kaio-99
commented
5 months ago Thank you for your interest in our work. Here is an example of processing pipeline using MAHNOB-HCI dataset, you can make minor changes to migrate to DEAP dataset and set looping conditions to batch process all data.
raw = mne.io.read_raw_bdf(file_path, preload=True, exclude=['Status'])
# get the total length of EEG signal
total_duration = raw.times[-1]
# calculate the time range
start_time = total_duration - 30
end_time = total_duration
if start_time < 0:
break;
raw_last_30s = raw.copy().crop(tmin=start_time, tmax=end_time)
# choose the bands
freq_ranges = {'4-8Hz': (4, 8),
'8-10Hz': (8, 10),
'8-12Hz': (8, 12),
'12-30Hz': (12, 30),
'30-40Hz': (30, 40)}
# define the window size, number of windows, and overlap ratio
win_size_sec = 3
win_size = int(win_size_sec * raw_last_30s.info['sfreq'])
overlap = 0
n_channels = len(raw_last_30s.info['ch_names'])
n_samples = raw_last_30s.n_times
n_windows = int(np.floor((n_samples - win_size) / (win_size - overlap))) + 1
freq_features = np.zeros((len(freq_ranges), n_channels, n_windows))
for i in range(n_windows):
start = i * (win_size - overlap)
end = start + win_size
for ch_idx in range(n_channels):
data = raw_last_30s.get_data(picks=ch_idx, start=start, stop=end)[0]
freqs, psd = welch(data, fs=raw_last_30s.info['sfreq'], nperseg=win_size, noverlap=overlap)
for f_idx, (f_name, f_range) in enumerate(freq_ranges.items()):
f_mask = (freqs >= f_range[0]) & (freqs <= f_range[1])
freq_features[f_idx, ch_idx, i] = np.mean(psd[f_mask])
feat = freq_features.reshape(1600)
After browsing the code, I found that the two inputs and labels of deap were processed into numpy arrays. Could you please tell me how to get from the original deap data set to these three arrays? Could you please provide the relevant codes? Looking forward to your reply, thank you very much.