Source code for torcheeg.datasets.module.emotion_recognition.faced
import os
import pickle as pkl
from typing import Any, Callable, Dict, Tuple, Union
from ....utils import get_random_dir_path
from ..base_dataset import BaseDataset
VALENCE_DICT = {
1: -1, # negative
2: -1,
3: -1,
4: -1,
5: -1,
6: -1,
7: -1,
8: -1,
9: -1,
10: -1,
11: -1,
12: -1,
13: 0, # neutral
14: 0,
15: 0,
16: 0,
17: 1, # positive
18: 1,
19: 1,
20: 1,
21: 1,
22: 1,
23: 1,
24: 1,
25: 1,
26: 1,
27: 1,
28: 1
}
EMOTION_DICT = {
1: 0, # anger
2: 0,
3: 0,
4: 1, # disgust
5: 1,
6: 1,
7: 2, # fear
8: 2,
9: 2,
10: 3, # sadness
11: 3,
12: 3,
13: 4, # neutral
14: 4,
15: 4,
16: 4,
17: 5, # amusement
18: 5,
19: 5,
20: 6, # inspiration
21: 6,
22: 6,
23: 7, # joy
24: 7,
25: 7,
26: 8, # tenderness
27: 8,
28: 8
}
[docs]class FACEDDataset(BaseDataset):
r'''
The FACED dataset was provided by the Tsinghua Laboratory of Brain and Intelligence. The Finer-grained Affective Computing EEG Dataset (FACED) recorded 32-channel EEG signals from 123 subjects. During the experiment, subjects watched 28 emotion-elicitation video clips covering nine emotion categories (amusement, inspiration, joy, tenderness; anger, fear, disgust, sadness, and neutral emotion), providing a fine-grained and balanced categorization on both the positive and negative sides of emotion.
This class generates training samples and test samples according to the given parameters, and caches the generated results in a unified input and output format (IO). The relevant information of the dataset is as follows:
- Author: Please refer to the downloaded URL.
- Year: 2023
- Download URL: https://www.synapse.org/#!Synapse:syn50614194/files/
- Reference: Please refer to the downloaded URL.
- Stimulus: video clips.
- Signals: Electroencephalogram (30 channels at 250Hz) and two channels of left/right mastoid signals from 123 subjects.
- Rating: 28 video clips are annotated in valence and discrete emotion dimensions. The valence is divided into positive (1), negative (-1), and neutral (0). Discrete emotions are divided into anger (0), disgust (1), fear (2), sadness (3), neutral (4), amusement (5), inspiration (6), joy (7), and tenderness (8).
In order to use this dataset, the download folder :obj:`Processed_data` (download from this url: https://www.synapse.org/#!Synapse:syn50615881) is required, containing the following files:
.. code-block:: python
Processed_data/
├── sub000.pkl
├── sub001.pkl
├── sub002.pkl
└── ...
An example dataset for CNN-based methods:
.. code-block:: python
from torcheeg.datasets import FACEDDataset
from torcheeg import transforms
from torcheeg.datasets.constants import FACED_CHANNEL_LOCATION_DICT
dataset = FACEDDataset(root_path='./Processed_data',
offline_transform=transforms.Compose([
transforms.BandDifferentialEntropy(),
transforms.ToGrid(FACED_CHANNEL_LOCATION_DICT)
]),
online_transform=transforms.ToTensor(),
label_transform=transforms.Compose([
transforms.Select('emotion'),
transforms.Lambda(lambda x: x + 1)
]))
print(dataset[0])
# EEG signal (torch.Tensor[4, 8, 9]),
# coresponding baseline signal (torch.Tensor[4, 8, 9]),
# label (int)
Another example dataset for CNN-based methods:
.. code-block:: python
from torcheeg.datasets import FACEDDataset
from torcheeg import transforms
dataset = FACEDDataset(root_path='./Processed_data',
online_transform=transforms.Compose(
[transforms.ToTensor(),
transforms.To2d()]),
label_transform=transforms.Compose([
transforms.Select('emotion'),
transforms.Lambda(lambda x: x + 1)
]))
print(dataset[0])
# EEG signal (torch.Tensor[1, 30, 250]),
# coresponding baseline signal (torch.Tensor[1, 30, 250]),
# label (int)
An example dataset for GNN-based methods:
.. code-block:: python
from torcheeg.datasets import FACEDDataset
from torcheeg import transforms
from torcheeg.datasets.constants import FACED_ADJACENCY_MATRIX
from torcheeg.transforms.pyg import ToG
dataset = FACEDDataset(root_path='./Processed_data',
online_transform=transforms.Compose([ToG(FACED_ADJACENCY_MATRIX)]),
label_transform=transforms.Compose(
[transforms.Select('emotion'),
transforms.Lambda(lambda x: x + 1)]))
print(dataset[0])
# EEG signal (torch_geometric.data.Data),
# coresponding baseline signal (torch_geometric.data.Data),
# label (int)
Args:
root_path (str): Downloaded data files in matlab (unzipped Processed_data.zip) formats (default: :obj:`'./Processed_data'`)
chunk_size (int): Number of data points included in each EEG chunk as training or test samples. If set to -1, the EEG signal of a trial is used as a sample of a chunk. (default: :obj:`250`)
overlap (int): The number of overlapping data points between different chunks when dividing EEG chunks. (default: :obj:`0`)
num_channel (int): Number of channels used, of which the first 30 channels are EEG signals. (default: :obj:`30`)
online_transform (Callable, optional): The transformation of the EEG signals and baseline EEG signals. The input is a :obj:`np.ndarray`, and the ouput is used as the first and second value of each element in the dataset. (default: :obj:`None`)
offline_transform (Callable, optional): The usage is the same as :obj:`online_transform`, but executed before generating IO intermediate results. (default: :obj:`None`)
label_transform (Callable, optional): The transformation of the label. The input is an information dictionary, and the ouput is used as the third value of each element in the dataset. (default: :obj:`None`)
before_trial (Callable, optional): The hook performed on the trial to which the sample belongs. It is performed before the offline transformation and thus typically used to implement context-dependent sample transformations, such as moving averages, etc. The input of this hook function is a 2D EEG signal with shape (number of electrodes, number of data points), whose ideal output shape is also (number of electrodes, number of data points).
after_trial (Callable, optional): The hook performed on the trial to which the sample belongs. It is performed after the offline transformation and thus typically used to implement context-dependent sample transformations, such as moving averages, etc. The input and output of this hook function should be a sequence of dictionaries representing a sequence of EEG samples. Each dictionary contains two key-value pairs, indexed by :obj:`eeg` (the EEG signal matrix) and :obj:`key` (the index in the database) respectively.
io_path (str): The path to generated unified data IO, cached as an intermediate result. If set to None, a random path will be generated. (default: :obj:`None`)
io_size (int): Maximum size database may grow to; used to size the memory mapping. If database grows larger than ``map_size``, an exception will be raised and the user must close and reopen. (default: :obj:`1048576`)
io_mode (str): Storage mode of EEG signal. When io_mode is set to :obj:`lmdb`, TorchEEG provides an efficient database (LMDB) for storing EEG signals. LMDB may not perform well on limited operating systems, where a file system based EEG signal storage is also provided. When io_mode is set to :obj:`pickle`, pickle-based persistence files are used. When io_mode is set to :obj:`memory`, memory are used. (default: :obj:`lmdb`)
num_worker (int): Number of subprocesses to use for data loading. 0 means that the data will be loaded in the main process. (default: :obj:`0`)
verbose (bool): Whether to display logs during processing, such as progress bars, etc. (default: :obj:`True`)
'''
def __init__(self,
root_path: str = './Processed_data',
chunk_size: int = 250,
overlap: int = 0,
num_channel: int = 30,
online_transform: Union[None, Callable] = None,
offline_transform: Union[None, Callable] = None,
label_transform: Union[None, Callable] = None,
before_trial: Union[None, Callable] = None,
after_trial: Union[Callable, None] = None,
after_subject: Union[Callable, None] = None,
io_path: Union[None, str] = None,
io_size: int = 1048576,
io_mode: str = 'lmdb',
num_worker: int = 0,
verbose: bool = True):
if io_path is None:
io_path = get_random_dir_path(dir_prefix='datasets')
# pass all arguments to super class
params = {
'root_path': root_path,
'chunk_size': chunk_size,
'overlap': overlap,
'num_channel': num_channel,
'online_transform': online_transform,
'offline_transform': offline_transform,
'label_transform': label_transform,
'before_trial': before_trial,
'after_trial': after_trial,
'after_subject': after_subject,
'io_path': io_path,
'io_size': io_size,
'io_mode': io_mode,
'num_worker': num_worker,
'verbose': verbose
}
super().__init__(**params)
# save all arguments to __dict__
self.__dict__.update(params)
@staticmethod
def process_record(file: Any = None,
root_path: str = './Processed_data',
chunk_size: int = 250,
overlap: int = 0,
num_channel: int = 30,
before_trial: Union[None, Callable] = None,
offline_transform: Union[None, Callable] = None,
**kwargs):
file_name = os.path.basename(
file) # an element from file name list, such as 'sub087.pkl'
subject_id = int(file_name.split('.')[0]
[3:]) # get subject_id from 'sub087.pkl', such as 87
# derive the given arguments (kwargs)
with open(os.path.join(root_path, file_name), 'rb') as f:
samples = pkl.load(
f, encoding='iso-8859-1'
) # 28(trials), 32(channels), 30s*250hz(time points)
write_pointer = 0
for trial_id in range(len(samples)):
trial_samples = samples[
trial_id, :
num_channel] # default 30(remove A1 and A2 from 32 channels), 30s*250hz(time points)
if before_trial:
trial_samples = before_trial(trial_samples)
# record the common meta info
trial_meta_info = {
'subject_id': subject_id,
'trial_id': trial_id,
'valence': VALENCE_DICT[trial_id + 1],
'emotion': EMOTION_DICT[trial_id + 1],
}
start_at = 0
if chunk_size <= 0:
dynamic_chunk_size = trial_samples.shape[1] - start_at
else:
dynamic_chunk_size = chunk_size
# chunk with chunk size
end_at = start_at + dynamic_chunk_size
# calculate moving step
step = dynamic_chunk_size - overlap
while end_at <= trial_samples.shape[1]:
clip_sample = trial_samples[:, start_at:end_at]
t_eeg = clip_sample
if not offline_transform is None:
t_eeg = offline_transform(eeg=clip_sample)['eeg']
clip_id = f'{file_name}_{write_pointer}'
write_pointer += 1
# record meta info for each signal
record_info = {
'start_at': start_at,
'end_at': end_at,
'clip_id': clip_id
}
record_info.update(trial_meta_info)
yield {'eeg': t_eeg, 'key': clip_id, 'info': record_info}
start_at = start_at + step
end_at = start_at + dynamic_chunk_size
def set_records(self, root_path: str = './Processed_data', **kwargs):
assert os.path.exists(
root_path
), f'root_path ({root_path}) does not exist. Please download the dataset and set the root_path to the downloaded path.'
file_path_list = os.listdir(root_path)
file_path_list.sort()
file_path_list = [
file_path for file_path in file_path_list if file_path.endswith('.pkl')
]
return file_path_list
def __getitem__(self, index: int) -> Tuple[any, any, int, int, int]:
info = self.read_info(index)
eeg_index = str(info['clip_id'])
eeg_record = str(info['_record_id'])
eeg = self.read_eeg(eeg_record, eeg_index)
signal = eeg
label = info
if self.online_transform:
signal = self.online_transform(eeg=eeg)['eeg']
if self.label_transform:
label = self.label_transform(y=info)['y']
return signal, label
@property
def repr_body(self) -> Dict:
return dict(
super().repr_body, **{
'root_path': self.root_path,
'chunk_size': self.chunk_size,
'overlap': self.overlap,
'num_channel': self.num_channel,
'online_transform': self.online_transform,
'offline_transform': self.offline_transform,
'label_transform': self.label_transform,
'before_trial': self.before_trial,
'after_trial': self.after_trial,
'num_worker': self.num_worker,
'verbose': self.verbose,
'io_size': self.io_size
})
