Note
Go to the end to download the full example code.
Cross-session motor imagery with deep learning EEGNet v4 model#
This example shows how to use BrainDecode in combination with MOABB evaluation. In this example, we use the architecture EEGNet.
# Authors: Igor Carrara <igor.carrara@inria.fr>
# Bruno Aristimunha <b.aristimunha@gmail.com>
#
# License: BSD (3-clause)
import matplotlib.pyplot as plt
import mne
import seaborn as sns
import torch
from moabb.datasets import BNCI2014_001
from moabb.evaluations import CrossSessionEvaluation
from moabb.paradigms import MotorImagery
from moabb.utils import setup_seed
from sklearn.pipeline import make_pipeline
from skorch.callbacks import EarlyStopping, EpochScoring
from skorch.dataset import ValidSplit
from braindecode import EEGClassifier
from braindecode.models import EEGNet
mne.set_log_level(False)
# Print Information PyTorch
print(f"Torch Version: {torch.__version__}")
# Set up GPU if it is there
cuda = torch.cuda.is_available()
device = "cuda" if cuda else "cpu"
print("GPU is", "AVAILABLE" if cuda else "NOT AVAILABLE")
Torch Version: 2.9.1+cu128
GPU is NOT AVAILABLE
In this example, we will use only the dataset BNCI2014_001.
Running the benchmark#
This example uses the CrossSession evaluation procedure. We focus on the dataset BNCI2014_001 and only on 1 subject to reduce computational time.
To keep the computational time low, the epoch is reduced. In a real situation, we suggest using the following: EPOCH = 1000 PATIENCE = 300
This code is implemented to run on the CPU. If you’re using a GPU, do not use multithreading (i.e. set n_jobs=1)
# Set random seed to be able to reproduce results
seed = 42
setup_seed(seed)
# Ensure that all operations are deterministic on GPU (if used) for reproducibility
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
# Hyperparameter
LEARNING_RATE = 0.0625 * 0.01 # parameter taken from Braindecode
WEIGHT_DECAY = 0 # parameter taken from Braindecode
BATCH_SIZE = 64 # parameter taken from BrainDecode
EPOCH = 10
PATIENCE = 3
fmin = 4
fmax = 100
tmin = 0
tmax = None
# Load the dataset
dataset = BNCI2014_001()
events = ["right_hand", "left_hand"]
paradigm = MotorImagery(
events=events, n_classes=len(events), fmin=fmin, fmax=fmax, tmin=tmin, tmax=tmax
)
subjects = [1]
X, _, _ = paradigm.get_data(dataset=dataset, subjects=subjects)
Create Pipelines#
In order to create a pipeline, we need to load a model from braindecode.
the second step is to define a skorch model using EEGClassifier from braindecode
that allows converting the PyTorch model in a scikit-learn classifier.
Here, we will use the EEGNet v4 model [1] .
This model has mandatory hyperparameters (the number of channels, the number of classes,
and the temporal length of the input) but we do not need to specify them because they will
be set dynamically by EEGClassifier using the input data during the call to the .fit() method.
# Define a Skorch classifier
clf = EEGClassifier(
module=EEGNet,
optimizer=torch.optim.Adam,
optimizer__lr=LEARNING_RATE,
batch_size=BATCH_SIZE,
max_epochs=EPOCH,
train_split=ValidSplit(0.2, random_state=seed),
device=device,
callbacks=[
EarlyStopping(monitor="valid_loss", patience=PATIENCE),
EpochScoring(
scoring="accuracy", on_train=True, name="train_acc", lower_is_better=False
),
EpochScoring(
scoring="accuracy", on_train=False, name="valid_acc", lower_is_better=False
),
],
verbose=1, # Not printing the results for each epoch
)
# Create the pipelines
pipes = {}
pipes["EEGNet"] = make_pipeline(clf)
Evaluation#
dataset.subject_list = dataset.subject_list[:2]
evaluation = CrossSessionEvaluation(
paradigm=paradigm,
datasets=dataset,
suffix="braindecode_example",
overwrite=True,
return_epochs=True,
n_jobs=1,
)
results = evaluation.process(pipes)
print(results.head())
BNCI2014-001-CrossSession: 0%| | 0/2 [00:00<?, ?it/s] epoch train_acc train_loss valid_acc valid_loss dur
------- ----------- ------------ ----------- ------------ ------
1 0.5781 0.6714 0.4828 0.6912 0.2205
2 0.6406 0.6701 0.4828 0.6908 0.2371
3 0.5156 0.7125 0.5172 0.6905 0.2441
4 0.5312 0.6878 0.5517 0.6904 0.2436
5 0.5625 0.6829 0.5862 0.6903 0.2384
6 0.6719 0.6372 0.5862 0.6902 0.2254
7 0.6406 0.6667 0.6207 0.6902 0.1944
8 0.5781 0.6599 0.5862 0.6901 0.1937
9 0.5156 0.6776 0.5862 0.6900 0.1959
10 0.5469 0.6801 0.5862 0.6898 0.1958
epoch train_acc train_loss valid_acc valid_loss dur
------- ----------- ------------ ----------- ------------ ------
1 0.5312 0.6910 0.6552 0.6793 0.1955
2 0.5312 0.6948 0.6897 0.6801 0.1944
3 0.6250 0.6834 0.6552 0.6808 0.1947
Stopping since valid_loss has not improved in the last 3 epochs.
BNCI2014-001-CrossSession: 50%|█████ | 1/2 [00:05<00:05, 5.65s/it] epoch train_acc train_loss valid_acc valid_loss dur
------- ----------- ------------ ----------- ------------ ------
1 0.5156 0.6962 0.6207 0.6878 0.1967
2 0.4688 0.6905 0.6207 0.6879 0.1953
3 0.5781 0.6831 0.6552 0.6883 0.1943
Stopping since valid_loss has not improved in the last 3 epochs.
epoch train_acc train_loss valid_acc valid_loss dur
------- ----------- ------------ ----------- ------------ ------
1 0.5625 0.6807 0.5172 0.6902 0.1956
2 0.5312 0.6775 0.6552 0.6909 0.1950
3 0.5938 0.6857 0.5862 0.6915 0.1936
Stopping since valid_loss has not improved in the last 3 epochs.
BNCI2014-001-CrossSession: 100%|██████████| 2/2 [00:09<00:00, 4.80s/it]
BNCI2014-001-CrossSession: 100%|██████████| 2/2 [00:09<00:00, 4.92s/it]
score time samples ... n_sessions dataset pipeline
0 0.588542 2.406188 144.0 ... 2 BNCI2014-001 EEGNet
1 0.552662 0.951012 144.0 ... 2 BNCI2014-001 EEGNet
2 0.455054 0.956856 144.0 ... 2 BNCI2014-001 EEGNet
3 0.459105 0.951726 144.0 ... 2 BNCI2014-001 EEGNet
[4 rows x 9 columns]
Plot Results#
plt.figure()
sns.barplot(data=results, y="score", x="subject", palette="viridis")
plt.show()
References#
Total running time of the script: (0 minutes 14.408 seconds)
Estimated memory usage: 1347 MB
Run this example
