def load_train_valid_tuh(n_subjects, n_seconds, ids_to_load):
path = '/home/schirrmr/data/preproced-tuh/all-sensors-32-hz/'
log.info("Load concat dataset...")
dataset = load_concat_dataset(path, preload=False, ids_to_load=ids_to_load)
whole_train_set = dataset.split('session')['train']
n_max_minutes = int(np.ceil(n_seconds / 60) + 2)
sfreq = whole_train_set.datasets[0].raw.info['sfreq']
log.info("Preprocess concat dataset...")
preprocess(whole_train_set, [
MNEPreproc('crop', tmin=0, tmax=n_max_minutes * 60, include_tmax=True),
NumpyPreproc(fn=lambda x: np.clip(x, -80, 80)),
NumpyPreproc(fn=lambda x: x / 3),
NumpyPreproc(fn=exponential_moving_demean,
init_block_size=int(sfreq * 10),
factor_new=1 / (sfreq * 5)),
])
subject_datasets = whole_train_set.split('subject')
n_split = int(np.round(n_subjects * 0.75))
keys = list(subject_datasets.keys())
train_sets = [
d for i in range(n_split) for d in subject_datasets[keys[i]].datasets
]
train_set = BaseConcatDataset(train_sets)
valid_sets = [
d for i in range(n_split, n_subjects)
for d in subject_datasets[keys[i]].datasets
]
valid_set = BaseConcatDataset(valid_sets)
train_set = create_fixed_length_windows(
train_set,
start_offset_samples=60 * 32,
stop_offset_samples=60 * 32 + 32 * n_seconds,
preload=True,
window_size_samples=128,
window_stride_samples=64,
drop_last_window=True,
)
valid_set = create_fixed_length_windows(
valid_set,
start_offset_samples=60 * 32,
stop_offset_samples=60 * 32 + 32 * n_seconds,
preload=True,
window_size_samples=128,
window_stride_samples=64,
drop_last_window=True,
)
return train_set, valid_set
def test_windows_from_events_different_events(tmpdir_factory):
description_expected = 5 * ['T0', 'T1'] + 4 * ['T2', 'T3'] + 2 * ['T1']
raw = _get_raw(tmpdir_factory, description_expected[:10])
base_ds = BaseDataset(raw, description=pd.Series({'file_id': 1}))
raw_1 = _get_raw(tmpdir_factory, description_expected[10:])
base_ds_1 = BaseDataset(raw_1, description=pd.Series({'file_id': 2}))
concat_ds = BaseConcatDataset([base_ds, base_ds_1])
windows = create_windows_from_events(
concat_ds=concat_ds, trial_start_offset_samples=0,
trial_stop_offset_samples=0, window_size_samples=100,
window_stride_samples=100, drop_last_window=False)
description = []
events = []
for ds in windows.datasets:
description += ds.windows.metadata['target'].to_list()
events += ds.windows.events[:, 0].tolist()
assert len(description) == 20
np.testing.assert_array_equal(description,
5 * [0, 1] + 4 * [2, 3] + 2 * [1])
np.testing.assert_array_equal(
np.concatenate(
[raw.time_as_index(raw.annotations.onset, use_rounding=True),
raw_1.time_as_index(raw.annotations.onset, use_rounding=True)]),
events)
def preprocess(concat_ds,
preprocessors,
save_dir=None,
overwrite=False,
n_jobs=None):
"""Apply preprocessors to a concat dataset.
Parameters
----------
concat_ds: BaseConcatDataset
A concat of BaseDataset or WindowsDataset datasets to be preprocessed.
preprocessors: list(Preprocessor)
List of Preprocessor objects to apply to the dataset.
save_dir : str | None
If a string, the preprocessed data will be saved under the specified
directory and the datasets in ``concat_ds`` will be reloaded with
`preload=False`.
overwrite : bool
When `save_dir` is provided, controls whether to delete the old
subdirectories that will be written to under `save_dir`. If False and
the corresponding subdirectories already exist, a ``FileExistsError``
will be raised.
n_jobs : int | None
Number of jobs for parallel execution.
Returns
-------
BaseConcatDataset:
Preprocessed dataset.
"""
# In case of serialization, make sure directory is available before
# preprocessing
if save_dir is not None and not overwrite:
_check_save_dir_empty(save_dir)
if not isinstance(preprocessors, Iterable):
raise ValueError(
'preprocessors must be a list of Preprocessor objects.')
for elem in preprocessors:
assert hasattr(
elem, 'apply'), ('Preprocessor object needs an `apply` method.')
list_of_ds = Parallel(n_jobs=n_jobs)(
delayed(_preprocess)(ds, i, preprocessors, save_dir, overwrite)
for i, ds in enumerate(concat_ds.datasets))
if save_dir is not None: # Reload datasets and replace in concat_ds
concat_ds_reloaded = load_concat_dataset(save_dir,
preload=False,
target_name=None)
_replace_inplace(concat_ds, concat_ds_reloaded)
else:
if n_jobs is None or n_jobs == 1: # joblib did not make copies, the
# preprocessing happened in-place
# Recompute cumulative sizes as transforms might have changed them
concat_ds.cumulative_sizes = concat_ds.cumsum(concat_ds.datasets)
else: # joblib made copies
_replace_inplace(concat_ds, BaseConcatDataset(list_of_ds))
return concat_ds
def load_concat_dataset(path, preload, ids_to_load=None, target_name=None):
"""Load a stored BaseConcatDataset of BaseDatasets or WindowsDatasets from
files.
Parameters
----------
path: str
Path to the directory of the .fif / -epo.fif and .json files.
preload: bool
Whether to preload the data.
ids_to_load: None | list(int)
Ids of specific files to load.
target_name: None or str
Load specific description column as target. If not given, take saved target name.
Returns
-------
concat_dataset: BaseConcatDataset of BaseDatasets or WindowsDatasets
"""
# assume we have a single concat dataset to load
concat_of_raws = os.path.isfile(os.path.join(path, '0-raw.fif'))
assert not (not concat_of_raws and target_name is not None), (
'Setting a new target is only supported for raws.')
concat_of_epochs = os.path.isfile(os.path.join(path, '0-epo.fif'))
paths = [path]
# assume we have multiple concat datasets to load
if not (concat_of_raws or concat_of_epochs):
concat_of_raws = os.path.isfile(os.path.join(path, '0', '0-raw.fif'))
concat_of_epochs = os.path.isfile(os.path.join(path, '0', '0-epo.fif'))
path = os.path.join(path, '*', '')
paths = glob(path)
paths = sorted(paths, key=lambda p: int(p.split(os.sep)[-2]))
if ids_to_load is not None:
paths = [paths[i] for i in ids_to_load]
ids_to_load = None
# if we have neither a single nor multiple datasets, something went wrong
assert concat_of_raws or concat_of_epochs, (
f'Expect either raw or epo to exist in {path} or in '
f'{os.path.join(path, "0")}')
datasets = []
for path in paths:
if concat_of_raws and target_name is None:
target_file_name = os.path.join(path, 'target_name.json')
target_name = json.load(open(target_file_name, "r"))['target_name']
all_signals, description = _load_signals_and_description(
path=path, preload=preload, raws=concat_of_raws,
ids_to_load=ids_to_load
)
for i_signal, signal in enumerate(all_signals):
if concat_of_raws:
datasets.append(
BaseDataset(signal, description.iloc[i_signal],
target_name=target_name))
else:
datasets.append(
WindowsDataset(signal, description.iloc[i_signal])
)
return BaseConcatDataset(datasets)
def lazy_loadable_dataset(tmpdir_factory):
"""Make a dataset of fif files that can be loaded lazily.
"""
raw = _get_raw(tmpdir_factory)
base_ds = BaseDataset(raw, description=pd.Series({'file_id': 1}))
concat_ds = BaseConcatDataset([base_ds])
return concat_ds
def concat_ds_targets():
raws, description = fetch_data_with_moabb(
dataset_name="BNCI2014001", subject_ids=4)
events, _ = mne.events_from_annotations(raws[0])
targets = events[:, -1] - 1
ds = BaseDataset(raws[0], description.iloc[0])
concat_ds = BaseConcatDataset([ds])
return concat_ds, targets
def _preprocess(ds, ds_index, preprocessors, save_dir=None, overwrite=False):
"""Apply preprocessor(s) to Raw or Epochs object.
Parameters
----------
ds: BaseDataset | WindowsDataset
Dataset object to preprocess.
ds_index : int
Index of the BaseDataset in its BaseConcatDataset. Ignored if save_dir
is None.
preprocessors: list(Preprocessor)
List of preprocessors to apply to the dataset.
save_dir : str | None
If provided, save the preprocessed BaseDataset in the
specified directory.
overwrite : bool
If True, overwrite existing file with the same name.
"""
def _preprocess_raw_or_epochs(raw_or_epochs, preprocessors):
for preproc in preprocessors:
preproc.apply(raw_or_epochs)
if hasattr(ds, 'raw'):
_preprocess_raw_or_epochs(ds.raw, preprocessors)
elif hasattr(ds, 'windows'):
_preprocess_raw_or_epochs(ds.windows, preprocessors)
else:
raise ValueError(
'Can only preprocess concatenation of BaseDataset or '
'WindowsDataset, with either a `raw` or `windows` attribute.')
# Store preprocessing keyword arguments in the dataset
_set_preproc_kwargs(ds, preprocessors)
if save_dir is not None:
concat_ds = BaseConcatDataset([ds])
concat_ds.save(save_dir, overwrite=overwrite, offset=ds_index)
else:
return ds
def test_windows_from_events_mapping_filter(tmpdir_factory):
raw = _get_raw(tmpdir_factory, 5 * ['T0', 'T1'])
base_ds = BaseDataset(raw, description=pd.Series({'file_id': 1}))
concat_ds = BaseConcatDataset([base_ds])
windows = create_windows_from_events(
concat_ds=concat_ds, trial_start_offset_samples=0,
trial_stop_offset_samples=0, window_size_samples=100,
window_stride_samples=100, drop_last_window=False, mapping={'T1': 0})
description = windows.datasets[0].windows.metadata['target'].to_list()
assert len(description) == 5
np.testing.assert_array_equal(description, np.zeros(5))
# dataset should contain only 'T1' events
np.testing.assert_array_equal(
(raw.time_as_index(raw.annotations.onset[1::2], use_rounding=True)),
windows.datasets[0].windows.events[:, 0])
def test_windows_from_events_n_jobs(lazy_loadable_dataset):
longer_dataset = BaseConcatDataset([lazy_loadable_dataset.datasets[0]] * 8)
windows = [create_windows_from_events(
concat_ds=longer_dataset, trial_start_offset_samples=0,
trial_stop_offset_samples=0, window_size_samples=100,
window_stride_samples=100, drop_last_window=False, preload=True,
n_jobs=n_jobs) for n_jobs in [1, 2]]
assert windows[0].description.equals(windows[1].description)
for ds1, ds2 in zip(windows[0].datasets, windows[1].datasets):
# assert ds1.windows == ds2.windows # Runs locally, fails in CI
assert np.allclose(ds1.windows.get_data(), ds2.windows.get_data())
assert pd.Series(ds1.windows.info).to_json() == \
pd.Series(ds2.windows.info).to_json()
assert ds1.description.equals(ds2.description)
assert np.array_equal(ds1.y, ds2.y)
assert np.array_equal(ds1.crop_inds, ds2.crop_inds)
def test_fixed_length_windower(start_offset_samples, window_size_samples,
window_stride_samples, drop_last_window,
mapping):
rng = np.random.RandomState(42)
info = mne.create_info(ch_names=['0', '1'], sfreq=50, ch_types='eeg')
data = rng.randn(2, 1000)
raw = mne.io.RawArray(data=data, info=info)
desc = pd.Series({'pathological': True, 'gender': 'M', 'age': 48})
base_ds = BaseDataset(raw, desc, target_name="age")
concat_ds = BaseConcatDataset([base_ds])
if window_size_samples is None:
window_size_samples = base_ds.raw.n_times
stop_offset_samples = data.shape[1] - start_offset_samples
epochs_ds = create_fixed_length_windows(
concat_ds,
start_offset_samples=start_offset_samples,
stop_offset_samples=stop_offset_samples,
window_size_samples=window_size_samples,
window_stride_samples=window_stride_samples,
drop_last_window=drop_last_window,
mapping=mapping)
if mapping is not None:
assert base_ds.target == 48
assert all(epochs_ds.datasets[0].windows.metadata['target'] == 0)
epochs_data = epochs_ds.datasets[0].windows.get_data()
idxs = np.arange(start_offset_samples,
stop_offset_samples - window_size_samples + 1,
window_stride_samples)
if not drop_last_window and idxs[
-1] != stop_offset_samples - window_size_samples:
idxs = np.append(idxs, stop_offset_samples - window_size_samples)
assert len(idxs) == epochs_data.shape[0], (
'Number of epochs different than expected')
assert window_size_samples == epochs_data.shape[2], (
'Window size different than expected')
for j, idx in enumerate(idxs):
np.testing.assert_allclose(base_ds.raw.get_data()[:, idx:idx +
window_size_samples],
epochs_data[j, :],
err_msg=f'Epochs different for epoch {j}')
def dataset_target_time_series():
rng = np.random.RandomState(42)
signal_sfreq = 50
info = mne.create_info(ch_names=['0', '1', 'target_0', 'target_1'],
sfreq=signal_sfreq,
ch_types=['eeg', 'eeg', 'misc', 'misc'])
signal = rng.randn(2, 1000)
targets = np.full((2, 1000), np.nan)
targets_sfreq = 10
targets_stride = int(signal_sfreq / targets_sfreq)
targets[:, ::targets_stride] = rng.randn(2, int(targets.shape[1] / targets_stride))
raw = mne.io.RawArray(np.concatenate([signal, targets]), info=info)
desc = pd.Series({'pathological': True, 'gender': 'M', 'age': 48})
base_dataset = BaseDataset(raw, desc, target_name=None)
concat_ds = BaseConcatDataset([base_dataset])
windows_dataset = create_windows_from_target_channels(
concat_ds,
window_size_samples=100,
)
return concat_ds, windows_dataset, targets, signal
def exp(subject_id):
import torch
test_subj = np.r_[subject_id]
print('test subj:' + str(test_subj))
# train_subj = np.setdiff1d(np.r_[1:10], test_subj)
train_subj = np.setdiff1d(np.r_[1, 3, 7, 8], test_subj)
tr = []
val = []
for ids in train_subj:
train_size = int(0.99 * len(splitted[ids]))
test_size = len(splitted[ids]) - train_size
tr_i, val_i = torch.utils.data.random_split(splitted[ids],
[train_size, test_size])
tr.append(tr_i)
val.append(val_i)
train_set = torch.utils.data.ConcatDataset(tr)
valid_set = torch.utils.data.ConcatDataset(val)
valid_set = BaseConcatDataset([splitted[ids] for ids in test_subj])
######################################################################
# Create model
# ------------
#
######################################################################
# Now we create the deep learning model! Braindecode comes with some
# predefined convolutional neural network architectures for raw
# time-domain EEG. Here, we use the shallow ConvNet model from `Deep
# learning with convolutional neural networks for EEG decoding and
# visualization <https://arxiv.org/abs/1703.05051>`__. These models are
# pure `PyTorch <https://pytorch.org>`__ deep learning models, therefore
# to use your own model, it just has to be a normal PyTorch
# `nn.Module <https://pytorch.org/docs/stable/nn.html#torch.nn.Module>`__.
#
import torch
from braindecode.util import set_random_seeds
from braindecode.models import ShallowFBCSPNet, Deep4Net
cuda = torch.cuda.is_available(
) # check if GPU is available, if True chooses to use it
device = 'cuda:0' if cuda else 'cpu'
if cuda:
torch.backends.cudnn.benchmark = True
seed = 20200220 # random seed to make results reproducible
# Set random seed to be able to reproduce results
set_random_seeds(seed=seed, cuda=cuda)
n_classes = 3
# Extract number of chans and time steps from dataset
n_chans = train_set[0][0].shape[0]
input_window_samples = train_set[0][0].shape[1]
#
# model = ShallowFBCSPNet(
# n_chans,
# n_classes,
# input_window_samples=input_window_samples,
# final_conv_length='auto',
# )
from mynetworks import Deep4Net_origin, ConvClfNet, FcClfNet
model = Deep4Net(
n_chans,
n_classes,
input_window_samples=input_window_samples,
final_conv_length="auto",
)
#
# embedding_net = Deep4Net_origin(4, 22, input_window_samples)
# model = FcClfNet(embedding_net)
# #
print(model)
# Send model to GPU
if cuda:
model.cuda()
######################################################################
# Training
# --------
#
######################################################################
# Now we train the network! EEGClassifier is a Braindecode object
# responsible for managing the training of neural networks. It inherits
# from skorch.NeuralNetClassifier, so the training logic is the same as in
# `Skorch <https://skorch.readthedocs.io/en/stable/>`__.
#
######################################################################
# **Note**: In this tutorial, we use some default parameters that we
# have found to work well for motor decoding, however we strongly
# encourage you to perform your own hyperparameter optimization using
# cross validation on your training data.
#
from skorch.callbacks import LRScheduler
from skorch.helper import predefined_split
from braindecode import EEGClassifier
# # These values we found good for shallow network:
lr = 0.0625 * 0.01
weight_decay = 0
# For deep4 they should be:
# lr = 1 * 0.01
# weight_decay = 0.5 * 0.001
batch_size = 8
n_epochs = 100
clf = EEGClassifier(
model,
criterion=torch.nn.NLLLoss,
optimizer=torch.optim.AdamW,
train_split=predefined_split(
valid_set), # using valid_set for validation
optimizer__lr=lr,
optimizer__weight_decay=weight_decay,
batch_size=batch_size,
callbacks=[
"accuracy",
("lr_scheduler",
LRScheduler('CosineAnnealingLR', T_max=n_epochs - 1)),
],
device=device,
)
# Model training for a specified number of epochs. `y` is None as it is already supplied
# in the dataset.
clf.fit(train_set, y=None, epochs=n_epochs)
######################################################################
# Plot Results
# ------------
#
######################################################################
# Now we use the history stored by Skorch throughout training to plot
# accuracy and loss curves.
#
import matplotlib.pyplot as plt
from matplotlib.lines import Line2D
import pandas as pd
# Extract loss and accuracy values for plotting from history object
results_columns = [
'train_loss', 'valid_loss', 'train_accuracy', 'valid_accuracy'
]
df = pd.DataFrame(clf.history[:, results_columns],
columns=results_columns,
index=clf.history[:, 'epoch'])
# get percent of misclass for better visual comparison to loss
df = df.assign(train_misclass=100 - 100 * df.train_accuracy,
valid_misclass=100 - 100 * df.valid_accuracy)
plt.style.use('seaborn')
fig, ax1 = plt.subplots(figsize=(8, 3))
df.loc[:, ['train_loss', 'valid_loss']].plot(ax=ax1,
style=['-', ':'],
marker='o',
color='tab:blue',
legend=False,
fontsize=14)
ax1.tick_params(axis='y', labelcolor='tab:blue', labelsize=14)
ax1.set_ylabel("Loss", color='tab:blue', fontsize=14)
ax2 = ax1.twinx() # instantiate a second axes that shares the same x-axis
df.loc[:, ['train_misclass', 'valid_misclass']].plot(ax=ax2,
style=['-', ':'],
marker='o',
color='tab:red',
legend=False)
ax2.tick_params(axis='y', labelcolor='tab:red', labelsize=14)
ax2.set_ylabel("Misclassification Rate [%]", color='tab:red', fontsize=14)
ax2.set_ylim(ax2.get_ylim()[0], 85) # make some room for legend
ax1.set_xlabel("Epoch", fontsize=14)
# where some data has already been plotted to ax
handles = []
handles.append(
Line2D([0], [0],
color='black',
linewidth=1,
linestyle='-',
label='Train'))
handles.append(
Line2D([0], [0],
color='black',
linewidth=1,
linestyle=':',
label='Valid'))
plt.legend(handles, [h.get_label() for h in handles], fontsize=14)
plt.tight_layout()
# plt.show()
return df
def get_sleep_stages(windows_dataset):
sleep_stages = [x[1] for x in BaseConcatDataset(windows_dataset.datasets)]
return np.array(sleep_stages)
def exp(subject_id):
cuda = torch.cuda.is_available(
) # check if GPU is available, if True chooses to use it
device = 'cuda:1' if cuda else 'cpu'
if cuda:
torch.backends.cudnn.benchmark = True
seed = 10 # random seed to make results reproducible
# Set random seed to be able to reproduce results
set_random_seeds(seed=seed, cuda=cuda)
test_subj = np.r_[subject_id]
print('test subj:' + str(test_subj))
train_subj = np.setdiff1d(np.r_[1:10], test_subj)
tr = []
val = []
#10%씩 떼어내서 val만듬
for ids in train_subj:
train_size = int(0.9 * len(splitted[ids]))
test_size = len(splitted[ids]) - train_size
tr_i, val_i = torch.utils.data.random_split(splitted[ids],
[train_size, test_size])
tr.append(tr_i)
val.append(val_i)
train_set = torch.utils.data.ConcatDataset(tr)
valid_set = torch.utils.data.ConcatDataset(val)
test_set = BaseConcatDataset([splitted[ids] for ids in test_subj])
# model = Deep4Net(
# n_chans,
# n_classes,
# input_window_samples=input_window_samples,
# final_conv_length="auto",
# )
crop_size = 1125
embedding_net = EEGNet_v2_old(n_classes, n_chans, crop_size)
model = FcClfNet(embedding_net)
print(model)
epochs = 100
batch_size = 64
train_loader = torch.utils.data.DataLoader(train_set,
batch_size=batch_size,
shuffle=True)
valid_loader = torch.utils.data.DataLoader(valid_set,
batch_size=batch_size,
shuffle=False)
test_loader = torch.utils.data.DataLoader(test_set,
batch_size=batch_size,
shuffle=False)
# Send model to GPU
if cuda:
model.cuda(device=device)
from torch.optim import lr_scheduler
import torch.optim as optim
import argparse
parser = argparse.ArgumentParser(
description='cross subject domain adaptation')
parser.add_argument('--batch-size',
type=int,
default=50,
metavar='N',
help='input batch size for training (default: 64)')
parser.add_argument('--test-batch-size',
type=int,
default=50,
metavar='N',
help='input batch size for testing (default: 1000)')
parser.add_argument('--epochs',
type=int,
default=100,
metavar='N',
help='number of epochs to train (default: 10)')
parser.add_argument('--lr',
type=float,
default=0.001,
metavar='LR',
help='learning rate (default: 0.01)')
parser.add_argument('--momentum',
type=float,
default=0.5,
metavar='M',
help='SGD momentum (default: 0.5)')
parser.add_argument('--no-cuda',
action='store_true',
default=False,
help='disables CUDA training')
parser.add_argument(
'--log-interval',
type=int,
default=10,
metavar='N',
help='how many batches to wait before logging training status')
parser.add_argument('--save-model',
action='store_true',
default=True,
help='For Saving the current Model')
args = parser.parse_args()
args.gpuidx = 1
args.seed = 0
args.use_tensorboard = False
args.save_model = False
optimizer = optim.AdamW(model.parameters(),
lr=0.001,
weight_decay=0.5 * 0.001)
# scheduler = lr_scheduler.CosineAnnealingLR(optimizer, T_max=200)
scheduler = lr_scheduler.CosineAnnealingWarmRestarts(optimizer, T_0=50)
#
# #test lr
# lr = []
# for i in range(200):
# scheduler.step()
# lr.append(scheduler.get_lr())
#
# import matplotlib.pyplot as plt
# plt.plot(lr)
import pandas as pd
results_columns = [
'val_loss', 'test_loss', 'val_accuracy', 'test_accuracy'
]
df = pd.DataFrame(columns=results_columns)
for epochidx in range(1, epochs):
print(epochidx)
train(10, model, device, train_loader, optimizer, scheduler, cuda,
device)
val_loss, val_score = eval(model, device, valid_loader)
test_loss, test_score = eval(model, device, test_loader)
results = {
'val_loss': val_loss,
'test_loss': test_loss,
'val_accuracy': val_score,
'test_accuracy': test_score
}
df = df.append(results, ignore_index=True)
print(results)
return df
def create_from_X_y(X,
y,
drop_last_window,
sfreq=None,
ch_names=None,
window_size_samples=None,
window_stride_samples=None):
"""Create a BaseConcatDataset of WindowsDatasets from X and y to be used for
decoding with skorch and braindecode, where X is a list of pre-cut trials
and y are corresponding targets.
Parameters
----------
X: array-like
list of pre-cut trials as n_trials x n_channels x n_times
y: array-like
targets corresponding to the trials
sfreq: common sampling frequency of all trials
ch_names: array-like
channel names of the trials
drop_last_window: bool
whether or not have a last overlapping window, when
windows/windows do not equally divide the continuous signal
window_size_samples: int
window size
window_stride_samples: int
stride between windows
Returns
-------
windows_datasets: BaseConcatDataset
X and y transformed to a dataset format that is compatible with skorch
and braindecode
"""
n_samples_per_x = []
base_datasets = []
if sfreq is None:
sfreq = 100
log.info("No sampling frequency given, set to 100 Hz.")
if ch_names is None:
ch_names = [str(i) for i in range(X.shape[1])]
log.info(f"No channel names given, set to 0-{X.shape[1]}).")
for x, target in zip(X, y):
n_samples_per_x.append(x.shape[1])
info = mne.create_info(ch_names=ch_names, sfreq=sfreq)
raw = mne.io.RawArray(x, info)
base_dataset = BaseDataset(raw,
pd.Series({"target": target}),
target_name="target")
base_datasets.append(base_dataset)
base_datasets = BaseConcatDataset(base_datasets)
if window_size_samples is None and window_stride_samples is None:
if not len(np.unique(n_samples_per_x)) == 1:
raise ValueError(f"if 'window_size_samples' and "
f"'window_stride_samples' are None, "
f"all trials have to have the same length")
window_size_samples = n_samples_per_x[0]
window_stride_samples = n_samples_per_x[0]
windows_datasets = create_fixed_length_windows(
base_datasets,
start_offset_samples=0,
stop_offset_samples=0,
window_size_samples=window_size_samples,
window_stride_samples=window_stride_samples,
drop_last_window=drop_last_window)
return windows_datasets
def exp(subject_id):
test_subj = np.r_[subject_id]
print('test subj:' + str(test_subj))
train_subj = np.setdiff1d(np.r_[1:11], test_subj)
train_set = BaseConcatDataset([splitted[ids] for ids in train_subj])
valid_set = BaseConcatDataset([splitted[ids] for ids in test_subj])
# #
model = ShallowFBCSPNet(
n_chans,
n_classes,
input_window_samples=input_window_samples,
final_conv_length=30,
)
# #
# embedding_net = Deep4Net_origin(4, 22, input_window_samples)
# model = FcClfNet(embedding_net)
print(model)
# Send model to GPU
if cuda:
model.cuda()
from braindecode.models.util import to_dense_prediction_model, get_output_shape
to_dense_prediction_model(model)
######################################################################
# Training
# --------
#
######################################################################
# In difference to trialwise decoding, we now should supply
# ``cropped=True`` to the EEGClassifier, and ``CroppedLoss`` as the
# criterion, as well as ``criterion__loss_function`` as the loss function
# applied to the meaned predictions.
#
######################################################################
# .. note::
# In this tutorial, we use some default parameters that we
# have found to work well for motor decoding, however we strongly
# encourage you to perform your own hyperparameter optimization using
# cross validation on your training data.
#
from skorch.callbacks import LRScheduler
from skorch.helper import predefined_split
from braindecode import EEGClassifier
from braindecode.training.losses import CroppedLoss
from braindecode.training.scoring import trial_preds_from_window_preds
# # These values we found good for shallow network:
lr = 0.0625 * 0.01
weight_decay = 0
# # For deep4 they should be:
# lr = 1 * 0.01
# weight_decay = 0.5 * 0.001
batch_size = 400
n_epochs = 100
clf = EEGClassifier(
model,
cropped=True,
criterion=CroppedLoss,
criterion__loss_function=torch.nn.functional.nll_loss,
optimizer=torch.optim.AdamW,
train_split=predefined_split(valid_set),
optimizer__lr=lr,
optimizer__weight_decay=weight_decay,
iterator_train__shuffle=True,
batch_size=batch_size,
callbacks=[
"accuracy",
("lr_scheduler",
LRScheduler('CosineAnnealingLR', T_max=n_epochs - 1)),
],
device=device,
)
# Model training for a specified number of epochs. `y` is None as it is already supplied
# in the dataset.
clf.fit(train_set, y=None, epochs=n_epochs)
######################################################################
# Plot Results
# ------------
#
######################################################################
# This is again the same code as in trialwise decoding.
#
# .. note::
# Note that we drop further in the classification error and
# loss as in the trialwise decoding tutorial.
#
import matplotlib.pyplot as plt
from matplotlib.lines import Line2D
import pandas as pd
# Extract loss and accuracy values for plotting from history object
results_columns = [
'train_loss', 'valid_loss', 'train_accuracy', 'valid_accuracy'
]
df = pd.DataFrame(clf.history[:, results_columns],
columns=results_columns,
index=clf.history[:, 'epoch'])
# get percent of misclass for better visual comparison to loss
df = df.assign(train_misclass=100 - 100 * df.train_accuracy,
valid_misclass=100 - 100 * df.valid_accuracy)
plt.style.use('seaborn')
fig, ax1 = plt.subplots(figsize=(8, 3))
df.loc[:, ['train_loss', 'valid_loss']].plot(ax=ax1,
style=['-', ':'],
marker='o',
color='tab:blue',
legend=False,
fontsize=14)
ax1.tick_params(axis='y', labelcolor='tab:blue', labelsize=14)
ax1.set_ylabel("Loss", color='tab:blue', fontsize=14)
ax2 = ax1.twinx() # instantiate a second axes that shares the same x-axis
df.loc[:, ['train_misclass', 'valid_misclass']].plot(ax=ax2,
style=['-', ':'],
marker='o',
color='tab:red',
legend=False)
ax2.tick_params(axis='y', labelcolor='tab:red', labelsize=14)
ax2.set_ylabel("Misclassification Rate [%]", color='tab:red', fontsize=14)
ax2.set_ylim(ax2.get_ylim()[0], 85) # make some room for legend
ax1.set_xlabel("Epoch", fontsize=14)
# where some data has already been plotted to ax
handles = []
handles.append(
Line2D([0], [0],
color='black',
linewidth=1,
linestyle='-',
label='Train'))
handles.append(
Line2D([0], [0],
color='black',
linewidth=1,
linestyle=':',
label='Valid'))
plt.legend(handles, [h.get_label() for h in handles], fontsize=14)
plt.tight_layout()
return df
