Custom Dataset Example

This example shows how to convert data X and y as numpy arrays to a braindecode compatible data format.

# Authors: Lukas Gemein <l.gemein@gmail.com>
#
# License: BSD (3-clause)

import mne

from braindecode.datasets import create_from_X_y

To set up the example, we first fetch some data using mne:

# 5, 6, 7, 10, 13, 14 are codes for executed and imagined hands/feet
subject_id = 22
event_codes = [5, 6, 9, 10, 13, 14]
# event_codes = [3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14]

# This will download the files if you don't have them yet,
# and then return the paths to the files.
physionet_paths = mne.datasets.eegbci.load_data(
    subject_id, event_codes, update_path=False)

# Load each of the files
parts = [mne.io.read_raw_edf(path, preload=True, stim_channel='auto')
         for path in physionet_paths]

Out:

Extracting EDF parameters from /home/runner/mne_data/MNE-eegbci-data/files/eegmmidb/1.0.0/S022/S022R05.edf...
EDF file detected
Setting channel info structure...
Creating raw.info structure...
Reading 0 ... 19999  =      0.000 ...   124.994 secs...
Extracting EDF parameters from /home/runner/mne_data/MNE-eegbci-data/files/eegmmidb/1.0.0/S022/S022R06.edf...
EDF file detected
Setting channel info structure...
Creating raw.info structure...
Reading 0 ... 19999  =      0.000 ...   124.994 secs...
Extracting EDF parameters from /home/runner/mne_data/MNE-eegbci-data/files/eegmmidb/1.0.0/S022/S022R09.edf...
EDF file detected
Setting channel info structure...
Creating raw.info structure...
Reading 0 ... 19999  =      0.000 ...   124.994 secs...
Extracting EDF parameters from /home/runner/mne_data/MNE-eegbci-data/files/eegmmidb/1.0.0/S022/S022R10.edf...
EDF file detected
Setting channel info structure...
Creating raw.info structure...
Reading 0 ... 19999  =      0.000 ...   124.994 secs...
Extracting EDF parameters from /home/runner/mne_data/MNE-eegbci-data/files/eegmmidb/1.0.0/S022/S022R13.edf...
EDF file detected
Setting channel info structure...
Creating raw.info structure...
Reading 0 ... 19999  =      0.000 ...   124.994 secs...
Extracting EDF parameters from /home/runner/mne_data/MNE-eegbci-data/files/eegmmidb/1.0.0/S022/S022R14.edf...
EDF file detected
Setting channel info structure...
Creating raw.info structure...
Reading 0 ... 19999  =      0.000 ...   124.994 secs...

We take the required data, targets and additional information sampling frequency and channel names from the loaded data. Note that this data and information can originate from any source.

X = [raw.get_data() for raw in parts]
y = event_codes
sfreq = parts[0].info["sfreq"]
ch_names = parts[0].info["ch_names"]

Convert to data format compatible with skorch and braindecode:

windows_dataset = create_from_X_y(
    X, y, drop_last_window=False, sfreq=sfreq, ch_names=ch_names,
    window_stride_samples=500,
    window_size_samples=500,
)

windows_dataset.description  # look as dataset description

Out:

Creating RawArray with float64 data, n_channels=64, n_times=20000
    Range : 0 ... 19999 =      0.000 ...   124.994 secs
Ready.
Creating RawArray with float64 data, n_channels=64, n_times=20000
    Range : 0 ... 19999 =      0.000 ...   124.994 secs
Ready.
Creating RawArray with float64 data, n_channels=64, n_times=20000
    Range : 0 ... 19999 =      0.000 ...   124.994 secs
Ready.
Creating RawArray with float64 data, n_channels=64, n_times=20000
    Range : 0 ... 19999 =      0.000 ...   124.994 secs
Ready.
Creating RawArray with float64 data, n_channels=64, n_times=20000
    Range : 0 ... 19999 =      0.000 ...   124.994 secs
Ready.
Creating RawArray with float64 data, n_channels=64, n_times=20000
    Range : 0 ... 19999 =      0.000 ...   124.994 secs
Ready.
Adding metadata with 4 columns
Replacing existing metadata with 4 columns
40 matching events found
No baseline correction applied
0 projection items activated
Loading data for 40 events and 500 original time points ...
0 bad epochs dropped
Adding metadata with 4 columns
Replacing existing metadata with 4 columns
40 matching events found
No baseline correction applied
0 projection items activated
Loading data for 40 events and 500 original time points ...
0 bad epochs dropped
Adding metadata with 4 columns
Replacing existing metadata with 4 columns
40 matching events found
No baseline correction applied
0 projection items activated
Loading data for 40 events and 500 original time points ...
0 bad epochs dropped
Adding metadata with 4 columns
Replacing existing metadata with 4 columns
40 matching events found
No baseline correction applied
0 projection items activated
Loading data for 40 events and 500 original time points ...
0 bad epochs dropped
Adding metadata with 4 columns
Replacing existing metadata with 4 columns
40 matching events found
No baseline correction applied
0 projection items activated
Loading data for 40 events and 500 original time points ...
0 bad epochs dropped
Adding metadata with 4 columns
Replacing existing metadata with 4 columns
40 matching events found
No baseline correction applied
0 projection items activated
Loading data for 40 events and 500 original time points ...
0 bad epochs dropped

	target
0	5
1	6
2	9
3	10
4	13
5	14

You can manipulate the dataset

print(len(windows_dataset))  # get the number of samples

Out:

240

You can now index the data

i = 0
x_i, y_i, window_ind = windows_dataset[0]
n_channels, n_times = x_i.shape  # the EEG data
_, start_ind, stop_ind = window_ind
print(f"n_channels={n_channels}  -- n_times={n_times} -- y_i={y_i}")
print(f"start_ind={start_ind} -- stop_ind={stop_ind}")

Out:

Loading data for 1 events and 500 original time points ...
n_channels=64  -- n_times=500 -- y_i=5
start_ind=0 -- stop_ind=500

Estimated memory usage: 96 MB