def create_model_lstm(n_classes, input_time_length, in_chans=22):
set_random_seeds(seed=20170629, cuda=cuda)
# This will determine how many crops are processed in parallel
# final_conv_length determines the size of the receptive field of the ConvNet
lstm_size = 30
lstm_layers = 1
print("#### LSTM SIZE {} ####".format(lstm_size))
print("#### LSTM LAYERS {} ####".format(lstm_layers))
model = ShallowFBCSPLSTM(in_chans=in_chans,
n_classes=n_classes,
input_time_length=input_time_length,
lstm_size=lstm_size,
lstm_layers=lstm_layers,
n_filters_time=lstm_size,
n_filters_spat=lstm_size,
final_conv_length=4,
pool_time_length=20,
pool_time_stride=5).create_network()
to_dense_prediction_model(model)
if cuda:
model.cuda()
return model
def preprocessing(data_folder, subject_id, low_cut_hz):
global train_set, test_set, valid_set, n_classes, n_chans
global n_iters, input_time_length
# def run_exp(data_folder, subject_id, low_cut_hz, model, cuda):
train_filename = 'A{:02d}T.gdf'.format(subject_id)
test_filename = 'A{:02d}E.gdf'.format(subject_id)
train_filepath = os.path.join(data_folder, train_filename)
test_filepath = os.path.join(data_folder, test_filename)
train_label_filepath = train_filepath.replace('.gdf', '.mat')
test_label_filepath = test_filepath.replace('.gdf', '.mat')
train_loader = BCICompetition4Set2A(
train_filepath, labels_filename=train_label_filepath)
test_loader = BCICompetition4Set2A(
test_filepath, labels_filename=test_label_filepath)
train_cnt = train_loader.load()
test_cnt = test_loader.load()
train_cnt = train_cnt.drop_channels(['STI 014', 'EOG-left',
'EOG-central', 'EOG-right'])
assert len(train_cnt.ch_names) == 22
# lets convert to millvolt for numerical stability of next operations
train_cnt = mne_apply(lambda a: a * 1e6, train_cnt)
train_cnt = mne_apply(
lambda a: bandpass_cnt(a, low_cut_hz, 38, train_cnt.info['sfreq'],
filt_order=3,
axis=1), train_cnt)
train_cnt = mne_apply(
lambda a: exponential_running_standardize(a.T, factor_new=1e-3,
init_block_size=1000,
eps=1e-4).T,
train_cnt)
test_cnt = test_cnt.drop_channels(['STI 014', 'EOG-left',
'EOG-central', 'EOG-right'])
assert len(test_cnt.ch_names) == 22
test_cnt = mne_apply(lambda a: a * 1e6, test_cnt)
test_cnt = mne_apply(
lambda a: bandpass_cnt(a, low_cut_hz, 38, test_cnt.info['sfreq'],
filt_order=3,
axis=1), test_cnt)
test_cnt = mne_apply(
lambda a: exponential_running_standardize(a.T, factor_new=1e-3,
init_block_size=1000,
eps=1e-4).T,
test_cnt)
marker_def = OrderedDict([('Left Hand', [1]), ('Right Hand', [2],),
('Foot', [3]), ('Tongue', [4])])
ival = [-500, 4000]
train_set = create_signal_target_from_raw_mne(train_cnt, marker_def, ival)
test_set = create_signal_target_from_raw_mne(test_cnt, marker_def, ival)
train_set, valid_set = split_into_two_sets(train_set,
first_set_fraction=0.8)
set_random_seeds(seed=20190706, cuda=cuda)
n_classes = 4
n_chans = int(train_set.X.shape[1])
input_time_length=1000
def setup_training(self):
"""
Setup training, i.e. set random seeds, transform model to cuda,
initialize monitoring.
"""
# reset remember best extension in case you rerun some experiment
self.rememberer = RememberBest(self.remember_best_column)
self.epochs_df = pd.DataFrame()
set_random_seeds(seed=2382938, cuda=self.cuda)
if self.cuda:
assert th.cuda.is_available(), "Cuda not available"
self.model.cuda()
def setup_training(self):
"""
Setup training, i.e. set random seeds, transform model to cuda,
initialize monitoring.
"""
# reset remember best extension in case you rerun some experiment
if self.do_early_stop:
self.rememberer = RememberBest(self.remember_best_column)
if self.loggers == ('print', ):
self.loggers = [Printer()]
self.epochs_df = pd.DataFrame()
set_random_seeds(seed=self.seed, cuda=self.cuda)
if self.cuda:
assert th.cuda.is_available(), "Cuda not available"
self.model.cuda()
def network_model(model, train_set, test_set, valid_set, n_chans, input_time_length, cuda):
max_epochs = 30
max_increase_epochs = 10
batch_size = 64
init_block_size = 1000
set_random_seeds(seed=20190629, cuda=cuda)
n_classes = 2
n_chans = n_chans
input_time_length = input_time_length
if model == 'deep':
model = Deep4Net(n_chans, n_classes, input_time_length=input_time_length,
final_conv_length='auto').create_network()
elif model == 'shallow':
model = ShallowFBCSPNet(n_chans, n_classes, input_time_length=input_time_length,
final_conv_length='auto').create_network()
if cuda:
model.cuda()
log.info("%s model: ".format(str(model)))
optimizer = AdamW(model.parameters(), lr=0.00625, weight_decay=0)
iterator = BalancedBatchSizeIterator(batch_size=batch_size)
stop_criterion = Or([MaxEpochs(max_epochs),
NoDecrease('valid_misclass', max_increase_epochs)])
monitors = [LossMonitor(), MisclassMonitor(), RuntimeMonitor()]
model_constraint = None
print(train_set.X.shape[0])
model_test = Experiment(model, train_set, valid_set, test_set, iterator=iterator,
loss_function=F.nll_loss, optimizer=optimizer,
model_constraint=model_constraint, monitors=monitors,
stop_criterion=stop_criterion, remember_best_column='valid_misclass',
run_after_early_stop=True, cuda=cuda)
model_test.run()
return model_test
def create_model(n_classes, input_time_length, in_chans=22):
set_random_seeds(seed=20170629, cuda=cuda)
# This will determine how many crops are processed in parallel
# final_conv_length determines the size of the receptive field of the ConvNet
model = ShallowFBCSPNet(in_chans=in_chans,
n_classes=n_classes,
input_time_length=input_time_length,
final_conv_length=4,
pool_time_length=20,
pool_time_stride=5).create_network()
to_dense_prediction_model(model)
if cuda:
model.cuda()
return model
def __init__(self,
n_filters_time=10,
filter_time_length=75,
n_filters_spat=5,
pool_time_length=60,
pool_time_stride=30,
nb_epoch=160):
# random generator
self.rng = RandomState(None)
# init meta info
self.cuda = torch.cuda.is_available()
set_random_seeds(seed=randint(1,20180505), cuda=self.cuda)
# copy all network parameters
self.n_filters_time=n_filters_time
self.filter_time_length=filter_time_length
self.n_filters_spat=n_filters_spat
self.pool_time_length=pool_time_length
self.pool_time_stride=pool_time_stride
self.nb_epoch = nb_epoch
return
def __init__(self,
n_filters_time=10,
filter_time_length=75,
n_filters_spat=5,
pool_time_length=60,
pool_time_stride=30,
nb_epoch=160):
# init meta info
self.cuda = torch.cuda.is_available()
#set_random_seeds(seed=20180505, cuda=self.cuda) # TODO: Fix random seed
set_random_seeds(seed=randint(1, 20180505),
cuda=self.cuda) # TODO: Fix random seed
# copy all network parameters
self.n_filters_time = n_filters_time
self.filter_time_length = filter_time_length
self.n_filters_spat = n_filters_spat
self.pool_time_length = pool_time_length
self.pool_time_stride = pool_time_stride
self.nb_epoch = nb_epoch
return
def build_exp(model_name, cuda, data, batch_size, max_epochs, max_increase_epochs):
log.info("==============================")
log.info("Loading Data...")
log.info("==============================")
train_set = data.train_set
valid_set = data.validation_set
test_set = data.test_set
log.info("==============================")
log.info("Setting Up Model...")
log.info("==============================")
set_random_seeds(seed=20190706, cuda=cuda)
n_classes = 4
n_chans = int(train_set.X.shape[1])
input_time_length = train_set.X.shape[2]
if model_name == "shallow":
model = NewShallowNet(
n_chans, n_classes, input_time_length, final_conv_length="auto"
)
# model = ShallowFBCSPNet(
# n_chans,
# n_classes,
# input_time_length=input_time_length,
# final_conv_length="auto",
# ).create_network()
elif model_name == "deep":
model = NewDeep4Net(n_chans, n_classes, input_time_length, "auto")
# model = Deep4Net(
# n_chans,
# n_classes,
# input_time_length=input_time_length,
# final_conv_length="auto",
# ).create_network()
elif model_name == "eegnet":
# model = EEGNet(n_chans, n_classes,
# input_time_length=input_time_length)
# model = EEGNetv4(n_chans, n_classes,
# input_time_length=input_time_length).create_network()
model = NewEEGNet(n_chans, n_classes, input_time_length=input_time_length)
if cuda:
model.cuda()
log.info("==============================")
log.info("Logging Model Architecture:")
log.info("==============================")
log.info("Model: \n{:s}".format(str(model)))
log.info("==============================")
log.info("Building Experiment:")
log.info("==============================")
optimizer = optim.Adam(model.parameters())
iterator = BalancedBatchSizeIterator(batch_size=batch_size)
stop_criterion = Or(
[MaxEpochs(max_epochs), NoDecrease("valid_misclass", max_increase_epochs)]
)
monitors = [LossMonitor(), MisclassMonitor(), RuntimeMonitor()]
model_constraint = MaxNormDefaultConstraint()
exp = Experiment(
model,
train_set,
valid_set,
test_set,
iterator=iterator,
loss_function=F.nll_loss,
optimizer=optimizer,
model_constraint=model_constraint,
monitors=monitors,
stop_criterion=stop_criterion,
remember_best_column="valid_misclass",
run_after_early_stop=True,
cuda=cuda,
)
return exp
def test_cropped_decoding():
import mne
from mne.io import concatenate_raws
# 5,6,7,10,13,14 are codes for executed and imagined hands/feet
subject_id = 1
event_codes = [5, 6, 9, 10, 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)
# Load each of the files
parts = [mne.io.read_raw_edf(path, preload=True, stim_channel='auto',
verbose='WARNING')
for path in physionet_paths]
# Concatenate them
raw = concatenate_raws(parts)
# Find the events in this dataset
events = mne.find_events(raw, shortest_event=0, stim_channel='STI 014')
# Use only EEG channels
eeg_channel_inds = mne.pick_types(raw.info, meg=False, eeg=True, stim=False,
eog=False,
exclude='bads')
# Extract trials, only using EEG channels
epoched = mne.Epochs(raw, events, dict(hands=2, feet=3), tmin=1, tmax=4.1,
proj=False, picks=eeg_channel_inds,
baseline=None, preload=True)
import numpy as np
from braindecode.datautil.signal_target import SignalAndTarget
# Convert data from volt to millivolt
# Pytorch expects float32 for input and int64 for labels.
X = (epoched.get_data() * 1e6).astype(np.float32)
y = (epoched.events[:, 2] - 2).astype(np.int64) # 2,3 -> 0,1
train_set = SignalAndTarget(X[:60], y=y[:60])
test_set = SignalAndTarget(X[60:], y=y[60:])
from braindecode.models.shallow_fbcsp import ShallowFBCSPNet
from torch import nn
from braindecode.torch_ext.util import set_random_seeds
from braindecode.models.util import to_dense_prediction_model
# Set if you want to use GPU
# You can also use torch.cuda.is_available() to determine if cuda is available on your machine.
cuda = False
set_random_seeds(seed=20170629, cuda=cuda)
# This will determine how many crops are processed in parallel
input_time_length = 450
n_classes = 2
in_chans = train_set.X.shape[1]
# final_conv_length determines the size of the receptive field of the ConvNet
model = ShallowFBCSPNet(in_chans=in_chans, n_classes=n_classes,
input_time_length=input_time_length,
final_conv_length=12).create_network()
to_dense_prediction_model(model)
if cuda:
model.cuda()
from torch import optim
optimizer = optim.Adam(model.parameters())
from braindecode.torch_ext.util import np_to_var
# determine output size
test_input = np_to_var(
np.ones((2, in_chans, input_time_length, 1), dtype=np.float32))
if cuda:
test_input = test_input.cuda()
out = model(test_input)
n_preds_per_input = out.cpu().data.numpy().shape[2]
print("{:d} predictions per input/trial".format(n_preds_per_input))
from braindecode.datautil.iterators import CropsFromTrialsIterator
iterator = CropsFromTrialsIterator(batch_size=32,
input_time_length=input_time_length,
n_preds_per_input=n_preds_per_input)
from braindecode.torch_ext.util import np_to_var, var_to_np
import torch.nn.functional as F
from numpy.random import RandomState
import torch as th
from braindecode.experiments.monitors import compute_preds_per_trial_from_crops
rng = RandomState((2017, 6, 30))
losses = []
accuracies = []
for i_epoch in range(4):
# Set model to training mode
model.train()
for batch_X, batch_y in iterator.get_batches(train_set, shuffle=False):
net_in = np_to_var(batch_X)
if cuda:
net_in = net_in.cuda()
net_target = np_to_var(batch_y)
if cuda:
net_target = net_target.cuda()
# Remove gradients of last backward pass from all parameters
optimizer.zero_grad()
outputs = model(net_in)
# Mean predictions across trial
# Note that this will give identical gradients to computing
# a per-prediction loss (at least for the combination of log softmax activation
# and negative log likelihood loss which we are using here)
outputs = th.mean(outputs, dim=2, keepdim=False)
loss = F.nll_loss(outputs, net_target)
loss.backward()
optimizer.step()
# Print some statistics each epoch
model.eval()
print("Epoch {:d}".format(i_epoch))
for setname, dataset in (('Train', train_set), ('Test', test_set)):
# Collect all predictions and losses
all_preds = []
all_losses = []
batch_sizes = []
for batch_X, batch_y in iterator.get_batches(dataset,
shuffle=False):
net_in = np_to_var(batch_X)
if cuda:
net_in = net_in.cuda()
net_target = np_to_var(batch_y)
if cuda:
net_target = net_target.cuda()
outputs = model(net_in)
all_preds.append(var_to_np(outputs))
outputs = th.mean(outputs, dim=2, keepdim=False)
loss = F.nll_loss(outputs, net_target)
loss = float(var_to_np(loss))
all_losses.append(loss)
batch_sizes.append(len(batch_X))
# Compute mean per-input loss
loss = np.mean(np.array(all_losses) * np.array(batch_sizes) /
np.mean(batch_sizes))
print("{:6s} Loss: {:.5f}".format(setname, loss))
losses.append(loss)
# Assign the predictions to the trials
preds_per_trial = compute_preds_per_trial_from_crops(all_preds,
input_time_length,
dataset.X)
# preds per trial are now trials x classes x timesteps/predictions
# Now mean across timesteps for each trial to get per-trial predictions
meaned_preds_per_trial = np.array(
[np.mean(p, axis=1) for p in preds_per_trial])
predicted_labels = np.argmax(meaned_preds_per_trial, axis=1)
accuracy = np.mean(predicted_labels == dataset.y)
accuracies.append(accuracy * 100)
print("{:6s} Accuracy: {:.1f}%".format(
setname, accuracy * 100))
np.testing.assert_allclose(
np.array(losses),
np.array([1.703004002571106,
1.6295261979103088,
0.71168938279151917,
0.70825588703155518,
0.58231228590011597,
0.60176041722297668,
0.46629951894283295,
0.51184913516044617]),
rtol=1e-4, atol=1e-5)
np.testing.assert_allclose(
np.array(accuracies),
np.array(
[50.0,
46.666666666666664,
60.0,
53.333333333333336,
68.333333333333329,
66.666666666666657,
88.333333333333329,
83.333333333333343]),
rtol=1e-4, atol=1e-5)
def run_exp(epoches, batch_size, subject_num, model_type, cuda, single_subject,
single_subject_num):
# ival = [-500, 4000]
max_increase_epochs = 160
# Preprocessing
X, y = loadSubjects(subject_num, single_subject, single_subject_num)
X = X.astype(np.float32)
y = y.astype(np.int64)
X, y = shuffle(X, y)
trial_length = X.shape[2]
print("trial_length " + str(trial_length))
print("trying to run with {} sec trials ".format((trial_length - 1) / 256))
print("y")
print(y)
trainingSampleSize = int(len(X) * 0.6)
valudationSampleSize = int(len(X) * 0.2)
testSampleSize = int(len(X) * 0.2)
print("INFO : Training sample size: {}".format(trainingSampleSize))
print("INFO : Validation sample size: {}".format(valudationSampleSize))
print("INFO : Test sample size: {}".format(testSampleSize))
train_set = SignalAndTarget(X[:trainingSampleSize],
y=y[:trainingSampleSize])
valid_set = SignalAndTarget(
X[trainingSampleSize:(trainingSampleSize + valudationSampleSize)],
y=y[trainingSampleSize:(trainingSampleSize + valudationSampleSize)])
test_set = SignalAndTarget(X[(trainingSampleSize + valudationSampleSize):],
y=y[(trainingSampleSize +
valudationSampleSize):])
set_random_seeds(seed=20190706, cuda=cuda)
n_classes = 3
n_chans = int(train_set.X.shape[1])
input_time_length = train_set.X.shape[2]
if model_type == 'shallow':
model = ShallowFBCSPNet(n_chans,
n_classes,
input_time_length=input_time_length,
final_conv_length='auto').create_network()
elif model_type == 'deep':
model = Deep4Net(n_chans,
n_classes,
input_time_length=input_time_length,
final_conv_length='auto').create_network()
elif model_type == 'eegnet':
model = EEGNetv4(n_chans,
n_classes,
input_time_length=input_time_length,
final_conv_length='auto').create_network()
if cuda:
model.cuda()
log.info("Model: \n{:s}".format(str(model)))
optimizer = optim.Adam(model.parameters())
iterator = BalancedBatchSizeIterator(batch_size=batch_size)
stop_criterion = Or([
MaxEpochs(max_epochs),
NoDecrease('valid_misclass', max_increase_epochs)
])
monitors = [LossMonitor(), MisclassMonitor(), RuntimeMonitor()]
model_constraint = MaxNormDefaultConstraint()
exp = Experiment(model,
train_set,
valid_set,
test_set,
iterator=iterator,
loss_function=F.nll_loss,
optimizer=optimizer,
model_constraint=model_constraint,
monitors=monitors,
stop_criterion=stop_criterion,
remember_best_column='valid_misclass',
run_after_early_stop=True,
cuda=cuda)
exp.run()
# th.save(model, "models\{}-cropped-singleSubjectNum{}-{}sec-{}epoches-torch_model".format(model_type, single_subject_num, ((trial_length - 1) / 256), epoches))
return exp
def run_exp(data_folder, subject_id, low_cut_hz, model, cuda):
ival = [-500, 4000]
max_epochs = 1600
max_increase_epochs = 160
batch_size = 60
high_cut_hz = 38
factor_new = 1e-3
init_block_size = 1000
valid_set_fraction = 0.2
train_filename = "A{:02d}T.gdf".format(subject_id)
test_filename = "A{:02d}E.gdf".format(subject_id)
train_filepath = os.path.join(data_folder, train_filename)
test_filepath = os.path.join(data_folder, test_filename)
train_label_filepath = train_filepath.replace(".gdf", ".mat")
test_label_filepath = test_filepath.replace(".gdf", ".mat")
train_loader = BCICompetition4Set2A(
train_filepath, labels_filename=train_label_filepath
)
test_loader = BCICompetition4Set2A(
test_filepath, labels_filename=test_label_filepath
)
train_cnt = train_loader.load()
test_cnt = test_loader.load()
# Preprocessing
train_cnt = train_cnt.drop_channels(
["EOG-left", "EOG-central", "EOG-right"]
)
assert len(train_cnt.ch_names) == 22
# lets convert to millvolt for numerical stability of next operations
train_cnt = mne_apply(lambda a: a * 1e6, train_cnt)
train_cnt = mne_apply(
lambda a: bandpass_cnt(
a,
low_cut_hz,
high_cut_hz,
train_cnt.info["sfreq"],
filt_order=3,
axis=1,
),
train_cnt,
)
train_cnt = mne_apply(
lambda a: exponential_running_standardize(
a.T,
factor_new=factor_new,
init_block_size=init_block_size,
eps=1e-4,
).T,
train_cnt,
)
test_cnt = test_cnt.drop_channels(["EOG-left", "EOG-central", "EOG-right"])
assert len(test_cnt.ch_names) == 22
test_cnt = mne_apply(lambda a: a * 1e6, test_cnt)
test_cnt = mne_apply(
lambda a: bandpass_cnt(
a,
low_cut_hz,
high_cut_hz,
test_cnt.info["sfreq"],
filt_order=3,
axis=1,
),
test_cnt,
)
test_cnt = mne_apply(
lambda a: exponential_running_standardize(
a.T,
factor_new=factor_new,
init_block_size=init_block_size,
eps=1e-4,
).T,
test_cnt,
)
marker_def = OrderedDict(
[
("Left Hand", [1]),
("Right Hand", [2]),
("Foot", [3]),
("Tongue", [4]),
]
)
train_set = create_signal_target_from_raw_mne(train_cnt, marker_def, ival)
test_set = create_signal_target_from_raw_mne(test_cnt, marker_def, ival)
train_set, valid_set = split_into_two_sets(
train_set, first_set_fraction=1 - valid_set_fraction
)
set_random_seeds(seed=20190706, cuda=cuda)
n_classes = 4
n_chans = int(train_set.X.shape[1])
input_time_length = train_set.X.shape[2]
if model == "shallow":
model = ShallowFBCSPNet(
n_chans,
n_classes,
input_time_length=input_time_length,
final_conv_length="auto",
).create_network()
elif model == "deep":
model = Deep4Net(
n_chans,
n_classes,
input_time_length=input_time_length,
final_conv_length="auto",
).create_network()
if cuda:
model.cuda()
log.info("Model: \n{:s}".format(str(model)))
optimizer = optim.Adam(model.parameters())
iterator = BalancedBatchSizeIterator(batch_size=batch_size)
stop_criterion = Or(
[
MaxEpochs(max_epochs),
NoDecrease("valid_misclass", max_increase_epochs),
]
)
monitors = [LossMonitor(), MisclassMonitor(), RuntimeMonitor()]
model_constraint = MaxNormDefaultConstraint()
exp = Experiment(
model,
train_set,
valid_set,
test_set,
iterator=iterator,
loss_function=F.nll_loss,
optimizer=optimizer,
model_constraint=model_constraint,
monitors=monitors,
stop_criterion=stop_criterion,
remember_best_column="valid_misclass",
run_after_early_stop=True,
cuda=cuda,
)
exp.run()
return exp
def run_experiment(train_set, valid_set, test_set, model_name, optimizer_name,
init_lr, scheduler_name, use_norm_constraint, weight_decay,
schedule_weight_decay, restarts, max_epochs,
max_increase_epochs, np_th_seed):
set_random_seeds(np_th_seed, cuda=True)
#torch.backends.cudnn.benchmark = True# sometimes crashes?
if valid_set is not None:
assert max_increase_epochs is not None
assert (max_epochs is None) != (restarts is None)
if max_epochs is None:
max_epochs = np.sum(restarts)
n_classes = int(np.max(train_set.y) + 1)
n_chans = int(train_set.X.shape[1])
input_time_length = 1000
if model_name == 'deep':
model = Deep4Net(n_chans,
n_classes,
input_time_length=input_time_length,
final_conv_length=2).create_network()
elif model_name == 'shallow':
model = ShallowFBCSPNet(n_chans,
n_classes,
input_time_length=input_time_length,
final_conv_length=30).create_network()
elif model_name in [
'resnet-he-uniform', 'resnet-he-normal', 'resnet-xavier-normal',
'resnet-xavier-uniform'
]:
init_name = model_name.lstrip('resnet-')
from torch.nn import init
init_fn = {
'he-uniform': lambda w: init.kaiming_uniform(w, a=0),
'he-normal': lambda w: init.kaiming_normal(w, a=0),
'xavier-uniform': lambda w: init.xavier_uniform(w, gain=1),
'xavier-normal': lambda w: init.xavier_normal(w, gain=1)
}[init_name]
model = EEGResNet(in_chans=n_chans,
n_classes=n_classes,
input_time_length=input_time_length,
final_pool_length=10,
n_first_filters=48,
conv_weight_init_fn=init_fn).create_network()
else:
raise ValueError("Unknown model name {:s}".format(model_name))
if 'resnet' not in model_name:
to_dense_prediction_model(model)
model.cuda()
model.eval()
out = model(np_to_var(train_set.X[:1, :, :input_time_length, None]).cuda())
n_preds_per_input = out.cpu().data.numpy().shape[2]
if optimizer_name == 'adam':
optimizer = optim.Adam(model.parameters(),
weight_decay=weight_decay,
lr=init_lr)
elif optimizer_name == 'adamw':
optimizer = AdamW(model.parameters(),
weight_decay=weight_decay,
lr=init_lr)
iterator = CropsFromTrialsIterator(batch_size=60,
input_time_length=input_time_length,
n_preds_per_input=n_preds_per_input,
seed=np_th_seed)
if scheduler_name is not None:
assert schedule_weight_decay == (optimizer_name == 'adamw')
if scheduler_name == 'cosine':
n_updates_per_epoch = sum(
[1 for _ in iterator.get_batches(train_set, shuffle=True)])
if restarts is None:
n_updates_per_period = n_updates_per_epoch * max_epochs
else:
n_updates_per_period = np.array(restarts) * n_updates_per_epoch
scheduler = CosineAnnealing(n_updates_per_period)
optimizer = ScheduledOptimizer(
scheduler,
optimizer,
schedule_weight_decay=schedule_weight_decay)
elif scheduler_name == 'cut_cosine':
# TODO: integrate with if clause before, now just separate
# to avoid messing with code
n_updates_per_epoch = sum(
[1 for _ in iterator.get_batches(train_set, shuffle=True)])
if restarts is None:
n_updates_per_period = n_updates_per_epoch * max_epochs
else:
n_updates_per_period = np.array(restarts) * n_updates_per_epoch
scheduler = CutCosineAnnealing(n_updates_per_period)
optimizer = ScheduledOptimizer(
scheduler,
optimizer,
schedule_weight_decay=schedule_weight_decay)
else:
raise ValueError("Unknown scheduler")
monitors = [
LossMonitor(),
MisclassMonitor(col_suffix='sample_misclass'),
CroppedTrialMisclassMonitor(input_time_length=input_time_length),
RuntimeMonitor()
]
if use_norm_constraint:
model_constraint = MaxNormDefaultConstraint()
else:
model_constraint = None
# change here this cell
loss_function = lambda preds, targets: F.nll_loss(th.mean(preds, dim=2),
targets)
if valid_set is not None:
run_after_early_stop = True
do_early_stop = True
remember_best_column = 'valid_misclass'
stop_criterion = Or([
MaxEpochs(max_epochs),
NoDecrease('valid_misclass', max_increase_epochs)
])
else:
run_after_early_stop = False
do_early_stop = False
remember_best_column = None
stop_criterion = MaxEpochs(max_epochs)
exp = Experiment(model,
train_set,
valid_set,
test_set,
iterator=iterator,
loss_function=loss_function,
optimizer=optimizer,
model_constraint=model_constraint,
monitors=monitors,
stop_criterion=stop_criterion,
remember_best_column=remember_best_column,
run_after_early_stop=run_after_early_stop,
cuda=True,
do_early_stop=do_early_stop)
exp.run()
return exp
def run_exp(
data_folders,
n_recordings,
sensor_types,
n_chans,
max_recording_mins,
sec_to_cut,
duration_recording_mins,
test_recording_mins,
max_abs_val,
sampling_freq,
divisor,
test_on_eval,
n_folds,
i_test_fold,
shuffle,
model_name,
n_start_chans,
n_chan_factor,
input_time_length,
final_conv_length,
model_constraint,
init_lr,
batch_size,
max_epochs,
cuda,
):
import torch.backends.cudnn as cudnn
cudnn.benchmark = True
preproc_functions = []
preproc_functions.append(lambda data, fs: (
data[:, int(sec_to_cut * fs):-int(sec_to_cut * fs)], fs))
preproc_functions.append(lambda data, fs: (data[:, :int(
duration_recording_mins * 60 * fs)], fs))
if max_abs_val is not None:
preproc_functions.append(
lambda data, fs: (np.clip(data, -max_abs_val, max_abs_val), fs))
preproc_functions.append(lambda data, fs: (resampy.resample(
data, fs, sampling_freq, axis=1, filter='kaiser_fast'), sampling_freq))
if divisor is not None:
preproc_functions.append(lambda data, fs: (data / divisor, fs))
dataset = DiagnosisSet(n_recordings=n_recordings,
max_recording_mins=max_recording_mins,
preproc_functions=preproc_functions,
data_folders=data_folders,
train_or_eval='train',
sensor_types=sensor_types)
if test_on_eval:
if test_recording_mins is None:
test_recording_mins = duration_recording_mins
test_preproc_functions = copy(preproc_functions)
test_preproc_functions[1] = lambda data, fs: (data[:, :int(
test_recording_mins * 60 * fs)], fs)
test_dataset = DiagnosisSet(n_recordings=n_recordings,
max_recording_mins=None,
preproc_functions=test_preproc_functions,
data_folders=data_folders,
train_or_eval='eval',
sensor_types=sensor_types)
X, y = dataset.load()
max_shape = np.max([list(x.shape) for x in X], axis=0)
assert max_shape[1] == int(duration_recording_mins * sampling_freq * 60)
if test_on_eval:
test_X, test_y = test_dataset.load()
max_shape = np.max([list(x.shape) for x in test_X], axis=0)
assert max_shape[1] == int(test_recording_mins * sampling_freq * 60)
if not test_on_eval:
splitter = TrainValidTestSplitter(n_folds,
i_test_fold,
shuffle=shuffle)
train_set, valid_set, test_set = splitter.split(X, y)
else:
splitter = TrainValidSplitter(n_folds,
i_valid_fold=i_test_fold,
shuffle=shuffle)
train_set, valid_set = splitter.split(X, y)
test_set = SignalAndTarget(test_X, test_y)
del test_X, test_y
del X, y # shouldn't be necessary, but just to make sure
set_random_seeds(seed=20170629, cuda=cuda)
n_classes = 2
if model_name == 'shallow':
model = ShallowFBCSPNet(
in_chans=n_chans,
n_classes=n_classes,
n_filters_time=n_start_chans,
n_filters_spat=n_start_chans,
input_time_length=input_time_length,
final_conv_length=final_conv_length).create_network()
elif model_name == 'deep':
model = Deep4Net(n_chans,
n_classes,
n_filters_time=n_start_chans,
n_filters_spat=n_start_chans,
input_time_length=input_time_length,
n_filters_2=int(n_start_chans * n_chan_factor),
n_filters_3=int(n_start_chans * (n_chan_factor**2.0)),
n_filters_4=int(n_start_chans * (n_chan_factor**3.0)),
final_conv_length=final_conv_length,
stride_before_pool=True).create_network()
elif (model_name == 'deep_smac'):
if model_name == 'deep_smac':
do_batch_norm = False
else:
assert model_name == 'deep_smac_bnorm'
do_batch_norm = True
double_time_convs = False
drop_prob = 0.244445
filter_length_2 = 12
filter_length_3 = 14
filter_length_4 = 12
filter_time_length = 21
final_conv_length = 1
first_nonlin = elu
first_pool_mode = 'mean'
first_pool_nonlin = identity
later_nonlin = elu
later_pool_mode = 'mean'
later_pool_nonlin = identity
n_filters_factor = 1.679066
n_filters_start = 32
pool_time_length = 1
pool_time_stride = 2
split_first_layer = True
n_chan_factor = n_filters_factor
n_start_chans = n_filters_start
model = Deep4Net(n_chans,
n_classes,
n_filters_time=n_start_chans,
n_filters_spat=n_start_chans,
input_time_length=input_time_length,
n_filters_2=int(n_start_chans * n_chan_factor),
n_filters_3=int(n_start_chans * (n_chan_factor**2.0)),
n_filters_4=int(n_start_chans * (n_chan_factor**3.0)),
final_conv_length=final_conv_length,
batch_norm=do_batch_norm,
double_time_convs=double_time_convs,
drop_prob=drop_prob,
filter_length_2=filter_length_2,
filter_length_3=filter_length_3,
filter_length_4=filter_length_4,
filter_time_length=filter_time_length,
first_nonlin=first_nonlin,
first_pool_mode=first_pool_mode,
first_pool_nonlin=first_pool_nonlin,
later_nonlin=later_nonlin,
later_pool_mode=later_pool_mode,
later_pool_nonlin=later_pool_nonlin,
pool_time_length=pool_time_length,
pool_time_stride=pool_time_stride,
split_first_layer=split_first_layer,
stride_before_pool=True).create_network()
elif model_name == 'shallow_smac':
conv_nonlin = identity
do_batch_norm = True
drop_prob = 0.328794
filter_time_length = 56
final_conv_length = 22
n_filters_spat = 73
n_filters_time = 24
pool_mode = 'max'
pool_nonlin = identity
pool_time_length = 84
pool_time_stride = 3
split_first_layer = True
model = ShallowFBCSPNet(
in_chans=n_chans,
n_classes=n_classes,
n_filters_time=n_filters_time,
n_filters_spat=n_filters_spat,
input_time_length=input_time_length,
final_conv_length=final_conv_length,
conv_nonlin=conv_nonlin,
batch_norm=do_batch_norm,
drop_prob=drop_prob,
filter_time_length=filter_time_length,
pool_mode=pool_mode,
pool_nonlin=pool_nonlin,
pool_time_length=pool_time_length,
pool_time_stride=pool_time_stride,
split_first_layer=split_first_layer,
).create_network()
elif model_name == 'linear':
model = nn.Sequential()
model.add_module("conv_classifier",
nn.Conv2d(n_chans, n_classes, (600, 1)))
model.add_module('softmax', nn.LogSoftmax())
model.add_module('squeeze', Expression(lambda x: x.squeeze(3)))
else:
assert False, "unknown model name {:s}".format(model_name)
to_dense_prediction_model(model)
log.info("Model:\n{:s}".format(str(model)))
if cuda:
model.cuda()
# determine output size
test_input = np_to_var(
np.ones((2, n_chans, input_time_length, 1), dtype=np.float32))
if cuda:
test_input = test_input.cuda()
log.info("In shape: {:s}".format(str(test_input.cpu().data.numpy().shape)))
out = model(test_input)
log.info("Out shape: {:s}".format(str(out.cpu().data.numpy().shape)))
n_preds_per_input = out.cpu().data.numpy().shape[2]
log.info("{:d} predictions per input/trial".format(n_preds_per_input))
iterator = CropsFromTrialsIterator(batch_size=batch_size,
input_time_length=input_time_length,
n_preds_per_input=n_preds_per_input)
optimizer = optim.Adam(model.parameters(), lr=init_lr)
loss_function = lambda preds, targets: F.nll_loss(
th.mean(preds, dim=2, keepdim=False), targets)
if model_constraint is not None:
assert model_constraint == 'defaultnorm'
model_constraint = MaxNormDefaultConstraint()
monitors = [
LossMonitor(),
MisclassMonitor(col_suffix='sample_misclass'),
CroppedDiagnosisMonitor(input_time_length, n_preds_per_input),
RuntimeMonitor(),
]
stop_criterion = MaxEpochs(max_epochs)
batch_modifier = None
run_after_early_stop = True
exp = Experiment(model,
train_set,
valid_set,
test_set,
iterator,
loss_function,
optimizer,
model_constraint,
monitors,
stop_criterion,
remember_best_column='valid_misclass',
run_after_early_stop=run_after_early_stop,
batch_modifier=batch_modifier,
cuda=cuda)
exp.run()
return exp
def train_model(self, train_set, val_set, test_set, save_model):
"""
:param train_set: EEG data (n_trials*n_channels*n_samples)
:param val_set: EEG data (n_trials*n_channels*n_samples)
:param test_set: EEG data (n_trials*n_channels*n_samples) - can be None when training on inner-fold
:param save_model: Boolean: True if trained model is to be saved
:return: Accuracy and loss scores for the model trained with a given set of hyper-parameters
"""
predictions = None
model = None
model = self.call_model()
set_random_seeds(seed=20190629, cuda=self.cuda)
if self.cuda:
model.cuda()
torch.backends.cudnn.deterministic = True
log.info("%s model: ".format(str(model)))
optimizer = optim.Adam(model.parameters(),
lr=self.learning_rate,
weight_decay=0,
eps=1e-8,
amsgrad=False)
stop_criterion = Or([
MaxEpochs(self.epochs),
NoDecrease('valid_misclass', self.max_increase_epochs)
])
model_loss_function = None
#####Setup to run the selected model#####
model_test = Experiment(model,
train_set,
val_set,
test_set=test_set,
iterator=self.iterator,
loss_function=self.loss,
optimizer=optimizer,
model_constraint=self.model_constraint,
monitors=self.monitors,
stop_criterion=stop_criterion,
remember_best_column='valid_misclass',
run_after_early_stop=True,
model_loss_function=model_loss_function,
cuda=self.cuda,
data_type=self.data_type,
model_type=self.model_type,
subject_id=self.subject,
model_number=str(self.model_number),
save_model=save_model)
model_test.run()
model_acc = model_test.epochs_df['valid_misclass'].astype('float')
model_loss = model_test.epochs_df['valid_loss'].astype('float')
current_val_acc = 1 - current_acc(model_acc)
current_val_loss = current_loss(model_loss)
test_accuracy = None
if test_set is not None:
test_accuracy = round(
(1 - model_test.epochs_df['test_misclass'].min()) * 100, 3)
predictions = model_test.predictions
probabilities = model_test.probabilites
return current_val_acc, current_val_loss, test_accuracy, model_test, predictions, probabilities
for CV in np.arange(0, num_folds):
print('Subject No.{} CV {}'.format(i, CV))
# 5th phase: Model evaluation (test)
train_set = SignalAndTarget(X, y=y)
# 80% training, 20% test
train_set, test_set = split_into_train_test(
train_set,
n_folds=num_folds,
i_test_fold=CV,
rng=RandomState((2019, 28, 6))) #RandomState((2019, 28, 6))
# 5% training, 95% test
#test_set, train_set = split_into_train_test(train_set, n_folds = num_folds, i_test_fold = CV, rng=None)
cuda_avail = th.cuda.is_available()
set_random_seeds(seed=20190628, cuda=cuda_avail)
n_classes = 2
in_chans = train_set.X.shape[1] # number of channels = 128
input_time_length = 150 # length of time of each epoch/trial = 4000
model = ShallowFBCSPNet(
split_first_layer=False,
in_chans=in_chans,
n_classes=n_classes,
input_time_length=input_time_length,
final_conv_length='auto',
)
#model = Deep4Net(in_chans=in_chans, n_classes=n_classes,
# input_time_length=input_time_length,
def run_exp(data_folder, subject_id, low_cut_hz, model, cuda):
ival = [-500, 4000]
input_time_length = 1000
max_epochs = 800
max_increase_epochs = 80
batch_size = 60
high_cut_hz = 38
factor_new = 1e-3
init_block_size = 1000
valid_set_fraction = 0.2
train_filename = 'A{:02d}T.gdf'.format(subject_id)
test_filename = 'A{:02d}E.gdf'.format(subject_id)
train_filepath = os.path.join(data_folder, train_filename)
test_filepath = os.path.join(data_folder, test_filename)
train_label_filepath = train_filepath.replace('.gdf', '.mat')
test_label_filepath = test_filepath.replace('.gdf', '.mat')
train_loader = BCICompetition4Set2A(train_filepath,
labels_filename=train_label_filepath)
test_loader = BCICompetition4Set2A(test_filepath,
labels_filename=test_label_filepath)
train_cnt = train_loader.load()
test_cnt = test_loader.load()
# Preprocessing
train_cnt = train_cnt.drop_channels(
['STI 014', 'EOG-left', 'EOG-central', 'EOG-right'])
assert len(train_cnt.ch_names) == 22
# lets convert to millvolt for numerical stability of next operations
train_cnt = mne_apply(lambda a: a * 1e6, train_cnt)
train_cnt = mne_apply(
lambda a: bandpass_cnt(a,
low_cut_hz,
high_cut_hz,
train_cnt.info['sfreq'],
filt_order=3,
axis=1), train_cnt)
train_cnt = mne_apply(
lambda a: exponential_running_standardize(a.T,
factor_new=factor_new,
init_block_size=
init_block_size,
eps=1e-4).T, train_cnt)
test_cnt = test_cnt.drop_channels(
['STI 014', 'EOG-left', 'EOG-central', 'EOG-right'])
assert len(test_cnt.ch_names) == 22
test_cnt = mne_apply(lambda a: a * 1e6, test_cnt)
test_cnt = mne_apply(
lambda a: bandpass_cnt(a,
low_cut_hz,
high_cut_hz,
test_cnt.info['sfreq'],
filt_order=3,
axis=1), test_cnt)
test_cnt = mne_apply(
lambda a: exponential_running_standardize(a.T,
factor_new=factor_new,
init_block_size=
init_block_size,
eps=1e-4).T, test_cnt)
marker_def = OrderedDict([('Left Hand', [1]), (
'Right Hand',
[2],
), ('Foot', [3]), ('Tongue', [4])])
train_set = create_signal_target_from_raw_mne(train_cnt, marker_def, ival)
test_set = create_signal_target_from_raw_mne(test_cnt, marker_def, ival)
train_set, valid_set = split_into_two_sets(train_set,
first_set_fraction=1 -
valid_set_fraction)
set_random_seeds(seed=20190706, cuda=cuda)
n_classes = 4
n_chans = int(train_set.X.shape[1])
if model == 'shallow':
model = ShallowFBCSPNet(n_chans,
n_classes,
input_time_length=input_time_length,
final_conv_length=30).create_network()
elif model == 'deep':
model = Deep4Net(n_chans,
n_classes,
input_time_length=input_time_length,
final_conv_length=2).create_network()
to_dense_prediction_model(model)
if cuda:
model.cuda()
log.info("Model: \n{:s}".format(str(model)))
dummy_input = np_to_var(train_set.X[:1, :, :, None])
if cuda:
dummy_input = dummy_input.cuda()
out = model(dummy_input)
n_preds_per_input = out.cpu().data.numpy().shape[2]
optimizer = optim.Adam(model.parameters())
iterator = CropsFromTrialsIterator(batch_size=batch_size,
input_time_length=input_time_length,
n_preds_per_input=n_preds_per_input)
stop_criterion = Or([
MaxEpochs(max_epochs),
NoDecrease('valid_misclass', max_increase_epochs)
])
monitors = [
LossMonitor(),
MisclassMonitor(col_suffix='sample_misclass'),
CroppedTrialMisclassMonitor(input_time_length=input_time_length),
RuntimeMonitor()
]
model_constraint = MaxNormDefaultConstraint()
loss_function = lambda preds, targets: F.nll_loss(
th.mean(preds, dim=2, keepdim=False), targets)
exp = Experiment(model,
train_set,
valid_set,
test_set,
iterator=iterator,
loss_function=loss_function,
optimizer=optimizer,
model_constraint=model_constraint,
monitors=monitors,
stop_criterion=stop_criterion,
remember_best_column='valid_misclass',
run_after_early_stop=True,
cuda=cuda)
exp.run()
return exp
def run_exp_on_high_gamma_dataset(train_filename, test_filename, low_cut_hz,
model_name, max_epochs, max_increase_epochs,
np_th_seed, debug):
train_set, valid_set, test_set = load_train_valid_test(
train_filename=train_filename,
test_filename=test_filename,
low_cut_hz=low_cut_hz,
debug=debug)
if debug:
max_epochs = 4
set_random_seeds(np_th_seed, cuda=True)
#torch.backends.cudnn.benchmark = True# sometimes crashes?
n_classes = int(np.max(train_set.y) + 1)
n_chans = int(train_set.X.shape[1])
input_time_length = 1000
if model_name == 'deep':
model = Deep4Net(n_chans,
n_classes,
input_time_length=input_time_length,
final_conv_length=2).create_network()
elif model_name == 'shallow':
model = ShallowFBCSPNet(n_chans,
n_classes,
input_time_length=input_time_length,
final_conv_length=30).create_network()
to_dense_prediction_model(model)
model.cuda()
model.eval()
out = model(np_to_var(train_set.X[:1, :, :input_time_length, None]).cuda())
n_preds_per_input = out.cpu().data.numpy().shape[2]
optimizer = optim.Adam(model.parameters(), weight_decay=0, lr=1e-3)
iterator = CropsFromTrialsIterator(batch_size=60,
input_time_length=input_time_length,
n_preds_per_input=n_preds_per_input,
seed=np_th_seed)
monitors = [
LossMonitor(),
MisclassMonitor(col_suffix='sample_misclass'),
CroppedTrialMisclassMonitor(input_time_length=input_time_length),
RuntimeMonitor()
]
model_constraint = MaxNormDefaultConstraint()
loss_function = lambda preds, targets: F.nll_loss(th.mean(preds, dim=2),
targets)
run_after_early_stop = True
do_early_stop = True
remember_best_column = 'valid_misclass'
stop_criterion = Or([
MaxEpochs(max_epochs),
NoDecrease('valid_misclass', max_increase_epochs)
])
exp = Experiment(model,
train_set,
valid_set,
test_set,
iterator=iterator,
loss_function=loss_function,
optimizer=optimizer,
model_constraint=model_constraint,
monitors=monitors,
stop_criterion=stop_criterion,
remember_best_column=remember_best_column,
run_after_early_stop=run_after_early_stop,
cuda=True,
do_early_stop=do_early_stop)
exp.run()
return exp
def test_trialwise_decoding():
import mne
from mne.io import concatenate_raws
# 5,6,7,10,13,14 are codes for executed and imagined hands/feet
subject_id = 1
event_codes = [5, 6, 9, 10, 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)
# Load each of the files
parts = [
mne.io.read_raw_edf(path,
preload=True,
stim_channel='auto',
verbose='WARNING') for path in physionet_paths
]
# Concatenate them
raw = concatenate_raws(parts)
# Find the events in this dataset
events = mne.find_events(raw, shortest_event=0, stim_channel='STI 014')
# Use only EEG channels
eeg_channel_inds = mne.pick_types(raw.info,
meg=False,
eeg=True,
stim=False,
eog=False,
exclude='bads')
# Extract trials, only using EEG channels
epoched = mne.Epochs(raw,
events,
dict(hands=2, feet=3),
tmin=1,
tmax=4.1,
proj=False,
picks=eeg_channel_inds,
baseline=None,
preload=True)
import numpy as np
# Convert data from volt to millivolt
# Pytorch expects float32 for input and int64 for labels.
X = (epoched.get_data() * 1e6).astype(np.float32)
y = (epoched.events[:, 2] - 2).astype(np.int64) # 2,3 -> 0,1
from braindecode.datautil.signal_target import SignalAndTarget
train_set = SignalAndTarget(X[:60], y=y[:60])
test_set = SignalAndTarget(X[60:], y=y[60:])
from braindecode.models.shallow_fbcsp import ShallowFBCSPNet
from torch import nn
from braindecode.torch_ext.util import set_random_seeds
# Set if you want to use GPU
# You can also use torch.cuda.is_available() to determine if cuda is available on your machine.
cuda = False
set_random_seeds(seed=20170629, cuda=cuda)
n_classes = 2
in_chans = train_set.X.shape[1]
# final_conv_length = auto ensures we only get a single output in the time dimension
model = ShallowFBCSPNet(in_chans=in_chans,
n_classes=n_classes,
input_time_length=train_set.X.shape[2],
final_conv_length='auto').create_network()
if cuda:
model.cuda()
from torch import optim
optimizer = optim.Adam(model.parameters())
from braindecode.torch_ext.util import np_to_var, var_to_np
from braindecode.datautil.iterators import get_balanced_batches
import torch.nn.functional as F
from numpy.random import RandomState
rng = RandomState((2017, 6, 30))
losses = []
accuracies = []
for i_epoch in range(6):
i_trials_in_batch = get_balanced_batches(len(train_set.X),
rng,
shuffle=True,
batch_size=30)
# Set model to training mode
model.train()
for i_trials in i_trials_in_batch:
# Have to add empty fourth dimension to X
batch_X = train_set.X[i_trials][:, :, :, None]
batch_y = train_set.y[i_trials]
net_in = np_to_var(batch_X)
if cuda:
net_in = net_in.cuda()
net_target = np_to_var(batch_y)
if cuda:
net_target = net_target.cuda()
# Remove gradients of last backward pass from all parameters
optimizer.zero_grad()
# Compute outputs of the network
outputs = model(net_in)
# Compute the loss
loss = F.nll_loss(outputs, net_target)
# Do the backpropagation
loss.backward()
# Update parameters with the optimizer
optimizer.step()
# Print some statistics each epoch
model.eval()
print("Epoch {:d}".format(i_epoch))
for setname, dataset in (('Train', train_set), ('Test', test_set)):
# Here, we will use the entire dataset at once, which is still possible
# for such smaller datasets. Otherwise we would have to use batches.
net_in = np_to_var(dataset.X[:, :, :, None])
if cuda:
net_in = net_in.cuda()
net_target = np_to_var(dataset.y)
if cuda:
net_target = net_target.cuda()
outputs = model(net_in)
loss = F.nll_loss(outputs, net_target)
losses.append(float(var_to_np(loss)))
print("{:6s} Loss: {:.5f}".format(setname, float(var_to_np(loss))))
predicted_labels = np.argmax(var_to_np(outputs), axis=1)
accuracy = np.mean(dataset.y == predicted_labels)
accuracies.append(accuracy * 100)
print("{:6s} Accuracy: {:.1f}%".format(setname, accuracy * 100))
np.testing.assert_allclose(np.array(losses),
np.array([
1.1775966882705688, 1.2602351903915405,
0.7068756818771362, 0.9367912411689758,
0.394258975982666, 0.6598362326622009,
0.3359280526638031, 0.656258761882782,
0.2790488004684448, 0.6104397177696228,
0.27319177985191345, 0.5949864983558655
]),
rtol=1e-4,
atol=1e-5)
np.testing.assert_allclose(np.array(accuracies),
np.array([
51.666666666666671, 53.333333333333336,
63.333333333333329, 56.666666666666664,
86.666666666666671, 66.666666666666657,
90.0, 63.333333333333329,
96.666666666666671, 56.666666666666664,
96.666666666666671, 66.666666666666657
]),
rtol=1e-4,
atol=1e-5)
def test_experiment_class():
import mne
from mne.io import concatenate_raws
# 5,6,7,10,13,14 are codes for executed and imagined hands/feet
subject_id = 1
event_codes = [5, 6, 9, 10, 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)
# Load each of the files
parts = [mne.io.read_raw_edf(path, preload=True, stim_channel='auto',
verbose='WARNING')
for path in physionet_paths]
# Concatenate them
raw = concatenate_raws(parts)
# Find the events in this dataset
events, _ = mne.events_from_annotations(raw)
# Use only EEG channels
eeg_channel_inds = mne.pick_types(raw.info, meg=False, eeg=True, stim=False,
eog=False,
exclude='bads')
# Extract trials, only using EEG channels
epoched = mne.Epochs(raw, events, dict(hands=2, feet=3), tmin=1, tmax=4.1,
proj=False, picks=eeg_channel_inds,
baseline=None, preload=True)
import numpy as np
from braindecode.datautil.signal_target import SignalAndTarget
from braindecode.datautil.splitters import split_into_two_sets
# Convert data from volt to millivolt
# Pytorch expects float32 for input and int64 for labels.
X = (epoched.get_data() * 1e6).astype(np.float32)
y = (epoched.events[:, 2] - 2).astype(np.int64) # 2,3 -> 0,1
train_set = SignalAndTarget(X[:60], y=y[:60])
test_set = SignalAndTarget(X[60:], y=y[60:])
train_set, valid_set = split_into_two_sets(train_set,
first_set_fraction=0.8)
from braindecode.models.shallow_fbcsp import ShallowFBCSPNet
from torch import nn
from braindecode.torch_ext.util import set_random_seeds
from braindecode.models.util import to_dense_prediction_model
# Set if you want to use GPU
# You can also use torch.cuda.is_available() to determine if cuda is available on your machine.
cuda = False
set_random_seeds(seed=20170629, cuda=cuda)
# This will determine how many crops are processed in parallel
input_time_length = 450
n_classes = 2
in_chans = train_set.X.shape[1]
# final_conv_length determines the size of the receptive field of the ConvNet
model = ShallowFBCSPNet(in_chans=in_chans, n_classes=n_classes,
input_time_length=input_time_length,
final_conv_length=12).create_network()
to_dense_prediction_model(model)
if cuda:
model.cuda()
from torch import optim
optimizer = optim.Adam(model.parameters())
from braindecode.torch_ext.util import np_to_var
# determine output size
test_input = np_to_var(
np.ones((2, in_chans, input_time_length, 1), dtype=np.float32))
if cuda:
test_input = test_input.cuda()
out = model(test_input)
n_preds_per_input = out.cpu().data.numpy().shape[2]
print("{:d} predictions per input/trial".format(n_preds_per_input))
from braindecode.experiments.experiment import Experiment
from braindecode.datautil.iterators import CropsFromTrialsIterator
from braindecode.experiments.monitors import RuntimeMonitor, LossMonitor, \
CroppedTrialMisclassMonitor, MisclassMonitor
from braindecode.experiments.stopcriteria import MaxEpochs
import torch.nn.functional as F
import torch as th
from braindecode.torch_ext.modules import Expression
# Iterator is used to iterate over datasets both for training
# and evaluation
iterator = CropsFromTrialsIterator(batch_size=32,
input_time_length=input_time_length,
n_preds_per_input=n_preds_per_input)
# Loss function takes predictions as they come out of the network and the targets
# and returns a loss
loss_function = lambda preds, targets: F.nll_loss(
th.mean(preds, dim=2, keepdim=False), targets)
# Could be used to apply some constraint on the models, then should be object
# with apply method that accepts a module
model_constraint = None
# Monitors log the training progress
monitors = [LossMonitor(), MisclassMonitor(col_suffix='sample_misclass'),
CroppedTrialMisclassMonitor(input_time_length),
RuntimeMonitor(), ]
# Stop criterion determines when the first stop happens
stop_criterion = MaxEpochs(4)
exp = Experiment(model, train_set, valid_set, test_set, iterator,
loss_function, optimizer, model_constraint,
monitors, stop_criterion,
remember_best_column='valid_misclass',
run_after_early_stop=True, batch_modifier=None, cuda=cuda)
# need to setup python logging before to be able to see anything
import logging
import sys
logging.basicConfig(format='%(asctime)s %(levelname)s : %(message)s',
level=logging.DEBUG, stream=sys.stdout)
exp.run()
import pandas as pd
from io import StringIO
compare_df = pd.read_csv(StringIO(
'train_loss,valid_loss,test_loss,train_sample_misclass,valid_sample_misclass,'
'test_sample_misclass,train_misclass,valid_misclass,test_misclass\n'
'14.167170524597168,13.910758018493652,15.945781707763672,0.5,0.5,'
'0.5333333333333333,0.5,0.5,0.5333333333333333\n'
'1.1735659837722778,1.4342904090881348,1.8664429187774658,0.4629567736185384,'
'0.5120320855614973,0.5336007130124778,0.5,0.5,0.5333333333333333\n'
'1.3168460130691528,1.60431969165802,1.9181344509124756,0.49298128342245995,'
'0.5109180035650625,0.531729055258467,0.5,0.5,0.5333333333333333\n'
'0.8465543389320374,1.280307412147522,1.439755916595459,0.4413435828877005,'
'0.5461229946524064,0.5283422459893048,0.47916666666666663,0.5,'
'0.5333333333333333\n0.6977059841156006,1.1762590408325195,1.2779350280761719,'
'0.40290775401069523,0.588903743315508,0.5307486631016043,0.5,0.5,0.5\n'
'0.7934166193008423,1.1762590408325195,1.2779350280761719,0.4401069518716577,'
'0.588903743315508,0.5307486631016043,0.5,0.5,0.5\n0.5982189178466797,'
'0.8581563830375671,0.9598925113677979,0.32032085561497325,0.47660427807486627,'
'0.4672905525846702,0.31666666666666665,0.5,0.4666666666666667\n0.5044312477111816,'
'0.7133197784423828,0.8164243102073669,0.2591354723707665,0.45699643493761144,'
'0.4393048128342246,0.16666666666666663,0.41666666666666663,0.43333333333333335\n'
'0.4815250039100647,0.6736412644386292,0.8016976714134216,0.23413547237076648,'
'0.39505347593582885,0.42932263814616756,0.15000000000000002,0.41666666666666663,0.5\n'))
for col in compare_df:
np.testing.assert_allclose(np.array(compare_df[col]),
exp.epochs_df[col],
rtol=1e-3, atol=1e-4)
def run_exp(max_recording_mins, n_recordings, sec_to_cut,
duration_recording_mins, max_abs_val, shrink_val, sampling_freq,
divisor, n_folds, i_test_fold, final_conv_length, model_constraint,
batch_size, max_epochs, n_filters_time, n_filters_spat,
filter_time_length, conv_nonlin, pool_time_length,
pool_time_stride, pool_mode, pool_nonlin, split_first_layer,
do_batch_norm, drop_prob, time_cut_off_sec, start_time,
input_time_length, only_return_exp):
kwargs = locals()
for model_param in [
'final_conv_length',
'n_filters_time',
'n_filters_spat',
'filter_time_length',
'conv_nonlin',
'pool_time_length',
'pool_time_stride',
'pool_mode',
'pool_nonlin',
'split_first_layer',
'do_batch_norm',
'drop_prob',
]:
kwargs.pop(model_param)
nonlin_dict = {
'elu': elu,
'relu': relu,
'relu6': relu6,
'tanh': tanh,
'square': square,
'identity': identity,
'log': safe_log,
}
assert input_time_length == 6000
# copy over from early seizure
# make proper
n_classes = 2
in_chans = 21
cuda = True
set_random_seeds(seed=20170629, cuda=cuda)
model = ShallowFBCSPNet(in_chans=in_chans,
n_classes=n_classes,
input_time_length=input_time_length,
final_conv_length=final_conv_length,
n_filters_time=n_filters_time,
filter_time_length=filter_time_length,
n_filters_spat=n_filters_spat,
pool_time_length=pool_time_length,
pool_time_stride=pool_time_stride,
conv_nonlin=nonlin_dict[conv_nonlin],
pool_mode=pool_mode,
pool_nonlin=nonlin_dict[pool_nonlin],
split_first_layer=split_first_layer,
batch_norm=do_batch_norm,
batch_norm_alpha=0.1,
drop_prob=drop_prob).create_network()
to_dense_prediction_model(model)
if cuda:
model.cuda()
model.eval()
test_input = np_to_var(
np.ones((2, in_chans, input_time_length, 1), dtype=np.float32))
if cuda:
test_input = test_input.cuda()
try:
out = model(test_input)
except RuntimeError:
raise ValueError("Model receptive field too large...")
n_preds_per_input = out.cpu().data.numpy().shape[2]
n_receptive_field = input_time_length - n_preds_per_input
if n_receptive_field > 6000:
raise ValueError("Model receptive field ({:d}) too large...".format(
n_receptive_field))
# For future, here optionally add input time length instead
model = ShallowFBCSPNet(in_chans=in_chans,
n_classes=n_classes,
input_time_length=input_time_length,
final_conv_length=final_conv_length,
n_filters_time=n_filters_time,
filter_time_length=filter_time_length,
n_filters_spat=n_filters_spat,
pool_time_length=pool_time_length,
pool_time_stride=pool_time_stride,
conv_nonlin=nonlin_dict[conv_nonlin],
pool_mode=pool_mode,
pool_nonlin=nonlin_dict[pool_nonlin],
split_first_layer=split_first_layer,
batch_norm=do_batch_norm,
batch_norm_alpha=0.1,
drop_prob=drop_prob).create_network()
return common.run_exp(model=model, **kwargs)
def test_trialwise_decoding():
# 5,6,7,10,13,14 are codes for executed and imagined hands/feet
subject_id = 1
event_codes = [5, 6, 9, 10, 13, 14]
# event_codes = [6]
# 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)
# Load each of the files
parts = [
mne.io.read_raw_edf(path,
preload=True,
stim_channel='auto',
verbose='WARNING') for path in physionet_paths
]
# Concatenate them
raw = concatenate_raws(parts)
# Find the events in this dataset
# events = mne.find_events(raw, shortest_event=0, stim_channel='STI 014')
events, _ = mne.events_from_annotations(raw)
# Extract trials, only using EEG channels
eeg_channel_inds = mne.pick_types(raw.info,
meg=False,
eeg=True,
stim=False,
eog=False,
exclude='bads')
# Extract trials, only using EEG channels
epoched = mne.Epochs(raw,
events,
dict(hands=2, feet=3),
tmin=1,
tmax=4.1,
proj=False,
picks=eeg_channel_inds,
baseline=None,
preload=True)
# Convert data from volt to millivolt
# Pytorch expects float32 for input and int64 for labels.
# X:[90,64,497]
X = (epoched.get_data() * 1e6).astype(np.float32)
# y:[90]
y = (epoched.events[:, 2] - 2).astype(np.int64) # 2,3 -> 0,1
# X_train:[60,64,497], y_train:[60]
train_set = SignalAndTarget(X[:60], y=y[:60])
# X_test:[30,64,497], y_test:[30]
test_set = SignalAndTarget(X[60:], y=y[60:])
# Set if you want to use GPU
# You can also use torch.cuda.is_available() to determine if cuda is available on your machine.
cuda = False
set_random_seeds(seed=20170629, cuda=cuda)
n_classes = 2
in_chans = train_set.X.shape[1]
# final_conv_length = auto ensures we only get a single output in the time dimension
# def __init__(self, in_chans=64, n_classes=2, input_time_length=497, n_filters_time=40, filter_time_length=25, n_filters_spat=40, pool_time_length=75, pool_time_stride=15, final_conv_length='auto, conv_nonlin=square, pool_mode="mean", pool_nonlin=safe_log, split_first_layer=True, batch_norm=True, batch_norm_alpha=0.1, drop_prob=0.5, ):
# 感觉create_network()就是__init__的一部分, 现在改成用self.model调用了, 还是感觉不优雅, 主要是forward集成在nn.Sequential里面了
# 然后这个model的实际__init__不是ShallowFBCSPNet, 而是nn.Sequential, 感觉我更喜欢原来的定义方式, 这种方式看不到中间输出
# model = ShallowFBCSPNet(in_chans=in_chans, n_classes=n_classes, input_time_length=train_set.X.shape[2], final_conv_length='auto').create_network() #原来的
model = ShallowFBCSPNet(in_chans=in_chans,
n_classes=n_classes,
input_time_length=train_set.X.shape[2],
final_conv_length='auto').model
if cuda:
model.cuda()
optimizer = optim.Adam(model.parameters())
rng = RandomState((2017, 6, 30))
losses = []
accuracies = []
for i_epoch in range(6):
i_trials_in_batch = get_balanced_batches(len(train_set.X),
rng,
shuffle=True,
batch_size=10)
# Set model to training mode
model.train()
for i_trials in i_trials_in_batch:
# Have to add empty fourth dimension to X
batch_X = train_set.X[i_trials][:, :, :, None]
batch_y = train_set.y[i_trials]
net_in = np_to_var(batch_X)
if cuda:
net_in = net_in.cuda()
net_target = np_to_var(batch_y)
if cuda:
net_target = net_target.cuda()
# Remove gradients of last backward pass from all parameters
optimizer.zero_grad()
# Compute outputs of the network
#net_in: [10, 64, 497, 1]=[bsz, H_im, W_im, C_im]
#
outputs = model.forward(net_in)
# model=Sequential(
# (dimshuffle): Expression(expression=_transpose_time_to_spat)
# (conv_time): Conv2d(1, 40, kernel_size=(25, 1), stride=(1, 1))
# (conv_spat): Conv2d(40, 40, kernel_size=(1, 64), stride=(1, 1), bias=False)
# (bnorm): BatchNorm2d(40, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
# (conv_nonlin): Expression(expression=square)
# (pool): AvgPool2d(kernel_size=(75, 1), stride=(15, 1), padding=0)
# (pool_nonlin): Expression(expression=safe_log)
# (drop): Dropout(p=0.5)
# (conv_classifier): Conv2d(40, 2, kernel_size=(27, 1), stride=(1, 1))
# (softmax): LogSoftmax()
# (squeeze): Expression(expression=_squeeze_final_output)
# )
# Compute the loss
loss = F.nll_loss(outputs, net_target)
# Do the backpropagation
loss.backward()
# Update parameters with the optimizer
optimizer.step()
# Print some statistics each epoch
model.eval()
print("Epoch {:d}".format(i_epoch))
for setname, dataset in (('Train', train_set), ('Test', test_set)):
# Here, we will use the entire dataset at once, which is still possible
# for such smaller datasets. Otherwise we would have to use batches.
net_in = np_to_var(dataset.X[:, :, :, None])
if cuda:
net_in = net_in.cuda()
net_target = np_to_var(dataset.y)
if cuda:
net_target = net_target.cuda()
outputs = model(net_in)
loss = F.nll_loss(outputs, net_target)
losses.append(float(var_to_np(loss)))
print("{:6s} Loss: {:.5f}".format(setname, float(var_to_np(loss))))
predicted_labels = np.argmax(var_to_np(outputs), axis=1)
accuracy = np.mean(dataset.y == predicted_labels)
accuracies.append(accuracy * 100)
print("{:6s} Accuracy: {:.1f}%".format(setname, accuracy * 100))
np.testing.assert_allclose(np.array(losses),
np.array([
1.1775966882705688, 1.2602351903915405,
0.7068756818771362, 0.9367912411689758,
0.394258975982666, 0.6598362326622009,
0.3359280526638031, 0.656258761882782,
0.2790488004684448, 0.6104397177696228,
0.27319177985191345, 0.5949864983558655
]),
rtol=1e-4,
atol=1e-5)
np.testing.assert_allclose(np.array(accuracies),
np.array([
51.666666666666671, 53.333333333333336,
63.333333333333329, 56.666666666666664,
86.666666666666671, 66.666666666666657,
90.0, 63.333333333333329,
96.666666666666671, 56.666666666666664,
96.666666666666671, 66.666666666666657
]),
rtol=1e-4,
atol=1e-5)
def run_exp(max_recording_mins, n_recordings, sec_to_cut,
duration_recording_mins, max_abs_val, max_min_threshold,
max_min_expected, shrink_val, max_min_remove, batch_set_zero_val,
batch_set_zero_test, sampling_freq, low_cut_hz, high_cut_hz,
exp_demean, exp_standardize, moving_demean, moving_standardize,
channel_demean, channel_standardize, divisor, n_folds, i_test_fold,
input_time_length, final_conv_length, pool_stride, n_blocks_to_add,
sigmoid, model_constraint, batch_size, max_epochs,
only_return_exp):
cuda = True
preproc_functions = []
preproc_functions.append(lambda data, fs: (
data[:, int(sec_to_cut * fs):-int(sec_to_cut * fs)], fs))
preproc_functions.append(lambda data, fs: (data[:, :int(
duration_recording_mins * 60 * fs)], fs))
if max_abs_val is not None:
preproc_functions.append(
lambda data, fs: (np.clip(data, -max_abs_val, max_abs_val), fs))
if max_min_threshold is not None:
preproc_functions.append(lambda data, fs: (clean_jumps(
data, 200, max_min_threshold, max_min_expected, cuda), fs))
if max_min_remove is not None:
window_len = 200
preproc_functions.append(lambda data, fs: (set_jumps_to_zero(
data,
window_len=window_len,
threshold=max_min_remove,
cuda=cuda,
clip_min_max_to_zero=True), fs))
if shrink_val is not None:
preproc_functions.append(lambda data, fs: (shrink_spikes(
data,
shrink_val,
1,
9,
), fs))
preproc_functions.append(lambda data, fs: (resampy.resample(
data, fs, sampling_freq, axis=1, filter='kaiser_fast'), sampling_freq))
preproc_functions.append(lambda data, fs: (bandpass_cnt(
data, low_cut_hz, high_cut_hz, fs, filt_order=4, axis=1), fs))
if exp_demean:
preproc_functions.append(lambda data, fs: (exponential_running_demean(
data.T, factor_new=0.001, init_block_size=100).T, fs))
if exp_standardize:
preproc_functions.append(
lambda data, fs: (exponential_running_standardize(
data.T, factor_new=0.001, init_block_size=100).T, fs))
if moving_demean:
preproc_functions.append(lambda data, fs: (padded_moving_demean(
data, axis=1, n_window=201), fs))
if moving_standardize:
preproc_functions.append(lambda data, fs: (padded_moving_standardize(
data, axis=1, n_window=201), fs))
if channel_demean:
preproc_functions.append(lambda data, fs: (demean(data, axis=1), fs))
if channel_standardize:
preproc_functions.append(lambda data, fs:
(standardize(data, axis=1), fs))
if divisor is not None:
preproc_functions.append(lambda data, fs: (data / divisor, fs))
dataset = DiagnosisSet(n_recordings=n_recordings,
max_recording_mins=max_recording_mins,
preproc_functions=preproc_functions)
if not only_return_exp:
X, y = dataset.load()
splitter = Splitter(
n_folds,
i_test_fold,
)
if not only_return_exp:
train_set, valid_set, test_set = splitter.split(X, y)
del X, y # shouldn't be necessary, but just to make sure
else:
train_set = None
valid_set = None
test_set = None
set_random_seeds(seed=20170629, cuda=cuda)
if sigmoid:
n_classes = 1
else:
n_classes = 2
in_chans = 21
net = Deep4Net(
in_chans=in_chans,
n_classes=n_classes,
input_time_length=input_time_length,
final_conv_length=final_conv_length,
pool_time_length=pool_stride,
pool_time_stride=pool_stride,
n_filters_2=50,
n_filters_3=80,
n_filters_4=120,
)
model = net_with_more_layers(net, n_blocks_to_add, nn.MaxPool2d)
if sigmoid:
model = to_linear_plus_minus_net(model)
optimizer = optim.Adam(model.parameters())
to_dense_prediction_model(model)
log.info("Model:\n{:s}".format(str(model)))
if cuda:
model.cuda()
# determine output size
test_input = np_to_var(
np.ones((2, in_chans, input_time_length, 1), dtype=np.float32))
if cuda:
test_input = test_input.cuda()
out = model(test_input)
n_preds_per_input = out.cpu().data.numpy().shape[2]
log.info("{:d} predictions per input/trial".format(n_preds_per_input))
iterator = CropsFromTrialsIterator(batch_size=batch_size,
input_time_length=input_time_length,
n_preds_per_input=n_preds_per_input)
if sigmoid:
loss_function = lambda preds, targets: binary_cross_entropy_with_logits(
th.mean(preds, dim=2)[:, 1, 0], targets.type_as(preds))
else:
loss_function = lambda preds, targets: F.nll_loss(
th.mean(preds, dim=2)[:, :, 0], targets)
if model_constraint is not None:
model_constraint = MaxNormDefaultConstraint()
monitors = [
LossMonitor(),
MisclassMonitor(col_suffix='sample_misclass'),
CroppedTrialMisclassMonitor(input_time_length),
RuntimeMonitor(),
]
stop_criterion = MaxEpochs(max_epochs)
batch_modifier = None
if batch_set_zero_val is not None:
batch_modifier = RemoveMinMaxDiff(batch_set_zero_val,
clip_max_abs=True,
set_zero=True)
if (batch_set_zero_val is not None) and (batch_set_zero_test == True):
iterator = ModifiedIterator(
iterator,
batch_modifier,
)
batch_modifier = None
exp = Experiment(model,
train_set,
valid_set,
test_set,
iterator,
loss_function,
optimizer,
model_constraint,
monitors,
stop_criterion,
remember_best_column='valid_misclass',
run_after_early_stop=True,
batch_modifier=batch_modifier,
cuda=cuda)
if not only_return_exp:
exp.run()
else:
exp.dataset = dataset
exp.splitter = splitter
return exp
def train_model(self, train_set_1, val_set_1, test_set_1, train_set_2,
val_set_2, test_set_2, save_model):
"""
:param train_set_1: (np.array) n_trials*n_channels*n_samples
:param val_set_1: (np.array) n_trials*n_channels*n_samples
:param test_set_1: (np.array) n_trials*n_channels*n_samples - can be None when training on inner-fold
:param train_set_2: (np.array) n_trials*n_channels*n_samples
:param val_set_2: (np.array) n_trials*n_channels*n_samples
:param test_set_2: (np.array) n_trials*n_channels*n_samples - can be None when training on inner-fold
:param save_model: (Bool) True if trained model is to be saved
:return: Accuracy and loss scores for the model trained with a given set of hyper-parameters
"""
model = self.call_model()
predictions = None
set_random_seeds(seed=20190629, cuda=self.cuda)
if self.cuda:
model.cuda()
torch.backends.cudnn.deterministic = True
model = torch.nn.DataParallel(model)
log.info(f"Cuda in use")
log.info("%s model: ".format(str(model)))
optimizer = optim.Adam(model.parameters(),
lr=self.learning_rate,
weight_decay=0.01,
eps=1e-8,
amsgrad=False)
stop_criterion = Or([
MaxEpochs(self.epochs),
NoDecrease('valid_loss', self.max_increase_epochs)
])
model_loss_function = None
#####Setup to run the selected model#####
model_test = Experiment(model,
train_set_1,
val_set_1,
train_set_2,
val_set_2,
test_set_1=test_set_1,
test_set_2=test_set_2,
iterator=self.iterator,
loss_function=self.loss,
optimizer=optimizer,
lr_scheduler=self.lr_scheduler(
optimizer,
step_size=self.lr_step,
gamma=self.lr_gamma),
model_constraint=self.model_constraint,
monitors=self.monitors,
stop_criterion=stop_criterion,
remember_best_column='valid_misclass',
run_after_early_stop=True,
model_loss_function=model_loss_function,
cuda=self.cuda,
save_file=self.model_save_path,
tag=self.tag,
save_model=save_model)
model_test.run()
model_acc = model_test.epochs_df['valid_misclass'].astype('float')
model_loss = model_test.epochs_df['valid_loss'].astype('float')
current_val_acc = 1 - current_acc(model_acc)
current_val_loss = current_loss(model_loss)
test_accuracy = None
if train_set_1 is not None and test_set_2 is not None:
val_metric_index = self.get_model_index(model_test.epochs_df)
test_accuracy = round(
(1 -
model_test.epochs_df['test_misclass'].iloc[val_metric_index])
* 100, 3)
predictions = model_test.model_predictions
probabilities = model_test.model_probabilities
return current_val_acc, current_val_loss, test_accuracy, model_test, predictions, probabilities
def run_exp(data_folder, session_id, subject_id, low_cut_hz, model, cuda):
ival = [-500, 4000]
max_epochs = 1600
max_increase_epochs = 160
batch_size = 10
high_cut_hz = 38
factor_new = 1e-3
init_block_size = 1000
valid_set_fraction = .2
''' # BCIcompetition
train_filename = 'A{:02d}T.gdf'.format(subject_id)
test_filename = 'A{:02d}E.gdf'.format(subject_id)
train_filepath = os.path.join(data_folder, train_filename)
test_filepath = os.path.join(data_folder, test_filename)
train_label_filepath = train_filepath.replace('.gdf', '.mat')
test_label_filepath = test_filepath.replace('.gdf', '.mat')
train_loader = BCICompetition4Set2A(
train_filepath, labels_filename=train_label_filepath)
test_loader = BCICompetition4Set2A(
test_filepath, labels_filename=test_label_filepath)
train_cnt = train_loader.load()
test_cnt = test_loader.load()
'''
# GIGAscience
filename = 'sess{:02d}_subj{:02d}_EEG_MI.mat'.format(
session_id, subject_id)
filepath = os.path.join(data_folder, filename)
train_variable = 'EEG_MI_train'
test_variable = 'EEG_MI_test'
train_loader = GIGAscience(filepath, train_variable)
test_loader = GIGAscience(filepath, test_variable)
train_cnt = train_loader.load()
test_cnt = test_loader.load()
# Preprocessing
''' channel
['Fp1', 'Fp2', 'F7', 'F3', 'Fz', 'F4', 'F8', 'FC5', 'FC1', 'FC2', 'FC6', 'T7', 'C3', 'Cz', 'C4', 'T8', 'TP9', 'CP5',
'CP1', 'CP2', 'CP6', 'TP10', 'P7', 'P3', 'Pz', 'P4', 'P8', 'PO9', 'O1', 'Oz', 'O2', 'PO10', 'FC3', 'FC4', 'C5',
'C1', 'C2', 'C6', 'CP3', 'CPz', 'CP4', 'P1', 'P2', 'POz', 'FT9', 'FTT9h', 'TTP7h', 'TP7', 'TPP9h', 'FT10',
'FTT10h', 'TPP8h', 'TP8', 'TPP10h', 'F9', 'F10', 'AF7', 'AF3', 'AF4', 'AF8', 'PO3', 'PO4']
'''
train_cnt = train_cnt.pick_channels([
'FC5', 'FC3', 'FC1', 'Fz', 'FC2', 'FC4', 'FC6', 'C5', 'C3', 'C1', 'Cz',
'C2', 'C4', 'C6', 'CP5', 'CP3', 'CP1', 'CPz', 'CP2', 'CP4', 'CP6', 'Pz'
])
train_cnt, train_cnt.info['events'] = train_cnt.copy().resample(
250, npad='auto', events=train_cnt.info['events'])
assert len(train_cnt.ch_names) == 22
# lets convert to millvolt for numerical stability of next operations
train_cnt = mne_apply(lambda a: a * 1e6, train_cnt)
train_cnt = mne_apply(
lambda a: bandpass_cnt(a,
low_cut_hz,
high_cut_hz,
train_cnt.info['sfreq'],
filt_order=3,
axis=1), train_cnt)
train_cnt = mne_apply(
lambda a: exponential_running_standardize(a.T,
factor_new=factor_new,
init_block_size=
init_block_size,
eps=1e-4).T, train_cnt)
test_cnt = test_cnt.pick_channels([
'FC5', 'FC3', 'FC1', 'Fz', 'FC2', 'FC4', 'FC6', 'C5', 'C3', 'C1', 'Cz',
'C2', 'C4', 'C6', 'CP5', 'CP3', 'CP1', 'CPz', 'CP2', 'CP4', 'CP6', 'Pz'
])
test_cnt, test_cnt.info['events'] = test_cnt.copy().resample(
250, npad='auto', events=test_cnt.info['events'])
assert len(test_cnt.ch_names) == 22
test_cnt = mne_apply(lambda a: a * 1e6, test_cnt)
test_cnt = mne_apply(
lambda a: bandpass_cnt(a,
low_cut_hz,
high_cut_hz,
test_cnt.info['sfreq'],
filt_order=3,
axis=1), test_cnt)
test_cnt = mne_apply(
lambda a: exponential_running_standardize(a.T,
factor_new=factor_new,
init_block_size=
init_block_size,
eps=1e-4).T, test_cnt)
marker_def = OrderedDict([('Right Hand', [1]), ('Left Hand', [2])])
train_set = create_signal_target_from_raw_mne(train_cnt, marker_def, ival)
test_set = create_signal_target_from_raw_mne(test_cnt, marker_def, ival)
train_set, valid_set = split_into_two_sets(train_set,
first_set_fraction=1 -
valid_set_fraction)
set_random_seeds(seed=20190706, cuda=cuda)
n_classes = 2
n_chans = int(train_set.X.shape[1])
input_time_length = train_set.X.shape[2]
if model == 'shallow':
model = ShallowFBCSPNet(n_chans,
n_classes,
input_time_length=input_time_length,
final_conv_length='auto').create_network()
elif model == 'deep':
model = Deep4Net(n_chans,
n_classes,
input_time_length=input_time_length,
final_conv_length='auto').create_network()
if cuda:
model.cuda()
log.info("Model: \n{:s}".format(str(model)))
optimizer = optim.Adam(model.parameters())
iterator = BalancedBatchSizeIterator(batch_size=batch_size)
stop_criterion = Or([
MaxEpochs(max_epochs),
NoDecrease('valid_misclass', max_increase_epochs)
])
monitors = [LossMonitor(), MisclassMonitor(), RuntimeMonitor()]
model_constraint = MaxNormDefaultConstraint()
exp = Experiment(model,
train_set,
valid_set,
test_set,
iterator=iterator,
loss_function=F.nll_loss,
optimizer=optimizer,
model_constraint=model_constraint,
monitors=monitors,
stop_criterion=stop_criterion,
remember_best_column='valid_misclass',
run_after_early_stop=True,
cuda=cuda)
exp.run()
return exp
outpath = args.outpath
fold = args.fold
assert (fold >= 0 and fold < 54)
# Randomly shuffled subject.
subjs = [
35, 47, 46, 37, 13, 27, 12, 32, 53, 54, 4, 40, 19, 41, 18, 42, 34, 7, 49,
9, 5, 48, 29, 15, 21, 17, 31, 45, 1, 38, 51, 8, 11, 16, 28, 44, 24, 52, 3,
26, 39, 50, 6, 23, 2, 14, 25, 20, 10, 33, 22, 43, 36, 30
]
test_subj = subjs[fold]
cv_set = np.array(subjs[fold + 1:] + subjs[:fold])
kf = KFold(n_splits=6)
dfile = h5py.File(datapath, 'r')
torch.cuda.set_device(args.gpu)
set_random_seeds(seed=20200205, cuda=True)
BATCH_SIZE = 16
TRAIN_EPOCH = 200 # consider 200 for early stopping
# Get data from single subject.
def get_data(subj):
dpath = '/s' + str(subj)
X = dfile[pjoin(dpath, 'X')]
Y = dfile[pjoin(dpath, 'Y')]
return X, Y
def get_multi_data(subjs):
Xs = []
global_vars.set('dataset', dataset)
set_params_by_dataset()
global_vars.set('cuda', True)
model_select = 'deep4'
model_dir = '143_x_evolution_layers_cross_subject'
model_name = 'best_model_9_8_6_7_2_1_3_4_5.th'
train_set = {}
val_set = {}
test_set = {}
train_set[subject_id], val_set[subject_id], test_set[subject_id] =\
get_train_val_test(data_folder, subject_id)
# Set if you want to use GPU
# You can also use torch.cuda.is_available() to determine if cuda is available on your machine.
cuda = True
set_random_seeds(seed=20170629, cuda=cuda)
# This will determine how many crops are processed in parallel
input_time_length = 450
# final_conv_length determines the size of the receptive field of the ConvNet
models = {
'evolution': torch.load(f'../models/{model_dir}/{model_name}'),
'deep4': target_model('deep')
}
model = models[model_select]
input_time_length = global_vars.get('input_time_len')
stop_criterion, iterator, loss_function, monitors = get_normal_settings()
naiveNAS = NaiveNAS(iterator=iterator,
exp_folder=None,
exp_name=None,
train_set=train_set,
def run_exp(data_folder, subject_id, low_cut_hz, model, cuda):
train_filename = 'A{:02d}T.gdf'.format(subject_id)
test_filename = 'A{:02d}E.gdf'.format(subject_id)
train_filepath = os.path.join(data_folder, train_filename)
test_filepath = os.path.join(data_folder, test_filename)
train_label_filepath = train_filepath.replace('.gdf', '.mat')
test_label_filepath = test_filepath.replace('.gdf', '.mat')
train_loader = BCICompetition4Set2A(train_filepath,
labels_filename=train_label_filepath)
test_loader = BCICompetition4Set2A(test_filepath,
labels_filename=test_label_filepath)
train_cnt = train_loader.load()
test_cnt = test_loader.load()
# Preprocessing
train_cnt = train_cnt.drop_channels(
['STI 014', 'EOG-left', 'EOG-central', 'EOG-right'])
assert len(train_cnt.ch_names) == 22
# lets convert to millvolt for numerical stability of next operations
train_cnt = mne_apply(lambda a: a * 1e6, train_cnt)
train_cnt = mne_apply(
lambda a: bandpass_cnt(
a, low_cut_hz, 38, train_cnt.info['sfreq'], filt_order=3, axis=1),
train_cnt)
train_cnt = mne_apply(
lambda a: exponential_running_standardize(
a.T, factor_new=1e-3, init_block_size=1000, eps=1e-4).T, train_cnt)
test_cnt = test_cnt.drop_channels(
['STI 014', 'EOG-left', 'EOG-central', 'EOG-right'])
assert len(test_cnt.ch_names) == 22
test_cnt = mne_apply(lambda a: a * 1e6, test_cnt)
test_cnt = mne_apply(
lambda a: bandpass_cnt(
a, low_cut_hz, 38, test_cnt.info['sfreq'], filt_order=3, axis=1),
test_cnt)
test_cnt = mne_apply(
lambda a: exponential_running_standardize(
a.T, factor_new=1e-3, init_block_size=1000, eps=1e-4).T, test_cnt)
marker_def = OrderedDict([('Left Hand', [1]), (
'Right Hand',
[2],
), ('Foot', [3]), ('Tongue', [4])])
ival = [-500, 4000]
train_set = create_signal_target_from_raw_mne(train_cnt, marker_def, ival)
test_set = create_signal_target_from_raw_mne(test_cnt, marker_def, ival)
train_set, valid_set = split_into_two_sets(train_set,
first_set_fraction=0.8)
set_random_seeds(seed=20190706, cuda=cuda)
n_classes = 4
n_chans = int(train_set.X.shape[1])
input_time_length = train_set.X.shape[2]
if model == 'shallow':
model = ShallowFBCSPNet(n_chans,
n_classes,
input_time_length=input_time_length,
final_conv_length='auto').create_network()
elif model == 'deep':
model = Deep4Net(n_chans,
n_classes,
input_time_length=input_time_length,
final_conv_length='auto').create_network()
if cuda:
model.cuda()
log.info("Model: \n{:s}".format(str(model)))
optimizer = optim.Adam(model.parameters())
iterator = BalancedBatchSizeIterator(batch_size=60)
stop_criterion = Or([MaxEpochs(1600), NoDecrease('valid_misclass', 160)])
monitors = [LossMonitor(), MisclassMonitor(), RuntimeMonitor()]
model_constraint = MaxNormDefaultConstraint()
exp = Experiment(model,
train_set,
valid_set,
test_set,
iterator=iterator,
loss_function=F.nll_loss,
optimizer=optimizer,
model_constraint=model_constraint,
monitors=monitors,
stop_criterion=stop_criterion,
remember_best_column='valid_misclass',
run_after_early_stop=True,
cuda=cuda)
exp.run()
return exp
def run_exp(test_on_eval, sensor_types, n_chans, max_recording_mins,
test_recording_mins, n_recordings, sec_to_cut_at_start,
sec_to_cut_at_end, duration_recording_mins, max_abs_val,
clip_before_resample, sampling_freq, divisor, n_folds, i_test_fold,
shuffle, merge_train_valid, model_name, n_start_chans,
n_chan_factor, input_time_length, final_conv_length,
stride_before_pool, optimizer, learning_rate, weight_decay,
scheduler, model_constraint, batch_size, max_epochs, log_dir,
only_return_exp, np_th_seed):
cuda = True
if ('smac' in model_name) and (input_time_length is None):
input_time_length = 12000
fix_input_length_for_smac = True
else:
fix_input_length_for_smac = False
set_random_seeds(seed=np_th_seed, cuda=cuda)
n_classes = 2
if model_name == 'shallow':
model = ShallowFBCSPNet(
in_chans=n_chans,
n_classes=n_classes,
n_filters_time=n_start_chans,
n_filters_spat=n_start_chans,
input_time_length=input_time_length,
final_conv_length=final_conv_length).create_network()
elif model_name == 'deep':
model = Deep4Net(
n_chans,
n_classes,
n_filters_time=n_start_chans,
n_filters_spat=n_start_chans,
input_time_length=input_time_length,
n_filters_2=int(n_start_chans * n_chan_factor),
n_filters_3=int(n_start_chans * (n_chan_factor**2.0)),
n_filters_4=int(n_start_chans * (n_chan_factor**3.0)),
final_conv_length=final_conv_length,
stride_before_pool=stride_before_pool).create_network()
elif (model_name == 'deep_smac') or (model_name == 'deep_smac_bnorm'):
if model_name == 'deep_smac':
do_batch_norm = False
else:
assert model_name == 'deep_smac_bnorm'
do_batch_norm = True
double_time_convs = False
drop_prob = 0.244445
filter_length_2 = 12
filter_length_3 = 14
filter_length_4 = 12
filter_time_length = 21
final_conv_length = 1
first_nonlin = elu
first_pool_mode = 'mean'
first_pool_nonlin = identity
later_nonlin = elu
later_pool_mode = 'mean'
later_pool_nonlin = identity
n_filters_factor = 1.679066
n_filters_start = 32
pool_time_length = 1
pool_time_stride = 2
split_first_layer = True
n_chan_factor = n_filters_factor
n_start_chans = n_filters_start
model = Deep4Net(n_chans,
n_classes,
n_filters_time=n_start_chans,
n_filters_spat=n_start_chans,
input_time_length=input_time_length,
n_filters_2=int(n_start_chans * n_chan_factor),
n_filters_3=int(n_start_chans * (n_chan_factor**2.0)),
n_filters_4=int(n_start_chans * (n_chan_factor**3.0)),
final_conv_length=final_conv_length,
batch_norm=do_batch_norm,
double_time_convs=double_time_convs,
drop_prob=drop_prob,
filter_length_2=filter_length_2,
filter_length_3=filter_length_3,
filter_length_4=filter_length_4,
filter_time_length=filter_time_length,
first_nonlin=first_nonlin,
first_pool_mode=first_pool_mode,
first_pool_nonlin=first_pool_nonlin,
later_nonlin=later_nonlin,
later_pool_mode=later_pool_mode,
later_pool_nonlin=later_pool_nonlin,
pool_time_length=pool_time_length,
pool_time_stride=pool_time_stride,
split_first_layer=split_first_layer,
stride_before_pool=True).create_network()
elif model_name == 'shallow_smac':
conv_nonlin = identity
do_batch_norm = True
drop_prob = 0.328794
filter_time_length = 56
final_conv_length = 22
n_filters_spat = 73
n_filters_time = 24
pool_mode = 'max'
pool_nonlin = identity
pool_time_length = 84
pool_time_stride = 3
split_first_layer = True
model = ShallowFBCSPNet(
in_chans=n_chans,
n_classes=n_classes,
n_filters_time=n_filters_time,
n_filters_spat=n_filters_spat,
input_time_length=input_time_length,
final_conv_length=final_conv_length,
conv_nonlin=conv_nonlin,
batch_norm=do_batch_norm,
drop_prob=drop_prob,
filter_time_length=filter_time_length,
pool_mode=pool_mode,
pool_nonlin=pool_nonlin,
pool_time_length=pool_time_length,
pool_time_stride=pool_time_stride,
split_first_layer=split_first_layer,
).create_network()
elif model_name == 'deep_smac_new':
from torch.nn.functional import elu, relu, relu6, tanh
from braindecode.torch_ext.functions import identity, square, safe_log
n_filters_factor = 1.9532637176784269
n_filters_start = 61
deep_kwargs = {
"batch_norm": False,
"double_time_convs": False,
"drop_prob": 0.3622676569047184,
"filter_length_2": 9,
"filter_length_3": 6,
"filter_length_4": 10,
"filter_time_length": 17,
"final_conv_length": 5,
"first_nonlin": elu,
"first_pool_mode": "max",
"first_pool_nonlin": identity,
"later_nonlin": relu6,
"later_pool_mode": "max",
"later_pool_nonlin": identity,
"n_filters_time": n_filters_start,
"n_filters_spat": n_filters_start,
"n_filters_2": int(n_filters_start * n_filters_factor),
"n_filters_3": int(n_filters_start * (n_filters_factor**2.0)),
"n_filters_4": int(n_filters_start * (n_filters_factor**3.0)),
"pool_time_length": 1,
"pool_time_stride": 4,
"split_first_layer": True,
"stride_before_pool": True,
}
model = Deep4Net(n_chans,
n_classes,
input_time_length=input_time_length,
**deep_kwargs).create_network()
elif model_name == 'shallow_smac_new':
from torch.nn.functional import elu, relu, relu6, tanh
from braindecode.torch_ext.functions import identity, square, safe_log
shallow_kwargs = {
"conv_nonlin": square,
"batch_norm": True,
"drop_prob": 0.10198630723385381,
"filter_time_length": 51,
"final_conv_length": 1,
"n_filters_spat": 200,
"n_filters_time": 76,
"pool_mode": "max",
"pool_nonlin": safe_log,
"pool_time_length": 139,
"pool_time_stride": 49,
"split_first_layer": True,
}
model = ShallowFBCSPNet(in_chans=n_chans,
n_classes=n_classes,
input_time_length=input_time_length,
**shallow_kwargs).create_network()
elif model_name == 'linear':
model = nn.Sequential()
model.add_module("conv_classifier",
nn.Conv2d(n_chans, n_classes, (600, 1)))
model.add_module('softmax', nn.LogSoftmax())
model.add_module('squeeze', Expression(lambda x: x.squeeze(3)))
elif model_name == '3path':
virtual_chan_1x1_conv = True
mean_across_features = False
drop_prob = 0.5
n_start_filters = 10
early_bnorm = False
n_classifier_filters = 100
later_kernel_len = 5
extra_conv_stride = 4
# dont forget to reset n_preds_per_blabla
model = create_multi_start_path_net(
in_chans=n_chans,
virtual_chan_1x1_conv=virtual_chan_1x1_conv,
n_start_filters=n_start_filters,
early_bnorm=early_bnorm,
later_kernel_len=later_kernel_len,
extra_conv_stride=extra_conv_stride,
mean_across_features=mean_across_features,
n_classifier_filters=n_classifier_filters,
drop_prob=drop_prob)
else:
assert False, "unknown model name {:s}".format(model_name)
if not model_name == '3path':
to_dense_prediction_model(model)
log.info("Model:\n{:s}".format(str(model)))
time_cut_off_sec = np.inf
start_time = time.time()
# fix input time length in case of smac models
if fix_input_length_for_smac:
assert ('smac' in model_name) and (input_time_length == 12000)
if cuda:
model.cuda()
test_input = np_to_var(
np.ones((2, n_chans, input_time_length, 1), dtype=np.float32))
if cuda:
test_input = test_input.cuda()
try:
out = model(test_input)
except:
raise ValueError("Model receptive field too large...")
n_preds_per_input = out.cpu().data.numpy().shape[2]
n_receptive_field = input_time_length - n_preds_per_input
input_time_length = 2 * n_receptive_field
exp = common.run_exp(
max_recording_mins,
n_recordings,
sec_to_cut_at_start,
sec_to_cut_at_end,
duration_recording_mins,
max_abs_val,
clip_before_resample,
sampling_freq,
divisor,
n_folds,
i_test_fold,
shuffle,
merge_train_valid,
model,
input_time_length,
optimizer,
learning_rate,
weight_decay,
scheduler,
model_constraint,
batch_size,
max_epochs,
only_return_exp,
time_cut_off_sec,
start_time,
test_on_eval,
test_recording_mins,
sensor_types,
log_dir,
np_th_seed,
)
return exp
def network_model(subject_id, model_type, data_type, cropped, cuda, parameters, hyp_params):
best_params = dict() # dictionary to store hyper-parameter values
#####Parameter passed to funciton#####
max_epochs = parameters['max_epochs']
max_increase_epochs = parameters['max_increase_epochs']
batch_size = parameters['batch_size']
#####Constant Parameters#####
best_loss = 100.0 # instatiate starting point for loss
iterator = BalancedBatchSizeIterator(batch_size=batch_size)
stop_criterion = Or([MaxEpochs(max_epochs),
NoDecrease('valid_misclass', max_increase_epochs)])
monitors = [LossMonitor(), MisclassMonitor(), RuntimeMonitor()]
model_constraint = MaxNormDefaultConstraint()
epoch = 4096
#####Collect and format data#####
if data_type == 'words':
data, labels = format_data(data_type, subject_id, epoch)
data = data[:,:,768:1280] # within-trial window selected for classification
elif data_type == 'vowels':
data, labels = format_data(data_type, subject_id, epoch)
data = data[:,:,512:1024]
elif data_type == 'all_classes':
data, labels = format_data(data_type, subject_id, epoch)
data = data[:,:,768:1280]
x = lambda a: a * 1e6 # improves numerical stability
data = x(data)
data = normalize(data)
data, labels = balanced_subsample(data, labels) # downsampling the data to ensure equal classes
data, _, labels, _ = train_test_split(data, labels, test_size=0, random_state=42) # redundant shuffle of data/labels
#####model inputs#####
unique, counts = np.unique(labels, return_counts=True)
n_classes = len(unique)
n_chans = int(data.shape[1])
input_time_length = data.shape[2]
#####k-fold nested corss-validation#####
num_folds = 4
skf = StratifiedKFold(n_splits=num_folds, shuffle=True, random_state=10)
out_fold_num = 0 # outer-fold number
cv_scores = []
#####Outer=Fold#####
for inner_ind, outer_index in skf.split(data, labels):
inner_fold, outer_fold = data[inner_ind], data[outer_index]
inner_labels, outer_labels = labels[inner_ind], labels[outer_index]
out_fold_num += 1
# list for storing cross-validated scores
loss_with_params = dict()# for storing param values and losses
in_fold_num = 0 # inner-fold number
#####Inner-Fold#####
for train_idx, valid_idx in skf.split(inner_fold, inner_labels):
X_Train, X_val = inner_fold[train_idx], inner_fold[valid_idx]
y_train, y_val = inner_labels[train_idx], inner_labels[valid_idx]
in_fold_num += 1
train_set = SignalAndTarget(X_Train, y_train)
valid_set = SignalAndTarget(X_val, y_val)
loss_with_params[f"Fold_{in_fold_num}"] = dict()
####Nested cross-validation#####
for drop_prob in hyp_params['drop_prob']:
for loss_function in hyp_params['loss']:
for i in range(len(hyp_params['lr_adam'])):
model = None # ensure no duplication of models
# model, learning-rate and optimizer setup according to model_type
if model_type == 'shallow':
model = ShallowFBCSPNet(in_chans=n_chans, n_classes=n_classes, input_time_length=input_time_length,
n_filters_time=80, filter_time_length=40, n_filters_spat=80,
pool_time_length=75, pool_time_stride=25, final_conv_length='auto',
conv_nonlin=square, pool_mode='max', pool_nonlin=safe_log,
split_first_layer=True, batch_norm=True, batch_norm_alpha=0.1,
drop_prob=drop_prob).create_network()
lr = hyp_params['lr_ada'][i]
optimizer = optim.Adadelta(model.parameters(), lr=lr, rho=0.9, weight_decay=0.1, eps=1e-8)
elif model_type == 'deep':
model = Deep4Net(in_chans=n_chans, n_classes=n_classes, input_time_length=input_time_length,
final_conv_length='auto', n_filters_time=20, n_filters_spat=20, filter_time_length=10,
pool_time_length=3, pool_time_stride=3, n_filters_2=50, filter_length_2=15,
n_filters_3=100, filter_length_3=15, n_filters_4=400, filter_length_4=10,
first_nonlin=leaky_relu, first_pool_mode='max', first_pool_nonlin=safe_log, later_nonlin=leaky_relu,
later_pool_mode='max', later_pool_nonlin=safe_log, drop_prob=drop_prob,
double_time_convs=False, split_first_layer=False, batch_norm=True, batch_norm_alpha=0.1,
stride_before_pool=False).create_network() #filter_length_4 changed from 15 to 10
lr = hyp_params['lr_ada'][i]
optimizer = optim.Adadelta(model.parameters(), lr=lr, weight_decay=0.1, eps=1e-8)
elif model_type == 'eegnet':
model = EEGNetv4(in_chans=n_chans, n_classes=n_classes, final_conv_length='auto',
input_time_length=input_time_length, pool_mode='mean', F1=16, D=2, F2=32,
kernel_length=64, third_kernel_size=(8,4), drop_prob=drop_prob).create_network()
lr = hyp_params['lr_adam'][i]
optimizer = optim.Adam(model.parameters(), lr=lr, weight_decay=0, eps=1e-8, amsgrad=False)
set_random_seeds(seed=20190629, cuda=cuda)
if cuda:
model.cuda()
torch.backends.cudnn.deterministic = True
model = torch.nn.DataParallel(model)
log.info("%s model: ".format(str(model)))
loss_function = loss_function
model_loss_function = None
#####Setup to run the selected model#####
model_test = Experiment(model, train_set, valid_set, test_set=None, iterator=iterator,
loss_function=loss_function, optimizer=optimizer,
model_constraint=model_constraint, monitors=monitors,
stop_criterion=stop_criterion, remember_best_column='valid_misclass',
run_after_early_stop=True, model_loss_function=model_loss_function, cuda=cuda,
data_type=data_type, subject_id=subject_id, model_type=model_type,
cropped=cropped, model_number=str(out_fold_num))
model_test.run()
model_loss = model_test.epochs_df['valid_loss'].astype('float')
current_val_loss = current_loss(model_loss)
loss_with_params[f"Fold_{in_fold_num}"][f"{drop_prob}/{loss_function}/{lr}"] = current_val_loss
####Select and train optimized model#####
df = pd.DataFrame(loss_with_params)
df['mean'] = df.mean(axis=1) # compute mean loss across k-folds
writer_df = f"results_folder\\results\\S{subject_id}\\{model_type}_parameters.xlsx"
df.to_excel(writer_df)
best_dp, best_loss, best_lr = df.loc[df['mean'].idxmin()].__dict__['_name'].split("/") # extract best param values
if str(best_loss[10:13]) == 'nll':
best_loss = F.nll_loss
elif str(best_loss[10:13]) == 'cro':
best_loss = F.cross_entropy
print(f"Best parameters: dropout: {best_dp}, loss: {str(best_loss)[10:13]}, lr: {best_lr}")
#####Train model on entire inner fold set#####
torch.backends.cudnn.deterministic = True
model = None
#####Create outer-fold validation and test sets#####
X_valid, X_test, y_valid, y_test = train_test_split(outer_fold, outer_labels, test_size=0.5, random_state=42, stratify=outer_labels)
train_set = SignalAndTarget(inner_fold, inner_labels)
valid_set = SignalAndTarget(X_valid, y_valid)
test_set = SignalAndTarget(X_test, y_test)
if model_type == 'shallow':
model = ShallowFBCSPNet(in_chans=n_chans, n_classes=n_classes, input_time_length=input_time_length,
n_filters_time=60, filter_time_length=5, n_filters_spat=40,
pool_time_length=50, pool_time_stride=15, final_conv_length='auto',
conv_nonlin=relu6, pool_mode='mean', pool_nonlin=safe_log,
split_first_layer=True, batch_norm=True, batch_norm_alpha=0.1,
drop_prob=0.1).create_network() #50 works better than 75
optimizer = optim.Adadelta(model.parameters(), lr=2.0, rho=0.9, weight_decay=0.1, eps=1e-8)
elif model_type == 'deep':
model = Deep4Net(in_chans=n_chans, n_classes=n_classes, input_time_length=input_time_length,
final_conv_length='auto', n_filters_time=20, n_filters_spat=20, filter_time_length=5,
pool_time_length=3, pool_time_stride=3, n_filters_2=20, filter_length_2=5,
n_filters_3=40, filter_length_3=5, n_filters_4=1500, filter_length_4=10,
first_nonlin=leaky_relu, first_pool_mode='mean', first_pool_nonlin=safe_log, later_nonlin=leaky_relu,
later_pool_mode='mean', later_pool_nonlin=safe_log, drop_prob=0.1,
double_time_convs=False, split_first_layer=True, batch_norm=True, batch_norm_alpha=0.1,
stride_before_pool=False).create_network()
optimizer = AdamW(model.parameters(), lr=0.1, weight_decay=0)
elif model_type == 'eegnet':
model = EEGNetv4(in_chans=n_chans, n_classes=n_classes, final_conv_length='auto',
input_time_length=input_time_length, pool_mode='mean', F1=16, D=2, F2=32,
kernel_length=64, third_kernel_size=(8,4), drop_prob=0.1).create_network()
optimizer = optim.Adam(model.parameters(), lr=0.1, weight_decay=0, eps=1e-8, amsgrad=False)
if cuda:
model.cuda()
torch.backends.cudnn.deterministic = True
#model = torch.nn.DataParallel(model)
log.info("Optimized model")
model_loss_function=None
#####Setup to run the optimized model#####
optimized_model = op_exp(model, train_set, valid_set, test_set=test_set, iterator=iterator,
loss_function=best_loss, optimizer=optimizer,
model_constraint=model_constraint, monitors=monitors,
stop_criterion=stop_criterion, remember_best_column='valid_misclass',
run_after_early_stop=True, model_loss_function=model_loss_function, cuda=cuda,
data_type=data_type, subject_id=subject_id, model_type=model_type,
cropped=cropped, model_number=str(out_fold_num))
optimized_model.run()
log.info("Last 5 epochs")
log.info("\n" + str(optimized_model.epochs_df.iloc[-5:]))
writer = f"results_folder\\results\\S{subject_id}\\{data_type}_{model_type}_{str(out_fold_num)}.xlsx"
optimized_model.epochs_df.iloc[-30:].to_excel(writer)
accuracy = 1 - np.min(np.array(optimized_model.class_acc))
cv_scores.append(accuracy) # k accuracy scores for this param set.
#####Print and store fold accuracies and mean accuracy#####
print(f"Class Accuracy: {np.mean(np.array(cv_scores))}")
results_df = pd.DataFrame(dict(cv_scores=cv_scores,
cv_mean=np.mean(np.array(cv_scores))))
writer2 = f"results_folder\\results\\S{subject_id}\\{data_type}_{model_type}_cvscores.xlsx"
results_df.to_excel(writer2)
return optimized_model, np.mean(np.array(cv_scores))
