Documentation | TorchEEG Examples
TorchEEG is a library built on PyTorch for EEG signal analysis. TorchEEG aims to provide a plug-and-play EEG analysis tool, so that researchers can quickly reproduce EEG analysis work and start new EEG analysis research without paying attention to technical details unrelated to the research focus.
TorchEEG specifies a unified data input-output format (IO) and implement commonly used EEG databases, allowing users to quickly access benchmark datasets and define new custom datasets. The datasets that have been defined so far include emotion recognition and so on. According to papers published in the field of EEG analysis, TorchEEG provides data preprocessing methods commonly used for EEG signals, and provides plug-and-play API for both offline and online pre-proocessing. Offline processing allow users to process once and use any times, speeding up the training process. Online processing allows users to save time when creating new data processing methods. TorchEEG also provides deep learning models following published papers for EEG analysis, including convolutional neural networks, graph convolutional neural networks, and Transformers.
TorchEEG depends on PyTorch, please complete the installation of PyTorch according to the system, CUDA version and other information:
# Conda
# please refer to https://pytorch.org/get-started/locally/
# e.g. CPU version
conda install pytorch==1.11.0 torchvision torchaudio cpuonly -c pytorch
# e.g. GPU version
conda install pytorch==1.11.0 torchvision torchaudio cudatoolkit=11.3 -c pytorch
# Pip
# please refer to https://pytorch.org/get-started/previous-versions/
# e.g. CPU version
pip install torch==1.11.0+cpu torchvision==0.12.0+cpu torchaudio==0.11.0 --extra-index-url https://download.pytorch.org/whl/cpu
# e.g. GPU version
pip install torch==1.11.0+cu102 torchvision==0.12.0+cu102 torchaudio==0.11.0 --extra-index-url https://download.pytorch.org/whl/cu102
Since version v1.0.9, torcheeg supports installing with conda! You can simply install TorchEEG using Anaconda, just run the following command:
conda install -c tczhangzhi -c conda-forge torcheeg
TorchEEG allows pip-based installation, please use the following command:
pip install torcheeg
In case you want to experiment with the latest TorchEEG features which are not fully released yet, please run the following command to install from the main branch on github:
pip install git+https://github.com/tczhangzhi/torcheeg.git
TorchEEG provides plugins related to graph algorithms for converting EEG in datasets into graph structures and analyzing them using graph neural networks. This part of the implementation relies on PyG.
If you do not use graph-related algorithms, you can skip this part of the installation.
# Conda
# please refer to https://pytorch-geometric.readthedocås.io/en/latest/notes/installation.html
conda install pyg -c pyg
# Pip
# please refer to https://pytorch-geometric.readthedocs.io/en/latest/notes/installation.html
# e.g. CPU version
pip install torch-scatter torch-sparse torch-cluster torch-spline-conv torch-geometric -f https://data.pyg.org/whl/torch-1.11.0+cpu.html
# e.g. GPU version
pip install torch-scatter torch-sparse torch-cluster torch-spline-conv torch-geometric -f https://data.pyg.org/whl/torch-1.11.0+cu102.html
At a granular level, PyTorch is a library that consists of the following components:
Component | Description |
---|---|
torcheeg.io | A set of unified input and output API is used to store the processing results of various EEG databases for more efficient and convenient use. |
torcheeg.datasets | The packaged benchmark dataset implementation provides a multi-process preprocessing interface. |
torcheeg.transforms | Extensive EEG preprocessing methods help users extract features, construct EEG signal representations, and connect to commonly used deep learning libraries. |
torcheeg.model_selection | Extensive dataset partitioning methods for users to experiment with different settings. |
torcheeg.models | Extensive baseline method reproduction. |
We list currently supported datasets, transforms, data splitting, and deep learning models by category.
Datasets: All datasets rely on a set of efficient IO APIs, torcheeg.io, to store data preprocessing results on disk and read them quickly during training. Data preprocessing and storage support multiprocessing (speed up!).
- AMIGOS dataset from Miranda-Correa et al.: AMIGOS: A dataset for affect, personality and mood research on individuals and groups.
- DREAMER dataset from Katsigiannis et al.: DREAMER: A database for emotion recognition through EEG and ECG signals from wireless low-cost off-the-shelf devices.
- SEED dataset from Zheng et al.: Investigating critical frequency bands and channels for EEG-based emotion recognition with deep neural networks.
- DEAP dataset from Koelstra et al.: DEAP: A database for emotion analysis; using physiological signals.
Expand to see more...
- MAHNOB dataset from Soleymani et al.: A multimodal database for affect recognition and implicit tagging.
Transforms: TorchEEG provides extensive data transformation tools to help users build EEG data representations suitable for a variety of task formulation and a variety of model structures.
- Feature Engineering: BandDifferentialEntropy, BandPowerSpectralDensity, BandMeanAbsoluteDeviation, BandKurtosis, BandSkewness, Concatenate
- General Operation: PickElectrode, MeanStdNormalize, MinMaxNormalize
- For CNN: To2d, ToGrid, ToInterpolatedGrid
Expand to see more...
- For GNN: ToG
- For Augmentation: Resize, RandomNoise, RandomMask
- For Label Construction: Select, Binary, BinariesToCategory
Data Splitting: In current research in the field of EEG analysis, there are various settings based on different considerations for data partitioning. Please choose a reasonable data division method according to the research focus:
- Subject Dependent: KFoldPerSubjectGroupbyTrial, train_test_split_per_subject_groupby_trial
- Subject Independent: LeaveOneSubjectOut
- Conventional: KFold, train_test_split, KFoldGroupbyTrial, train_test_split_groupby_trial
Models: Coming soon after pushing to align with the official implementation or description. If the current version of CNNs, GNNs and Transformers is to be used, please refer to the implementation in torcheeg.models.
In this quick tour, we highlight the ease of starting an EEG analysis research with only modifying a few lines of PyTorch tutorial.
The torcheeg.datasets
module contains dataset classes for many real-world EEG datasets. In this tutorial, we use the DEAP
dataset. We first go to the official website to apply for data download permission according to the introduction of DEAP dataset, and download the dataset. Next, we need to specify the download location of the dataset in the root_path
parameter. For the DEAP dataset, we specify the path to the data_preprocessed_python
folder, e.g. ./tmp_in/data_preprocessed_python
.
from torcheeg.datasets import DEAPDataset
from torcheeg.datasets.constants.emotion_recognition.deap import DEAP_CHANNEL_LOCATION_DICT
dataset = DEAPDataset(io_path=f'./tmp_out/deap',
root_path='./tmp_in/data_preprocessed_python',
offline_transform=transforms.Compose(
[transforms.BandDifferentialEntropy(),
transforms.ToGrid(DEAP_CHANNEL_LOCATION_DICT)]),
online_transform=transforms.Compose([transforms.BaselineRemoval(),
transforms.ToTensor()]),
label_transform=transforms.Compose([
transforms.Select('valence'),
transforms.Binary(5.0),
]), num_worker=4)
The DEAPDataset
API further contains three parameters: online_transform
, offline_transform
, and label_transform
, which are used to modify samples and labels, respectively.
Here, offline_transform
will only be called once when the dataset is initialized to preprocess all samples in the dataset, and the processed dataset will be stored in io_path
to avoid time-consuming repeated transformations in subsequent use. If offline preprocessing is a computationally intensive operation, we also recommend setting multi-CPU parallelism for offline_transform, e.g., set num_worker
to 4.
online_transform
is used to transform samples on the fly. Please use online_transform
if you don't want to wait for the preprocessing of the entire dataset (suitable for scenarios where new transform
algorithms are designed) or expect data transformation with randomness each time a sample is indexed.
Next, we need to divide the dataset into a training set and a test set. In the field of EEG analysis, commonly used data partitioning methods include k-fold cross-validation and leave-one-out cross-validation. In this tutorial, we use k-fold cross-validation on the entire dataset (KFold
) as an example of dataset splitting.
from torcheeg.model_selection import KFold
k_fold = KFold(n_splits=10,
split_path=f'./tmp_out/split',
shuffle=True,
random_state=42)
Let's define a simple but effective CNN model according to CCNN:
class CNN(torch.nn.Module):
def __init__(self, in_channels=4, num_classes=3):
super().__init__()
self.conv1 = nn.Sequential(
nn.ZeroPad2d((1, 2, 1, 2)),
nn.Conv2d(in_channels, 64, kernel_size=4, stride=1),
nn.ReLU()
)
self.conv2 = nn.Sequential(
nn.ZeroPad2d((1, 2, 1, 2)),
nn.Conv2d(64, 128, kernel_size=4, stride=1),
nn.ReLU()
)
self.conv3 = nn.Sequential(
nn.ZeroPad2d((1, 2, 1, 2)),
nn.Conv2d(128, 256, kernel_size=4, stride=1),
nn.ReLU()
)
self.conv4 = nn.Sequential(
nn.ZeroPad2d((1, 2, 1, 2)),
nn.Conv2d(256, 64, kernel_size=4, stride=1),
nn.ReLU()
)
self.lin1 = nn.Linear(9 * 9 * 64, 1024)
self.lin2 = nn.Linear(1024, num_classes)
def forward(self, x):
x = self.conv1(x)
x = self.conv2(x)
x = self.conv3(x)
x = self.conv4(x)
x = x.flatten(start_dim=1)
x = self.lin1(x)
x = self.lin2(x)
return x
Specify the device and loss function used during training and test.
device = "cuda" if torch.cuda.is_available() else "cpu"
loss_fn = nn.CrossEntropyLoss()
batch_size = 64
The training and validation scripts for the model are taken from the PyTorch tutorial without much modification. Usually, the value of batch
contains two parts; the first part refers to the result of online_transform
, which generally corresponds to the Tensor
sequence representing EEG signals. The second part refers to the result of label_transform
, a sequence of integers representing the label.
def train(dataloader, model, loss_fn, optimizer):
size = len(dataloader.dataset)
model.train()
for batch_idx, batch in enumerate(dataloader):
X = batch[0].to(device)
y = batch[1].to(device)
# Compute prediction error
pred = model(X)
loss = loss_fn(pred, y)
# Backpropagation
optimizer.zero_grad()
loss.backward()
optimizer.step()
if batch_idx % 100 == 0:
loss, current = loss.item(), batch_idx * len(X)
print(f"loss: {loss:>7f} [{current:>5d}/{size:>5d}]")
def valid(dataloader, model, loss_fn):
size = len(dataloader.dataset)
num_batches = len(dataloader)
model.eval()
val_loss, correct = 0, 0
with torch.no_grad():
for batch in dataloader:
X = batch[0].to(device)
y = batch[1].to(device)
pred = model(X)
val_loss += loss_fn(pred, y).item()
correct += (pred.argmax(1) == y).type(torch.float).sum().item()
val_loss /= num_batches
correct /= size
print(f"Test Error: \n Accuracy: {(100*correct):>0.1f}%, Avg loss: {val_loss:>8f} \n")
Traverse k
folds and train the model separately for testing. It is worth noting that, in general, we need to specify shuffle=True
for the DataLoader
of the training data set to avoid the deviation of the model training caused by consecutive labels of the same category.
for i, (train_dataset, val_dataset) in enumerate(k_fold.split(dataset)):
model = CNN().to(device)
optimizer = torch.optim.Adam(model.parameters(), lr=1e-4)
train_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True)
val_loader = DataLoader(val_dataset, batch_size=batch_size, shuffle=False)
epochs = 50
for t in range(epochs):
print(f"Epoch {t+1}\n-------------------------------")
train(train_loader, model, loss_fn, optimizer)
valid(val_loader, model, loss_fn)
print("Done!")
For more specific usage of each module, please refer to the documentation.
TorchEEG is currently in beta; Please let us know if you encounter a bug by filing an issue. We also appreciate all contributions.
If you would like to contribute new datasets, deep learning methods, and extensions to the core, please first open an issue and then send a PR. If you are planning to contribute back bug fixes, please do so without any further discussion.
TorchEEG has a MIT license, as found in the LICENSE file.