = np.array(ch_names)
tmin = 0.
if session_list is None:
session_list = ['01T', '02T', '03T', '04E', '05E']
rawData, rawLabels = [], []
for sec in session_list:
raw=mne.io.read_raw_gdf(path+'B%02d%s.gdf'%(subject, sec), preload=True, verbose='ERROR')
raw_data = raw.get_data()[:3]
annotations = raw.annotations.to_data_frame()
first_timestamp = pd.to_datetime(annotations['onset'].iloc[0])
annotations['onset'] = (pd.to_datetime(annotations['onset']) - first_timestamp).dt.total_seconds()
annotations['description'] = annotations['description'].astype(int)
new_trial_time = np.array(annotations[annotations['description']==768]['onset'])
times_ = np.array(raw.times)
rawData_ = []
for trial_ in new_trial_time:
idx_ = np.where(times_ == trial_)[0][0]
rawData_.append(raw_data[:, idx_:idx_+2125])
rawData_ = np.array(rawData_)
rawLabels_ = np.array(scipy.io.loadmat(path+'B%02d%s.mat'%(subject, sec))['classlabel']).reshape(-1)
rawData.append(rawData_)
rawLabels.append(rawLabels_)
X, y = np.concatenate(rawData), np.concatenate(rawLabels)
X = np.reshape(X, (X.shape[0], 1, X.shape[1], X.shape[2]))
labels_dict = {1: 'left-hand', 2: 'right-hand'}
y = np.array([labels_dict[i] for i in y])
selected_labels = np.isin(y, labels)
X, y = X[selected_labels], y[selected_labels]
y_dict = {labels[i]: i for i in range(len(labels))}
y = np.array([y_dict[i] for i in y])
return {'X': X,
'y': y,
'sfreq': sfreq,
'y_dict': labels_dict,
'events': events,
'ch_names': ch_names,
'tmin': tmin,
'data_type': "epochs"}