def N170_test(session_data):
markers = N170_MARKERS
epochs = get_session_erp_epochs(session_data, markers)
conditions = OrderedDict()
for i in range(len(markers)):
conditions[markers[i]] = [i+1]
clfs = OrderedDict()
clfs['Vect + LR'] = make_pipeline(Vectorizer(), StandardScaler(), LogisticRegression())
clfs['Vect + RegLDA'] = make_pipeline(Vectorizer(), LDA(shrinkage='auto', solver='eigen'))
clfs['ERPCov + TS'] = make_pipeline(ERPCovariances(estimator='oas'), TangentSpace(), LogisticRegression())
clfs['ERPCov + MDM'] = make_pipeline(ERPCovariances(estimator='oas'), MDM())
clfs['XdawnCov + TS'] = make_pipeline(XdawnCovariances(estimator='oas'), TangentSpace(), LogisticRegression())
clfs['XdawnCov + MDM'] = make_pipeline(XdawnCovariances(estimator='oas'), MDM())
methods_list = ['Vect + LR','Vect + RegLDA','ERPCov + TS','ERPCov + MDM','XdawnCov + TS','XdawnCov + MDM']
# format data
epochs.pick_types(eeg=True)
X = epochs.get_data() * 1e6
times = epochs.times
y = epochs.events[:, -1]
# define cross validation
cv = StratifiedShuffleSplit(n_splits=20, test_size=0.25,
random_state=42)
# run cross validation for each pipeline
auc = []
methods = []
print('Calcul in progress...')
for m in clfs:
try:
res = cross_val_score(clfs[m], X, y==2, scoring='roc_auc',
cv=cv, n_jobs=-1)
auc.extend(res)
methods.extend([m]*len(res))
except Exception:
print("exception")
## Plot Decoding Results
results = pd.DataFrame(data=auc, columns=['AUC'])
results['Method'] = methods
n_row,n_column = results.shape
auc_means = []
for method in methods_list:
auc = []
for i in range(n_row):
if results.loc[i,'Method']== method:
auc.append(results.loc[i,'AUC'])
auc_means.append(np.mean(auc))
counter = 0
for i in range(len(methods_list)):
color = 'green' if auc_means[i]>=0.7 else 'red'
counter = counter +1 if auc_means[i]>=0.7 else counter
return counter > 0, counter
def xdawn_embedding(data, use_xdawn):
"""Perform embedding of EEG data in 2D Euclidean space
with Laplacian Eigenmaps.
Parameters
----------
data : dict
A dictionary containing training and testing data
Returns
-------
array
Embedded
"""
if use_xdawn:
nfilter = 3
xdwn = XdawnCovariances(estimator='scm', nfilter=nfilter)
covs = xdwn.fit(data['train_x'],
data['train_y']).transform(data['test_x'])
lapl = Embedding(metric='riemann', n_components=3)
embd = lapl.fit_transform(covs)
else:
tangent_space = Pipeline([
('cov_transform', Covariances(estimator='lwf')),
('tangent_space', TangentSpace(metric='riemann'))
])
t_space = tangent_space.fit(data['train_x'],
data['train_y']).transform(data['test_x'])
reducer = umap.UMAP(n_neighbors=30, min_dist=1, spread=2)
embd = reducer.fit_transform(t_space)
return embd
def get_sourcetarget_split_p300(source, target, ncovs_train):
X_source = source['signals']
y_source = source['labels'].flatten()
covs_source = XdawnCovariances(classes=[2]).fit_transform(
X_source, y_source)
source = {}
source['covs'] = covs_source
source['labels'] = y_source
X_target = target['signals']
y_target = target['labels'].flatten()
if ncovs_train is None:
ncovs_train = np.sum(y_target == 2)
sel = np.arange(len(y_target))
np.random.shuffle(sel)
X_target = X_target[sel]
y_target = y_target[sel]
idx_erps = np.where(y_target == 2)[0][:ncovs_train]
idx_rest = np.where(
y_target == 1)[0][:ncovs_train *
5] # because there's one ERP in every 6 flashes
idx_train = np.concatenate([idx_erps, idx_rest])
idx_test = np.array(
[i for i in range(len(y_target)) if i not in idx_train])
erp = XdawnCovariances(classes=[2])
erp.fit(X_target[idx_train], y_target[idx_train])
target_train = {}
covs_target_train = erp.transform(X_target[idx_train])
y_target_train = y_target[idx_train]
target_train['covs'] = covs_target_train
target_train['labels'] = y_target_train
target_test = {}
covs_target_test = erp.transform(X_target[idx_test])
y_target_test = y_target[idx_test]
target_test['covs'] = covs_target_test
target_test['labels'] = y_target_test
return source, target_train, target_test
def test_xdawn_covariances_applyfilters(rndstate, get_labels):
n_classes, nfilter = 2, 2
n_matrices, n_channels, n_times = 4, 6, 100
x = rndstate.randn(n_matrices, n_channels, n_times)
labels = get_labels(n_matrices, n_classes)
cov = XdawnCovariances(nfilter=nfilter, applyfilters=False)
covmats = cov.fit_transform(x, labels)
covsize = n_classes * nfilter + n_channels
assert covmats.shape == (n_matrices, covsize, covsize)
assert is_spsd(covmats)
def test_Xdawncovariances():
"""Test fit ERPCovariances"""
x = np.random.randn(10, 3, 100)
labels = np.array([0, 1]).repeat(5)
cov = XdawnCovariances()
cov.fit_transform(x, labels)
assert_equal(cov.get_params(), dict(nfilter=4, applyfilters=True,
classes=None, estimator='scm',
xdawn_estimator='scm',
baseline_cov=None))
def test_xdawn_covariances_nfilter(nfilter, rndstate, get_labels):
"""Test fit XdawnCovariances"""
n_classes, n_matrices, n_channels, n_times = 2, 4, 8, 100
x = rndstate.randn(n_matrices, n_channels, n_times)
labels = get_labels(n_matrices, n_classes)
cov = XdawnCovariances(nfilter=nfilter)
covmats = cov.fit_transform(x, labels)
assert cov.get_params() == dict(
nfilter=nfilter,
applyfilters=True,
classes=None,
estimator="scm",
xdawn_estimator="scm",
baseline_cov=None,
)
covsize = 2 * (n_classes * nfilter)
assert covmats.shape == (n_matrices, covsize, covsize)
assert is_spsd(covmats)
def xdawn_embedding(data):
"""Perform embedding of EEG data in 2D Euclidean space
with Laplacian Eigenmaps.
Parameters
----------
data : dict
A dictionary containing training and testing data
Returns
-------
array
Embedded
"""
nfilter = 3
xdwn = XdawnCovariances(estimator='scm', nfilter=nfilter)
covs = xdwn.fit(data['train_x'], data['train_y']).transform(data['test_x'])
lapl = Embedding(metric='riemann', n_components=3)
embd = lapl.fit_transform(covs)
return embd
reject={'eeg': 75e-6}, preload=True,
verbose=False, picks=[0,1,2,3])
print('sample drop %: ', (1 - len(epochs.events)/len(events)) * 100)
epochs
###################################################################################################
# Run classification
# ----------------------------
clfs = OrderedDict()
clfs['Vect + LR'] = make_pipeline(Vectorizer(), StandardScaler(), LogisticRegression())
clfs['Vect + RegLDA'] = make_pipeline(Vectorizer(), LDA(shrinkage='auto', solver='eigen'))
clfs['ERPCov + TS'] = make_pipeline(ERPCovariances(estimator='oas'), TangentSpace(), LogisticRegression())
clfs['ERPCov + MDM'] = make_pipeline(ERPCovariances(estimator='oas'), MDM())
clfs['XdawnCov + TS'] = make_pipeline(XdawnCovariances(estimator='oas'), TangentSpace(), LogisticRegression())
clfs['XdawnCov + MDM'] = make_pipeline(XdawnCovariances(estimator='oas'), MDM())
# format data
epochs.pick_types(eeg=True)
X = epochs.get_data() * 1e6
times = epochs.times
y = epochs.events[:, -1]
# define cross validation
cv = StratifiedShuffleSplit(n_splits=20, test_size=0.25,
random_state=42)
# run cross validation for each pipeline
auc = []
methods = []
subject_dir_list = test_list_np[i]
subject_epoch = np.empty((0, 56, 260), float)
for j in range(5):
subject_dir = subject_dir_list[j]
data = epoching('./data/test/' + subject_dir)
subject_epoch = np.vstack((subject_epoch, data))
subject_epoch = np.reshape(subject_epoch, (1, 340, 56, 260))
test_data_list = np.vstack((test_data_list, subject_epoch))
print('Epoched training data shape: ' + str(train_data_list.shape))
print('Epoched testing data shape: ' + str(test_data_list.shape))
########################## apply data preprocessing ############################
y_train = pd.read_csv('TrainLabels.csv')['Prediction'].values
y_test = np.reshape(pd.read_csv('true_labels.csv', header=None).values, 3400)
XC = XdawnCovariances(nfilter=5)
output_train = XC.fit_transform(
np.reshape(train_data_list, (16 * 340, 56, 260)), y_train)
X_train = TangentSpace(metric='riemann').fit_transform(output_train)
output_test = XC.fit_transform(np.reshape(test_data_list, (10 * 340, 56, 260)),
y_test)
X_test = TangentSpace(metric='riemann').fit_transform(output_test)
print('Preprocessed training data shape: ' + str(X_train.shape))
print('Preprocessed testing data shape: ' + str(X_test.shape))
############################## save data to disk ###############################
np.save('./data/train_data_56_260_1_40Hz.npy', train_data_list)
np.save('./data/test_data_56_260_1_40Hz.npy', test_data_list)
np.save('./data/X_train', X_train)
np.save('./data/X_test', X_test)
stats['VR'] = {}
stats['PC'] = {}
for condition in datasets.keys():
# get the epochs and labels
X, y, meta = paradigm.get_data(datasets[condition], subjects=[subject])
y = LabelEncoder().fit_transform(y)
data[condition]['X'] = X
data[condition]['y'] = y
# estimate xDawn covs
ncomps = 4
erp = XdawnCovariances(classes=[1],
estimator='lwf',
nfilter=ncomps,
xdawn_estimator='lwf')
#erp = ERPCovariances(classes=[1], estimator='lwf', svd=ncomps)
split = train_test_split(X, y, train_size=0.50, random_state=42)
Xtrain, Xtest, ytrain, ytest = split
covs = erp.fit(Xtrain, ytrain).transform(Xtest)
Mtarget = mean_riemann(covs[ytest == 1])
Mnontarget = mean_riemann(covs[ytest == 0])
stats[condition]['distance'] = distance_riemann(Mtarget, Mnontarget)
stats[condition]['dispersion_target'] = np.sum(
[distance_riemann(covi, Mtarget)**2
for covi in covs[ytest == 1]]) / len(covs[ytest == 1])
stats[condition]['dispersion_nontarget'] = np.sum([
distance_riemann(covi, Mnontarget)**2 for covi in covs[ytest == 0]
]) / len(covs[ytest == 0])
# ____________________________________________________________________________
# Create pipelines
# ----------------
# Pipelines must be a dict of sklearn pipeline transformer.
pipelines = {}
# we have to do this because the classes are called 'Target' and 'NonTarget'
# but the evaluation function uses a LabelEncoder, transforming them
# to 0 and 1
labels_dict = {'Target': 1, 'NonTarget': 0}
# %%
# from sklearn.preprocessing import StandardScaler
pipelines['RG + LRR'] = make_pipeline(
XdawnCovariances(nfilter=2,
classes=[labels_dict['Target']],
estimator='lwf',
xdawn_estimator='lwf'), TangentSpace(), LRR())
# %%
pipelines['Xdawn + LRR'] = make_pipeline(Xdawn(nfilter=2, estimator='lwf'),
Vectorizer(), LRR())
#%%
pipelines['LRR'] = make_pipeline(Vectorizer(), LRR())
# ____________________________________________________________________________
# Evaluation
# %%
paradigm = P300(resample=128)
dataset = BNCI2015003()
from pyriemann.channelselection import ElectrodeSelection
from sklearn.linear_model import LogisticRegression
from sklearn.pipeline import make_pipeline
from sklearn.cross_validation import KFold
from sklearn.metrics import roc_auc_score
from utils import (DownSampler, EpochsVectorizer, CospBoostingClassifier,
epoch_data)
dataframe1 = pd.read_csv('ecog_train_with_labels.csv')
array_clfs = OrderedDict()
# ERPs models
array_clfs['XdawnCov'] = make_pipeline(XdawnCovariances(6, estimator='oas'),
TangentSpace('riemann'),
LogisticRegression('l2'))
array_clfs['Xdawn'] = make_pipeline(Xdawn(12, estimator='oas'), DownSampler(5),
EpochsVectorizer(),
LogisticRegression('l2'))
# Induced activity models
baseclf = make_pipeline(
ElectrodeSelection(10, metric=dict(mean='logeuclid', distance='riemann')),
TangentSpace('riemann'), LogisticRegression('l1'))
array_clfs['Cosp'] = make_pipeline(
CospCovariances(fs=1000, window=32, overlap=0.95, fmax=300, fmin=1),
def pyR_decoding_on_full_epochs(X,
y,
plot_conf_matrix=0,
class_names=None,
test_size=0.2,
n_splits=5,
classifier='ERP_cov'):
""" This function decodes on the full epoch using the pyRiemannian decoder
cf https://github.com/Team-BK/Biomag2016/blob/master/Final_Submission.ipynb
Parameters
---------
X : data extracted from the epochs provided to the decoder
y : categorical variable (i.e. discrete but it can be more then 2 categories)
plot_confusion_matrix : set to 1 if you wanna see the confusion matrix
class_names: needed for the legend if confusion matrices are plotted ['cat1','cat2','cat3']
test_size : proportion of the data on which you wanna test the decoder
n_splits : when calculating the score, number of cross-validation folds
classifier : set it to 'ERP_cov', 'Xdawn_cov' or 'Hankel_cov' depending on the classification you want to do.
Returns: scores, y_test, y_pred, cnf_matrix or just scores if you don't want the confusion matrix
-------
"""
# ------- define the classifier -------
if classifier == 'ERP_cov':
spatial_filter = UnsupervisedSpatialFilter(PCA(20), average=False)
ERP_cov = ERPCovariances(estimator='lwf')
CSP_30 = CSP(30, log=False)
tang = TangentSpace('logeuclid')
clf = make_pipeline(spatial_filter, ERP_cov, CSP_30, tang,
LogisticRegression('l2'))
if classifier == 'Xdawn_cov':
clf = make_pipeline(
UnsupervisedSpatialFilter(PCA(50), average=False),
XdawnCovariances(12, estimator='lwf', xdawn_estimator='lwf'),
TangentSpace('logeuclid'), LogisticRegression('l2'))
if classifier == 'Hankel_cov':
clf = make_pipeline(
UnsupervisedSpatialFilter(PCA(70), average=False),
HankelCovariances(delays=[1, 8, 12, 64], estimator='oas'),
CSP(15, log=False), TangentSpace('logeuclid'),
LogisticRegression('l2'))
cv = StratifiedKFold(n_splits=n_splits, shuffle=True, random_state=4343)
y = np.asarray(y)
scores = []
for train_index, test_index in cv.split(X, y):
print(train_index)
print(test_index)
print('we are in the CV loop')
X_train, X_test = X[train_index], X[test_index]
y_train, y_test = y[train_index], y[test_index]
# Train on X_train, y_train
clf.fit(X_train, y_train)
# Predict the category on X_test
y_pred = clf.predict(X_test)
scores.append(accuracy_score(y_true=y_test, y_pred=y_pred))
scores = np.asarray(scores)
if plot_conf_matrix == 1:
X_train, X_test, y_train, y_test = train_test_split(
X, y, test_size=test_size, random_state=7, stratify=y)
print('train and test have been split')
y_pred = clf.fit(X_train, y_train).predict(X_test)
# Compute confusion matrix
cnf_matrix = confusion_matrix(y_test, y_pred)
np.set_printoptions(precision=2)
print(cnf_matrix)
# Plot non-normalized confusion matrix
plt.figure()
plot_confusion_matrix(cnf_matrix,
classes=class_names,
title='Confusion matrix, without normalization')
# Plot normalized confusion matrix
plt.figure()
plot_confusion_matrix(cnf_matrix,
classes=class_names,
normalize=True,
title='Normalized confusion matrix')
plt.show()
return scores, y_test, y_pred, cnf_matrix
return scores, y_test, y_pred, cnf_matrix
print('sample drop %: ', (1 - len(epochs.events) / len(events)) * 100)
epochs
###################################################################################################
# Run classification
# ----------------------------
clfs = OrderedDict()
clfs['Vect + LR'] = make_pipeline(Vectorizer(), StandardScaler(),
LogisticRegression())
clfs['Vect + RegLDA'] = make_pipeline(Vectorizer(),
LDA(shrinkage='auto', solver='eigen'))
clfs['ERPCov + TS'] = make_pipeline(ERPCovariances(estimator='oas'),
TangentSpace(), LogisticRegression())
clfs['ERPCov + MDM'] = make_pipeline(ERPCovariances(estimator='oas'), MDM())
clfs['XdawnCov + TS'] = make_pipeline(XdawnCovariances(estimator='oas'),
TangentSpace(), LogisticRegression())
clfs['XdawnCov + MDM'] = make_pipeline(XdawnCovariances(estimator='oas'),
MDM())
# format data
epochs.pick_types(eeg=True)
X = epochs.get_data() * 1e6
times = epochs.times
y = epochs.events[:, -1]
# define cross validation
cv = StratifiedShuffleSplit(n_splits=20, test_size=0.25, random_state=42)
# run cross validation for each pipeline
auc = []
tmin,
tmax,
proj=False,
picks=picks,
baseline=None,
preload=True,
verbose=False)
X = epochs.get_data()
y = epochs.events[:, -1]
###############################################################################
# Embedding the Xdawn covariance matrices with Laplacian Eigenmaps
nfilter = 4
xdwn = XdawnCovariances(estimator='scm', nfilter=nfilter)
split = train_test_split(X, y, train_size=0.25, random_state=42)
Xtrain, Xtest, ytrain, ytest = split
covs = xdwn.fit(Xtrain, ytrain).transform(Xtest)
lapl = Embedding(metric='riemann', n_components=2)
embd = lapl.fit_transform(covs)
###############################################################################
# Plot the three first components of the embedded points
fig, ax = plt.subplots(figsize=(7, 8), facecolor='white')
for cond, label in event_id.items():
idx = (ytest == label)
ax.scatter(embd[idx, 0], embd[idx, 1], s=36, label=cond)
def _init(self):
self.n_components = self.params.get('n_components', 3)
self.pipeline = make_pipeline(XdawnCovariances(
self.n_components), TangentSpace(metric='riemann'), LogisticRegression())
def test_Xdawncovariances():
"""Test fit ERPCovariances"""
x = np.random.randn(10, 3, 100)
labels = np.array([0, 1]).repeat(5)
cov = XdawnCovariances()
cov.fit_transform(x, labels)
def decode(epochs,
get_y_label_func,
epoch_filter=None,
decoding_method='standard',
sliding_window_size=None,
sliding_window_step=None,
n_jobs=multiprocessing.cpu_count(),
equalize_event_counts=True,
only_fit=False,
generalize_across_time=True):
"""
Basic flow for decoding
"""
config = dict(equalize_event_counts=equalize_event_counts,
only_fit=only_fit,
sliding_window_size=sliding_window_size,
sliding_window_step=sliding_window_step,
decoding_method=decoding_method,
generalize_across_time=generalize_across_time,
epoch_filter=str(epoch_filter))
if epoch_filter is not None:
epochs = epochs[epoch_filter]
#-- Classify epochs into groups (training epochs)
y_labels = get_y_label_func(epochs)
if equalize_event_counts:
epochs.events[:, 2] = y_labels
epochs.event_id = {str(label): label for label in np.unique(y_labels)}
min_n_items_per_y_label = min(
[len(epochs[cond]) for cond in epochs.event_id.keys()])
print("\nEqualizing the number of epochs to %d per condition..." %
min_n_items_per_y_label)
epochs.equalize_event_counts(epochs.event_id.keys())
y_labels = epochs.events[:, 2]
print("The epochs were classified into %d groups:" % len(set(y_labels)))
for g in set(y_labels):
print("Group {:}: {:} epochs".format(g, sum(np.array(y_labels) == g)))
#-- Create the decoding pipeline
print("Creating the classification pipeline...")
epochs_data = epochs.get_data()
preprocess_pipeline = None
if decoding_method.startswith('standard'):
if 'reg' in decoding_method:
clf = make_pipeline(StandardScaler(), Ridge())
else:
clf = make_pipeline(
StandardScaler(),
svm.SVC(C=1, kernel='linear', class_weight='balanced'))
if 'raw' not in decoding_method:
assert sliding_window_size is not None
assert sliding_window_step is not None
preprocess_pipeline = \
make_pipeline(umne.transformers.SlidingWindow(window_size=sliding_window_size, step=sliding_window_step, average=True))
elif decoding_method == 'ERP_cov':
clf = make_pipeline(
UnsupervisedSpatialFilter(PCA(20), average=False),
ERPCovariances(
estimator='lwf'), # todo how to apply sliding window?
CSP(30, log=False),
TangentSpace('logeuclid'),
LogisticRegression('l2')) # todo why logistic regression?
elif decoding_method == 'Xdawn_cov':
clf = make_pipeline(
UnsupervisedSpatialFilter(PCA(50), average=False),
XdawnCovariances(12, estimator='lwf', xdawn_estimator='lwf'),
TangentSpace('logeuclid'), LogisticRegression('l2'))
elif decoding_method == 'Hankel_cov':
clf = make_pipeline(
UnsupervisedSpatialFilter(PCA(70), average=False),
HankelCovariances(delays=[1, 8, 12, 64], estimator='oas'),
CSP(15, log=False), TangentSpace('logeuclid'),
LogisticRegression('l2'))
else:
raise Exception('Unknown decoding method: {:}'.format(decoding_method))
print('\nDecoding pipeline:')
for i in range(len(clf.steps)):
print('Step #{:}: {:}'.format(i + 1, clf.steps[i][1]))
if preprocess_pipeline is not None:
print('\nApplying the pre-processing pipeline:')
for i in range(len(preprocess_pipeline.steps)):
print('Step #{:}: {:}'.format(i + 1,
preprocess_pipeline.steps[i][1]))
epochs_data = preprocess_pipeline.fit_transform(epochs_data)
if only_fit:
#-- Only fit the decoders
procedure = 'only_fit'
scores = None
cv = None
if decoding_method.startswith('standard'):
if 'reg' in decoding_method:
if 'r2' in decoding_method:
scoring = metrics.make_scorer(metrics.r2_score)
else:
scoring = metrics.make_scorer(metrics.mean_squared_error)
else:
scoring = 'accuracy'
if generalize_across_time:
estimator = GeneralizingEstimator(clf,
scoring=scoring,
n_jobs=n_jobs)
else:
estimator = SlidingEstimator(clf,
scoring=scoring,
n_jobs=n_jobs)
else:
estimator = clf
estimator.fit(X=epochs_data, y=y_labels)
else:
#-- Classify & score -- cross-validation
procedure = 'fit_and_score'
print(
"\nCreating a classifier and calculating accuracy scores (this may take some time)..."
)
cv = StratifiedKFold(n_splits=5)
if decoding_method.startswith('standard'):
if 'reg' in decoding_method:
if 'r2' in decoding_method:
scoring = metrics.make_scorer(metrics.r2_score)
else:
scoring = metrics.make_scorer(metrics.mean_squared_error)
else:
scoring = 'accuracy'
if generalize_across_time:
estimator = GeneralizingEstimator(clf,
scoring=scoring,
n_jobs=n_jobs)
else:
estimator = SlidingEstimator(clf,
scoring=scoring,
n_jobs=n_jobs)
scores = cross_val_multiscore(estimator=estimator,
X=epochs_data,
y=np.array(y_labels),
cv=cv)
else:
scores = _run_cross_validation(X=epochs_data,
y=np.array(y_labels),
clf=clf,
cv=cv)
estimator = 'None' # Estimator is not defined in the case of Riemannian decoding
times = np.linspace(epochs.tmin, epochs.tmax, epochs_data.shape[2])
return dict(procedure=procedure,
estimator=estimator,
scores=scores,
pipeline=clf,
preprocess=preprocess_pipeline,
cv=cv,
times=times,
config=config)
def __run_strat_validation_RG(
self,
epochs,
n_strat_folds=5,
shuffle=False,
random_state=42,
RG_Pipeline_Num=0,
estimator='lwf',
class_names=['Rest', '13 Hz', '17 Hz', '21 Hz'],
accuracy_threshold=0.7):
"""
Complete a stratified cross-validation using Riemannian Geometery pipeline.
Parameters
----------
epochs : Epoch Object from MNE
Epoch data held in an appropriate MNE format. This could be derived from
mne.Epochs, or using the `build_epochs` command included in this script.
n_strat_folds : int, optional
Number of folds for the stratified K-Fold cross-validation.
This value should be chosen carefully to avoid unbalanced classes.
The default is 5.
shuffle : bool, optional
Shuffle training set data. See sklearn.model_selection.StratifiedKFold
for more details.
The default is False.
random_state : int, optional
The value to be used as the 'seed' for `numpy.random.RandomState`.
See sklearn.model_selection.StratifiedKFold for more details.
The default is 42.
RG_Pipeline_Num : int, optional
Which pre-defined Riemannian Geometery pipeline to run for analysis.
Can be 0,1,2,3:
Pipeline 0:
Covariance w/ estimator -> Riemannian KNN
Pipeline 1:
Covariance w/ estimator -> CSP -> TangentSpace -> LogisticRegression
LogReg uses a 'balanced' option for class weights, l2 penalty.
Pipeline 2:
XDawnCovariance w/ estimator -> TangentSpace -> LogisticRegression
LogReg uses elasticnet penalty, solver soga and a multinominal multi_class flag.
Pipeline 3:
Covariance w/ estimator -> MDM.
Minimum distance to mean (MDM) is the main classification scheme.
The default is 0.
estimator : str, optional
Covariance matrix estimator to use. For regularization consider 'lwf'
or 'oas'. For complete lists, see pyriemann.utils.covariance.
The default is 'lwf'.
class_names : List, optional
List of names for the confusion matrix plot.
The default is ['Rest','13 Hz','17 Hz','21 Hz'].
accuracy_threshold : float, optional
Threshold for determining which folds are 'good' fits. Accuracy found
above the threshold (e.g. 70% or greater) will be reported as good fit
folds.
The default is 0.7.
Returns
-------
DICT
Dictionary of outputs are returned for the user.
In order:
Fold accuracy -'Fold Acc'
Indices for `good` training folds > or = to accuracy_threshold value - 'Good Train Ind'
Indices for `good` test folds > or = to given accuracy_threshold value - 'Good Test Ind'
Indices for `bad` train folds < given accuracy_threshold value - 'Bad Train Ind'
Indices for `bad` test folds < given accuracy_threshold value - 'Bad Test Ind'
List of predicted classes from the RG Pipeline - 'Prediction List'
List of true classes from the RG Pipeline - 'True Class List'
See Also
--------
mne.Epochs
sklearn.model_selection.StratifiedKFold
sklearn.linear_model.LogisticRegression
pyriemann.estimation.Covariances
pyriemann.estimation.XdawnCovariances
pyriemann.spatialfilters.CSP
pyriemann.tangentspace.TangentSpace
pyriemann.classification.MDM
pyriemann.classification.KNearestNeighbor (riemmanian KNN)
"""
# Set the stratified CV model
cv_strat = StratifiedKFold(
n_splits=n_strat_folds, shuffle=True, random_state=random_state
) # Requires us to input in the ylabels as well...need to figure out how to get this.
# Run one of the pre-defined pipelines
if RG_Pipeline_Num == 1:
clf = make_pipeline(
Covariances(estimator=estimator), CSP(log=False),
TangentSpace(),
LogisticRegression(class_weight='balanced', max_iter=500))
elif RG_Pipeline_Num == 2:
clf = make_pipeline(
XdawnCovariances(estimator=estimator,
xdawn_estimator=estimator), TangentSpace(),
LogisticRegression(penalty='elasticnet',
class_weight=None,
solver='saga',
multi_class='multinomial',
l1_ratio=0.5,
max_iter=500))
elif RG_Pipeline_Num == 3:
clf = make_pipeline(Covariances(estimator=estimator),
MDM()) # This is the best so far
else:
print(
"...Running a default pipeline for RG using Covariance, and KNN..."
)
clf = make_pipeline(Covariances(estimator=estimator), riem_KNN())
# Get the labels for the data
labels = epochs.events[:, -1]
# Identify the data itself
X_data = epochs.get_data()
# Get the class names for the confusion matrix
class_names = class_names
# Make empty lists for each item in the stratified CV
acc_list = []
preds_list = []
true_class_list = []
good_train_indx = []
good_test_indx = []
bad_train_indx = []
bad_test_indx = []
# For loop testing each iteration of the stratified cross-validation
for train_idx, test_idx in cv_strat.split(X_data, labels):
# Get the x_train and x_test data for this fold
x_train, x_test = X_data[train_idx], X_data[test_idx]
# Get the y_train and y_test data for this fold
y_train, y_test = labels[train_idx], labels[test_idx]
# Fit the classifier
clf.fit(x_train, y_train)
# Find the predicted value on the test data in this fold
preds = clf.predict(x_test)
# Save in list
preds_list.append(preds)
# Save the true class labels in a list for this fold
true_class_list.append(y_test)
# Find the accuracy on average from this prediction
acc_mean = np.average(preds == y_test)
# Save the accuracy to a list
acc_list.append(acc_mean)
# Find out where the 'Good' training folds are. (Greater than threshold)
if acc_mean >= accuracy_threshold:
print(
"Train indices above accuracy threshold of " +
str(accuracy_threshold * 100) + "% are: ", train_idx)
print(
"Test indices above accuracy threshold of " +
str(accuracy_threshold * 100) + "% are: ", test_idx)
good_train_indx.append(train_idx)
good_test_indx.append(test_idx)
# Find out where the 'Bad' training folds are. (Less than threshold)
else:
bad_train_indx.append(train_idx)
bad_test_indx.append(test_idx)
# Make a plot for the confusion matrix
fig = plt.figure()
plot_confusion_matrix(y_test, preds, class_names)
# Print out the final results from across all folds on average
print(
"The overall accuracy with " + str(n_strat_folds) +
"-fold stratified CV was: ", np.average(acc_list))
# Return output vals
return dict({
'Fold Acc': acc_list,
'Good Train Ind': good_train_indx,
'Good Test Ind': good_test_indx,
'Bad Train Ind': bad_train_indx,
'Bad Test Ind': bad_test_indx,
'Prediction List': preds_list,
'True Class List': true_class_list
})
def __train_predefined_classifier(
self,
epochs,
RG_Pipeline_Num=0,
estimator='lwf',
estimate_accuracy=False,
random_state=44,
class_names=['Rest', '13 Hz', '17 Hz', '21 Hz']):
"""
Train a predefined Riemannian Geometery pipeline on a single dataset using
MNE and pyriemann.
Parameters
----------
epochs : Epoch Object from MNE
Epoch data held in an appropriate MNE format. This could be derived from
mne.Epochs, or using the `build_epochs` command included in this script.
RG_Pipeline_Num :int, optional
Which pre-defined Riemannian Geometery pipeline to run for analysis.
Can be 0,1,2,3:
Pipeline 0:
Covariance w/ estimator -> Riemannian KNN
Pipeline 1:
Covariance w/ estimator -> CSP -> TangentSpace -> LogisticRegression
LogReg uses a 'balanced' option for class weights, l2 penalty.
Pipeline 2:
XDawnCovariance w/ estimator -> TangentSpace -> LogisticRegression
LogReg uses elasticnet penalty, solver soga and a multinominal multi_class flag.
Pipeline 3:
Covariance w/ estimator -> MDM.
Minimum distance to mean (MDM) is the main classification scheme.
The default is 0.
estimator : str, optional
Covariance matrix estimator to use. For regularization consider 'lwf'
or 'oas'. For complete lists, see pyriemann.utils.covariance.
The default is 'lwf'.
estimate_accuracy : bool, optional
Estimate model accuracy roughly using a simple data-hold out train/test split.
A default hold out of 75/25% train, test respectively is used.
The default is False.
random_state : int, optional
The value to be used as the 'seed' for `numpy.random.RandomState`.
See sklearn.model_selection.StratifiedKFold for more details.
The default is 42.
class_names : List, optional
List of names for the confusion matrix plot.
The default is ['Rest','13 Hz','17 Hz','21 Hz'].
Returns
-------
clf : Classifier object (sklearn)
Returns a trained classifier object based on the given epoch data and
Riemannian Geometry pipeline.
See Also
--------
mne.Epochs
sklearn.model_selection.StratifiedKFold
sklearn.linear_model.LogisticRegression
pyriemann.estimation.Covariances
pyriemann.estimation.XdawnCovariances
pyriemann.spatialfilters.CSP
pyriemann.tangentspace.TangentSpace
pyriemann.classification.MDM
pyriemann.classification.KNearestNeighbor (riemmanian KNN)
"""
# Run one of the pre-defined pipelines
if RG_Pipeline_Num == 1:
clf = make_pipeline(
Covariances(estimator=estimator), CSP(log=False),
TangentSpace(),
LogisticRegression(class_weight='balanced', max_iter=500))
elif RG_Pipeline_Num == 2:
clf = make_pipeline(
XdawnCovariances(estimator=estimator,
xdawn_estimator=estimator), TangentSpace(),
LogisticRegression(l1,
class_weight=None,
solver='saga',
multi_class='multinomial',
max_iter=500))
elif RG_Pipeline_Num == 3:
clf = make_pipeline(Covariances(estimator=estimator),
MDM()) # This is the best so far
else:
print(
"...Running a default pipeline for RG using Covariance, and KNN..."
)
clf = make_pipeline(Covariances(estimator=estimator), riem_KNN())
# Get the labels for the data
labels = epochs.events[:, -1]
# Identify the data itself
X_data = epochs.get_data()
# Get the class names for the confusion matrix
class_names = class_names
# This is NOT a great measure of the model accuracy. This just will give you
# a rough estimate of how it is performing within its own dataset. This
# should be used sparingly!
if estimate_accuracy is True:
# Do a simple data-hold out for testing
x_train, x_test, y_train, y_test = train_test_split(
X_data, labels, test_size=0.25, random_state=random_state)
clf_estimate = clf
clf_estimate.fit(x_train, y_train)
pred_vals = clf_estimate.predict(x_test)
accuracy_val = np.mean(pred_vals == y_test)
fig = plt.figure()
plot_confusion_matrix(y_test, pred_vals, class_names)
# Fit the data to the given epoch information
clf.fit(X_data, labels)
return clf
labels = epochs.events[:, -1]
evoked = epochs.average()
###############################################################################
# Decoding in sensor space using a linear SVM
n_components = 3 # pick some components
# Define a monte-carlo cross-validation generator (reduce variance):
cv = KFold(len(labels), 10, shuffle=True, random_state=42)
pr = np.zeros(len(labels))
epochs_data = epochs.get_data()
print('Multiclass classification with XDAWN + MDM')
clf = Pipeline([('COV', XdawnCovariances(n_components)), ('MDM', MDM())])
for train_idx, test_idx in cv:
y_train, y_test = labels[train_idx], labels[test_idx]
clf.fit(epochs_data[train_idx], y_train)
pr[test_idx] = clf.predict(epochs_data[test_idx])
print classification_report(labels, pr)
print confusion_matrix(labels, pr)
print('Multiclass classification with XDAWN + FgMDM')
clf = Pipeline([('COV', XdawnCovariances(n_components)), ('MDM', FgMDM())])
for train_idx, test_idx in cv:
y_train, y_test = labels[train_idx], labels[test_idx]
##############################################################################
# Create pipelines
# ----------------
#
# Pipelines must be a dict of sklearn pipeline transformer.
processing_sampling_rate = 128
pipelines = {}
# we have to do this because the classes are called 'Target' and 'NonTarget'
# but the evaluation function uses a LabelEncoder, transforming them
# to 0 and 1
labels_dict = {"Target": 1, "NonTarget": 0}
# Riemannian geometry based classification
pipelines["RG+LDA"] = make_pipeline(
XdawnCovariances(nfilter=5, estimator="lwf", xdawn_estimator="scm"),
TangentSpace(),
LDA(solver="lsqr", shrinkage="auto"),
)
# Simple LDA pipeline using averaged feature values in certain time intervals
jumping_mean_ivals = [
[0.10, 0.139],
[0.14, 0.169],
[0.17, 0.199],
[0.20, 0.229],
[0.23, 0.269],
[0.27, 0.299],
[0.30, 0.349],
[0.35, 0.409],
[0.41, 0.449],
##############################################################################
# Create pipelines
# ----------------
#
# Pipelines must be a dict of sklearn pipeline transformer.
pipelines = {}
# we have to do this because the classes are called 'Target' and 'NonTarget'
# but the evaluation function uses a LabelEncoder, transforming them
# to 0 and 1
labels_dict = {"Target": 1, "NonTarget": 0}
pipelines["RG + LDA"] = make_pipeline(
XdawnCovariances(nfilter=2,
classes=[labels_dict["Target"]],
estimator="lwf",
xdawn_estimator="lwf"),
TangentSpace(),
LDA(solver="lsqr", shrinkage="auto"),
)
pipelines["Xdw + LDA"] = make_pipeline(Xdawn(nfilter=2, estimator="lwf"),
Vectorizer(),
LDA(solver="lsqr", shrinkage="auto"))
##############################################################################
# Evaluation
# ----------
#
# We define the paradigm (P300) and use all three datasets available for it.
# The evaluation will return a dataframe containing a single AUC score for
loss_ax.set_ylabel('loss')
acc_ax.set_ylabel('accuray')
loss_ax.legend(loc='upper left')
acc_ax.legend(loc='lower left')
plt.show()
############################# PyRiemann Portion ##############################
# code is taken from PyRiemann's ERP sample script, which is decoding in
# the tangent space with a logistic regression
n_components = 2 # pick some components
# set up sklearn pipeline
clf = make_pipeline(XdawnCovariances(n_components),
TangentSpace(metric='riemann'), LogisticRegression())
preds_rg = np.zeros(len(Y_test))
# reshape back to (trials, channels, samples)
X_train = X_train.reshape(X_train.shape[0], chans, samples)
X_test = X_test.reshape(X_test.shape[0], chans, samples)
# train a classifier with xDAWN spatial filtering + Riemannian Geometry (RG)
# labels need to be back in single-column format
history = clf.fit(X_train, Y_train.argmax(axis=-1))
preds_rg = clf.predict(X_test)
# Printing the results
acc2 = np.mean(preds_rg == Y_test.argmax(axis=-1))
return np.reshape(X, (X.shape[0], -1))
##############################################################################
# Create pipelines
# ----------------
# Pipelines must be a dict of sklearn pipeline transformer.
pipelines = {}
# we have to do this because the classes are called 'Target' and 'NonTarget'
# but the evaluation function uses a LabelEncoder, transforming them
# to 0 and 1
labels_dict = {'Target': 1, 'NonTarget': 0}
pipelines['RG + LDA'] = make_pipeline(
XdawnCovariances(
nfilter=2,
classes=[
labels_dict['Target']],
estimator='lwf',
xdawn_estimator='lwf'),
TangentSpace(),
LDA(solver='lsqr', shrinkage='auto'))
pipelines['Xdw + LDA'] = make_pipeline(Xdawn(nfilter=2, estimator='lwf'),
Vectorizer(), LDA(solver='lsqr',
shrinkage='auto'))
pipelines['ERPCov + TS'] = make_pipeline(ERPCovariances(classes=[0, 1], estimator='oas', svd=None),
TangentSpace(metric='riemann'),
LogisticRegression(solver='lbfgs'))
_, x_test, y_test = load(data_name,
s_id[s],
npp_params,
clean=True,
physical=True,
downsample=False)
data_size = y_train.shape[0]
shuffle_index = utils.shuffle_data(data_size)
x_train = x_train[shuffle_index]
x_train = np.squeeze(x_train)
y_train = y_train[shuffle_index]
# Build Model
xd = XdawnCovariances(nfilter=5,
applyfilters=True,
estimator='lwf')
# es = ElectrodeSelection(nelec=25, metric='riemann')
ts = TangentSpace(metric='logeuclid')
lr = LogisticRegression(solver='liblinear', max_iter=200, C=0.01)
model = Pipeline([('xDAWN', xd), ('TangentSpace', ts), ('LR', lr)])
model.fit(x_train, y_train)
# Test Model
y_pred = np.argmax(model.predict_proba(np.squeeze(x_test)), axis=1)
bca = utils.bca(y_test, y_pred)
acc = np.sum(y_pred == y_test).astype(np.float32) / len(y_pred)
print('{}: acc-{} bca-{}'.format(data_name, acc, bca))
)
labels = epochs.events[:, -1]
evoked = epochs.average()
###############################################################################
# Decoding in tangent space with a logistic regression
n_components = 2 # pick some components
# Define a monte-carlo cross-validation generator (reduce variance):
cv = KFold(n_splits=10, shuffle=True, random_state=42)
epochs_data = epochs.get_data()
clf = make_pipeline(
XdawnCovariances(n_components),
TangentSpace(metric="riemann"),
LogisticRegression(),
)
preds = np.zeros(len(labels))
for train_idx, test_idx in cv.split(epochs_data):
y_train, y_test = labels[train_idx], labels[test_idx]
clf.fit(epochs_data[train_idx], y_train)
preds[test_idx] = clf.predict(epochs_data[test_idx])
# Printing the results
acc = np.mean(preds == labels)
print("Classification accuracy: %f " % (acc))
