Filter bank CSP with Riemannian weighting for disability-centric motor imagery brain computer interface.

Code ↔ Paper

1 match between paragraphs of the paper and lines of its authors' code, computed by the harvester (lexical-v1). Click a colored paragraph or line to see its counterpart.

The 1 match

[1] § Materials and methods › Multi-classification ↔ BCI_R-FBCSP.ipynb, lines 1066 to 1125 · score 0.51 · Random Forest Classifier, Linear Discriminant, Layer, MLP, LDA · Report a mismatch

Paper

Loaded from Europe PMC by your browser; OSCR does not store it: doi.org · Europe PMC

The paper is loaded when this pane is shown.

The paper is loaded by your browser, with JavaScript: read it at doi.org or on Europe PMC.

The authors' code

Jupyter notebook · 1,259 lines · 47 KB · CC-BY-4.0 · 1 match

# %%
from scipy import signal
from scipy.signal import firwin, lfilter
import mne
import numpy as np
from sklearn.metrics import classification_report, accuracy_score
from sklearn.preprocessing import StandardScaler
from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
from scipy.signal import butter, sosfiltfilt, filtfilt
from sklearn.metrics import accuracy_score, cohen_kappa_score, precision_score, recall_score
from mne.decoding import CSP
from scipy.signal import welch
from sklearn.preprocessing import StandardScaler
import logging
import scipy.io as io
import os
# %%
class IV2aGdfReader:
stimcodes = ('769', '770', '771', '772')
channels_to_remove = ['EOG-left', 'EOG-central', 'EOG-right']
def __init__(self, data_dir: str=None):
self.data_dir = data_dir
if data_dir is not None:
self.filenames = [name for name in os.listdir(
self.data_dir) if name.endswith(".gdf")]
else:
self.filenames = []
def load_raw_data_gdf(self, filename: str):
if self.data_dir is not None:
filepath = os.path.join(self.data_dir, filename)
else:
filepath = filename
logging.info(f"Read {filepath}")
return mne.io.read_raw_gdf(filepath, verbose="ERROR")
def read_file(self, filename: str, tmin=0.5, tmax=3.0, baseline=None, matfile=None): #tmin=0.5, tmax=2.5
gdf_name = os.path.basename(filename).split(".")[0]
subject = gdf_name[1:3]
raw_data = self.load_raw_data_gdf(filename)
sample_freq = raw_data.info.get('sfreq')
events, event_ids = mne.events_from_annotations(raw_data)
if "T" == gdf_name[-1]:
stims = [value for key, value in event_ids.items()
if key in self.stimcodes]
elif "E" == gdf_name[-1]:
stims = [value for key, value in event_ids.items() if key == "783"]
else:
all_stimcodes = self.stimcodes + ["783"]
stims = [value for key, value in event_ids.items() if key in all_stimcodes]
epochs = mne.Epochs(raw_data, events, event_id=stims, tmin=tmin, tmax=tmax,
baseline=baseline, preload=True, proj=False)
epochs = epochs.drop_channels(self.channels_to_remove)
# Récupérer les noms et indices des canaux EEG restants
ch_names = epochs.ch_names # Liste des noms de canaux
ch_indices = {name: idx for idx, name in enumerate(ch_names)}
if matfile is not None:
mat = io.loadmat(matfile)
y_labels = mat["classlabel"].reshape(-1) - 1
elif self.data_dir is not None:
mat = io.loadmat(os.path.join(
self.data_dir, "true_labels", f"{gdf_name}.mat"))
y_labels = mat["classlabel"].reshape(-1) - 1
else:
y_labels = None
x_data = epochs.get_data()#*1e6
eeg_data = {'x_data': x_data,
'y_labels': y_labels,
'fs': sample_freq,
'ch_names':ch_names,
'ch_indices':ch_indices,
"subject": subject}
return eeg_data
# %%
base_path ="C:/Users/Client/Desktop/Jihen_BCI/BCI_IV_DSIIa"
reader = IV2aGdfReader(data_dir=base_path)
# %%
%%capture
# Liste des fichiers à traiter
file_extensions = ['T', 'E']
base_filename = "A{:02d}{}.gdf"
# Initialiser les listes pour stocker toutes les données et étiquettes .T et .E
all_x_data_T = []
all_x_data_E = []
all_y_labels_T = []
all_y_labels_E = []
all_ch_namesT = []
all_ch_indicesT = []
all_ch_namesE = []
all_ch_indicesE = []
# Boucle sur les extensions de fichiers
for ext in file_extensions:
# Boucle sur les fichiers (9 fichiers au total)
for i in range(1, 10):
filename = os.path.join(base_path, base_filename.format(i, ext))
data = reader.read_file(filename)
# Accéder aux données et aux étiquettes
x_data = data['x_data']
y_labels = data['y_labels']
ch_names = data['ch_names']
ch_indices = data['ch_indices']
# Ajouter les données et étiquettes aux listes correspondantes
if ext == 'T':
all_x_data_T.append(x_data)
all_y_labels_T.append(y_labels)
all_ch_namesT .append(ch_names)
all_ch_indicesT .append(ch_indices)
elif ext == 'E':
all_x_data_E.append(x_data)
all_y_labels_E.append(y_labels)
all_ch_namesE .append(ch_names)
all_ch_indicesE .append(ch_indices)
# Imprimer la taille des données et des étiquettes pour chaque fichier
print(f"Taille des données ({filename}):", x_data.shape)
print(f"Taille des étiquettes ({filename}):", y_labels.shape)
# Convertir les listes en tableaux numpy
all_x_data_T = np.array(all_x_data_T)
all_y_labels_T = np.array(all_y_labels_T)
all_x_data_E = np.array(all_x_data_E)
all_y_labels_E = np.array(all_y_labels_E)
# %%
# Imprimer la taille totale des données et des étiquettes pour chaque catégorie
print("Taille totale des données .T (all_x_data_T):", all_x_data_T.shape)
print("Taille totale des étiquettes .T (all_y_labels_T):", all_y_labels_T.shape)
print("Taille totale des données .E (all_x_data_E):", all_x_data_E.shape)
print("Taille totale des étiquettes .E (all_y_labels_E):", all_y_labels_E.shape)
# %%
def my_filter_function( data, lowcut, highcut, fs, order):
nyquist = 0.5 * fs
low = lowcut / nyquist
high = highcut / nyquist
sos = butter(order, [low, high], btype='band', output='sos')
# Increase the padlen parameter to increase the filter window size
filtered_data = sosfiltfilt(sos, data, axis=-1, padlen=250)
return filtered_data
# %% [markdown]
# # fonction R-FBCSP
# %%
import numpy as np
from pyriemann.utils.mean import mean_riemann
from scipy.linalg import eigh
def CSP_OVR_optim(epochs, labels, n_components=3):
n_trials, n_channels, _ = epochs.shape
classes = np.unique(labels)
all_features = []
all_W = []
# 1. Calcul des matrices de covariance normalisées
covmats = []
for epoch in epochs:
cov = epoch @ epoch.T
trace = np.trace(cov)
cov_norm = cov / (trace + 1e-10)
covmats.append(cov_norm)
covmats = np.array(covmats)
for c in classes:
y_binary = (labels == c).astype(int)
# 2. Moyennes riemanniennes
cov_A = mean_riemann(covmats[y_binary == 1])
cov_B = mean_riemann(covmats[y_binary == 0])
# 3. Blanchiment robuste
cov_sum = cov_A + cov_B
eigvals, eigvecs = eigh(cov_sum)
eigvals = np.clip(eigvals, 1e-6, None)
P = eigvecs @ np.diag(1.0 / np.sqrt(eigvals)) @ eigvecs.T
# 4. Diagonalisation
S_A = P @ cov_A @ P.T
eigvals_s, eigvecs_s = eigh(S_A)
# 5. Tri des vecteurs
idx = np.argsort(eigvals_s)[::-1]
W = eigvecs_s[:, idx].T @ P
W_c = np.vstack([W[:n_components], W[-n_components:]])
all_W.append(W_c)
# 6. Extraction des log-variances
features = []
for epoch in epochs:
Z = W_c @ epoch
var = np.var(Z, axis=1)
var_sum = np.sum(var)
log_var = np.log((var + 1e-10) / (var_sum + 1e-10))
features.append(log_var)
all_features.append(np.array(features))
# 7. Concatenation des features de tous les classifieurs OVR
final_features = np.concatenate(all_features, axis=1)
return final_features, all_W
# %% [markdown]
# # Global filtering
# %%
order_filter = 6
lowcut = 8.0
highcut = 30.0
fs = 250
n_components_csp = 22
# Train
all_person_featuresT = []
for person_index, (x_data_T, y_labels_T) in enumerate(zip(all_x_data_T, all_y_labels_T)):
filtered_dataT = my_filter_function(x_data_T, lowcut, highcut, fs, order=order_filter)
features_T, _ = CSP_OVR_optim(filtered_dataT, y_labels_T, n_components=n_components_csp)
all_person_featuresT.append(features_T)
# Test
all_person_featuresE = []
for person_index, (x_data_E, y_labels_E) in enumerate(zip(all_x_data_E, all_y_labels_E)):
filtered_dataE = my_filter_function(x_data_E, lowcut, highcut, fs, order=order_filter)
features_E, _ = CSP_OVR_optim(filtered_dataE, y_labels_E, n_components=n_components_csp)
all_person_featuresE.append(features_E)
# Convertir en numpy arrays
all_person_featuresT = np.array(all_person_featuresT)
all_person_featuresE = np.array(all_person_featuresE)
# %%
print(np.shape(all_person_featuresT))
print(np.shape(all_person_featuresE))
# %% [markdown]
# # LDA Classification
# %%
from sklearn.model_selection import KFold
from sklearn.preprocessing import StandardScaler
from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score
from sklearn.pipeline import make_pipeline
import numpy as np
# Initialisation des listes
cv_train_accuracies, cv_val_accuracies, cv_test_accuracies = [], [], []
cv_train_precisions, cv_val_precisions, cv_test_precisions = [], [], []
cv_train_recalls, cv_val_recalls, cv_test_recalls = [], [], []
cv_train_f1s, cv_val_f1s, cv_test_f1s = [], [], []
kfold = KFold(n_splits=5, shuffle=True, random_state=42)
for i, (features_train, y_labels_T, features_test, y_labels_E) in enumerate(zip(all_person_featuresT, all_y_labels_T,
all_person_featuresE, all_y_labels_E)):
print(f"\nProcessing subject {i + 1}")
# Pour accumuler les scores par pli
fold_train_accs, fold_val_accs = [], []
fold_train_prec, fold_val_prec = [], []
fold_train_rec, fold_val_rec = [], []
fold_train_f1, fold_val_f1 = [], []
model = make_pipeline(StandardScaler(),
LinearDiscriminantAnalysis(solver='lsqr', shrinkage='auto',
n_components=3, tol=1e-4))
# Validation croisée
for train_idx, val_idx in kfold.split(features_train):
X_train_fold, X_val_fold = features_train[train_idx], features_train[val_idx]
y_train_fold, y_val_fold = y_labels_T[train_idx], y_labels_T[val_idx]
model.fit(X_train_fold, y_train_fold)
# Prédictions
y_train_pred = model.predict(X_train_fold)
y_val_pred = model.predict(X_val_fold)
# Accuracy
fold_train_accs.append(accuracy_score(y_train_fold, y_train_pred))
fold_val_accs.append(accuracy_score(y_val_fold, y_val_pred))
# Precision, Recall, F1
fold_train_prec.append(precision_score(y_train_fold, y_train_pred, average='weighted', zero_division=0))
fold_val_prec.append(precision_score(y_val_fold, y_val_pred, average='weighted', zero_division=0))
fold_train_rec.append(recall_score(y_train_fold, y_train_pred, average='weighted', zero_division=0))
fold_val_rec.append(recall_score(y_val_fold, y_val_pred, average='weighted', zero_division=0))
fold_train_f1.append(f1_score(y_train_fold, y_train_pred, average='weighted', zero_division=0))
fold_val_f1.append(f1_score(y_val_fold, y_val_pred, average='weighted', zero_division=0))
# Réentraînement sur tout l'ensemble d'apprentissage
model.fit(features_train, y_labels_T)
y_test_pred = model.predict(features_test)
# Test scores
test_acc = accuracy_score(y_labels_E, y_test_pred)
test_prec = precision_score(y_labels_E, y_test_pred, average='weighted', zero_division=0)
test_rec = recall_score(y_labels_E, y_test_pred, average='weighted', zero_division=0)
test_f1 = f1_score(y_labels_E, y_test_pred, average='weighted', zero_division=0)
# Moyennes sur les plis
cv_train_accuracies.append(np.mean(fold_train_accs))
cv_val_accuracies.append(np.mean(fold_val_accs))
cv_test_accuracies.append(test_acc)
cv_train_precisions.append(np.mean(fold_train_prec))
cv_val_precisions.append(np.mean(fold_val_prec))
cv_test_precisions.append(test_prec)
cv_train_recalls.append(np.mean(fold_train_rec))
cv_val_recalls.append(np.mean(fold_val_rec))
cv_test_recalls.append(test_rec)
cv_train_f1s.append(np.mean(fold_train_f1))
cv_val_f1s.append(np.mean(fold_val_f1))
cv_test_f1s.append(test_f1)
# %%
import pandas as pd
# Conversion des scores en pourcentages
test_accuracies_percent = [acc * 100 for acc in cv_test_accuracies]
test_precisions_percent = [prec * 100 for prec in cv_test_precisions]
test_recalls_percent = [rec * 100 for rec in cv_test_recalls]
test_f1_scores_percent = [f1 * 100 for f1 in cv_test_f1s]
# Création d'un DataFrame avec les scores
results_df = pd.DataFrame({
"Subject": [f"Person {i+1}" for i in range(len(test_accuracies_percent))],
"Accuracy (%)": test_accuracies_percent,
"Precision (%)": test_precisions_percent,
"Recall (%)": test_recalls_percent,
"F1-score (%)": test_f1_scores_percent
})
# Arrondir les valeurs à 2 décimales
results_df = results_df.round(2)
# Affichage du tableau
print("\nTest Metrics per Subject:\n")
print(results_df.to_string(index=False))
# Moyennes globales
average_results = results_df.mean(numeric_only=True)
print("\nAverage Test Metrics Across All Subjects:")
print(f"Accuracy: {average_results['Accuracy (%)']:.2f}%")
print(f"Precision: {average_results['Precision (%)']:.2f}%")
print(f"Recall: {average_results['Recall (%)']:.2f}%")
print(f"F1-score: {average_results['F1-score (%)']:.2f}%")
# %% [markdown]
# # RFC Classification
# %%
from sklearn.model_selection import KFold
from sklearn.preprocessing import StandardScaler
from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score
from sklearn.pipeline import make_pipeline
import numpy as np
from sklearn.ensemble import RandomForestClassifier
# Initialisation des listes
cv_train_accuracies, cv_val_accuracies, cv_test_accuracies = [], [], []
cv_train_precisions, cv_val_precisions, cv_test_precisions = [], [], []
cv_train_recalls, cv_val_recalls, cv_test_recalls = [], [], []
cv_train_f1s, cv_val_f1s, cv_test_f1s = [], [], []
kfold = KFold(n_splits=5, shuffle=True, random_state=42)
for i, (features_train, y_labels_T, features_test, y_labels_E) in enumerate(zip(all_person_featuresT, all_y_labels_T,
all_person_featuresE, all_y_labels_E)):
print(f"\nProcessing subject {i + 1}")
# Pour accumuler les scores par pli
fold_train_accs, fold_val_accs = [], []
fold_train_prec, fold_val_prec = [], []
fold_train_rec, fold_val_rec = [], []
fold_train_f1, fold_val_f1 = [], []
model = make_pipeline(StandardScaler(),
RandomForestClassifier(n_estimators=100, class_weight='balanced', random_state=42))
# Validation croisée
for train_idx, val_idx in kfold.split(features_train):
X_train_fold, X_val_fold = features_train[train_idx], features_train[val_idx]
y_train_fold, y_val_fold = y_labels_T[train_idx], y_labels_T[val_idx]
model.fit(X_train_fold, y_train_fold)
# Prédictions
y_train_pred = model.predict(X_train_fold)
y_val_pred = model.predict(X_val_fold)
# Accuracy
fold_train_accs.append(accuracy_score(y_train_fold, y_train_pred))
fold_val_accs.append(accuracy_score(y_val_fold, y_val_pred))
# Precision, Recall, F1
fold_train_prec.append(precision_score(y_train_fold, y_train_pred, average='weighted', zero_division=0))
fold_val_prec.append(precision_score(y_val_fold, y_val_pred, average='weighted', zero_division=0))
fold_train_rec.append(recall_score(y_train_fold, y_train_pred, average='weighted', zero_division=0))
fold_val_rec.append(recall_score(y_val_fold, y_val_pred, average='weighted', zero_division=0))
fold_train_f1.append(f1_score(y_train_fold, y_train_pred, average='weighted', zero_division=0))
fold_val_f1.append(f1_score(y_val_fold, y_val_pred, average='weighted', zero_division=0))
# Réentraînement sur tout l'ensemble d'apprentissage
model.fit(features_train, y_labels_T)
y_test_pred = model.predict(features_test)
# Test scores
test_acc = accuracy_score(y_labels_E, y_test_pred)
test_prec = precision_score(y_labels_E, y_test_pred, average='weighted', zero_division=0)
test_rec = recall_score(y_labels_E, y_test_pred, average='weighted', zero_division=0)
test_f1 = f1_score(y_labels_E, y_test_pred, average='weighted', zero_division=0)
# Moyennes sur les plis
cv_train_accuracies.append(np.mean(fold_train_accs))
cv_val_accuracies.append(np.mean(fold_val_accs))
cv_test_accuracies.append(test_acc)
cv_train_precisions.append(np.mean(fold_train_prec))
cv_val_precisions.append(np.mean(fold_val_prec))
cv_test_precisions.append(test_prec)
cv_train_recalls.append(np.mean(fold_train_rec))
cv_val_recalls.append(np.mean(fold_val_rec))
cv_test_recalls.append(test_rec)
cv_train_f1s.append(np.mean(fold_train_f1))
cv_val_f1s.append(np.mean(fold_val_f1))
cv_test_f1s.append(test_f1)
# %%
import pandas as pd
# Conversion des scores en pourcentages
test_accuracies_percent = [acc * 100 for acc in cv_test_accuracies]
test_precisions_percent = [prec * 100 for prec in cv_test_precisions]
test_recalls_percent = [rec * 100 for rec in cv_test_recalls]
test_f1_scores_percent = [f1 * 100 for f1 in cv_test_f1s]
# Création d'un DataFrame avec les scores
results_df = pd.DataFrame({
"Subject": [f"Person {i+1}" for i in range(len(test_accuracies_percent))],
"Accuracy (%)": test_accuracies_percent,
"Precision (%)": test_precisions_percent,
"Recall (%)": test_recalls_percent,
"F1-score (%)": test_f1_scores_percent
})
# Arrondir les valeurs à 2 décimales
results_df = results_df.round(2)
# Affichage du tableau
print("\nTest Metrics per Subject:\n")
print(results_df.to_string(index=False))
# Moyennes globales
average_results = results_df.mean(numeric_only=True)
print("\nAverage Test Metrics Across All Subjects:")
print(f"Accuracy: {average_results['Accuracy (%)']:.2f}%")
print(f"Precision: {average_results['Precision (%)']:.2f}%")
print(f"Recall: {average_results['Recall (%)']:.2f}%")
print(f"F1-score: {average_results['F1-score (%)']:.2f}%")
# %% [markdown]
# # MLP classification
# %%
from sklearn.model_selection import KFold
from sklearn.preprocessing import StandardScaler
from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score
from sklearn.pipeline import make_pipeline
import numpy as np
from sklearn.ensemble import RandomForestClassifier
from sklearn.neural_network import MLPClassifier
# Initialisation des listes
cv_train_accuracies, cv_val_accuracies, cv_test_accuracies = [], [], []
cv_train_precisions, cv_val_precisions, cv_test_precisions = [], [], []
cv_train_recalls, cv_val_recalls, cv_test_recalls = [], [], []
cv_train_f1s, cv_val_f1s, cv_test_f1s = [], [], []
kfold = KFold(n_splits=5, shuffle=True, random_state=42)
for i, (features_train, y_labels_T, features_test, y_labels_E) in enumerate(zip(all_person_featuresT, all_y_labels_T,
all_person_featuresE, all_y_labels_E)):
print(f"\nProcessing subject {i + 1}")
# Pour accumuler les scores par pli
fold_train_accs, fold_val_accs = [], []
fold_train_prec, fold_val_prec = [], []
fold_train_rec, fold_val_rec = [], []
fold_train_f1, fold_val_f1 = [], []
model = make_pipeline(StandardScaler(),
MLPClassifier(hidden_layer_sizes=(100,), max_iter=300, activation='tanh', solver='adam',
learning_rate_init=0.001, random_state=42 ))
# Validation croisée
for train_idx, val_idx in kfold.split(features_train):
X_train_fold, X_val_fold = features_train[train_idx], features_train[val_idx]
y_train_fold, y_val_fold = y_labels_T[train_idx], y_labels_T[val_idx]
model.fit(X_train_fold, y_train_fold)
# Prédictions
y_train_pred = model.predict(X_train_fold)
y_val_pred = model.predict(X_val_fold)
# Accuracy
fold_train_accs.append(accuracy_score(y_train_fold, y_train_pred))
fold_val_accs.append(accuracy_score(y_val_fold, y_val_pred))
# Precision, Recall, F1
fold_train_prec.append(precision_score(y_train_fold, y_train_pred, average='weighted', zero_division=0))
fold_val_prec.append(precision_score(y_val_fold, y_val_pred, average='weighted', zero_division=0))
fold_train_rec.append(recall_score(y_train_fold, y_train_pred, average='weighted', zero_division=0))
fold_val_rec.append(recall_score(y_val_fold, y_val_pred, average='weighted', zero_division=0))
fold_train_f1.append(f1_score(y_train_fold, y_train_pred, average='weighted', zero_division=0))
fold_val_f1.append(f1_score(y_val_fold, y_val_pred, average='weighted', zero_division=0))
# Réentraînement sur tout l'ensemble d'apprentissage
model.fit(features_train, y_labels_T)
y_test_pred = model.predict(features_test)
# Test scores
test_acc = accuracy_score(y_labels_E, y_test_pred)
test_prec = precision_score(y_labels_E, y_test_pred, average='weighted', zero_division=0)
test_rec = recall_score(y_labels_E, y_test_pred, average='weighted', zero_division=0)
test_f1 = f1_score(y_labels_E, y_test_pred, average='weighted', zero_division=0)
# Moyennes sur les plis
cv_train_accuracies.append(np.mean(fold_train_accs))
cv_val_accuracies.append(np.mean(fold_val_accs))
cv_test_accuracies.append(test_acc)
cv_train_precisions.append(np.mean(fold_train_prec))
cv_val_precisions.append(np.mean(fold_val_prec))
cv_test_precisions.append(test_prec)
cv_train_recalls.append(np.mean(fold_train_rec))
cv_val_recalls.append(np.mean(fold_val_rec))
cv_test_recalls.append(test_rec)
cv_train_f1s.append(np.mean(fold_train_f1))
cv_val_f1s.append(np.mean(fold_val_f1))
cv_test_f1s.append(test_f1)
# %%
import pandas as pd
# Conversion des scores en pourcentages
test_accuracies_percent = [acc * 100 for acc in cv_test_accuracies]
test_precisions_percent = [prec * 100 for prec in cv_test_precisions]
test_recalls_percent = [rec * 100 for rec in cv_test_recalls]
test_f1_scores_percent = [f1 * 100 for f1 in cv_test_f1s]
# Création d'un DataFrame avec les scores
results_df = pd.DataFrame({
"Subject": [f"Person {i+1}" for i in range(len(test_accuracies_percent))],
"Accuracy (%)": test_accuracies_percent,
"Precision (%)": test_precisions_percent,
"Recall (%)": test_recalls_percent,
"F1-score (%)": test_f1_scores_percent
})
# Arrondir les valeurs à 2 décimales
results_df = results_df.round(2)
# Affichage du tableau
print("\nTest Metrics per Subject:\n")
print(results_df.to_string(index=False))
# Moyennes globales
average_results = results_df.mean(numeric_only=True)
print("\nAverage Test Metrics Across All Subjects:")
print(f"Accuracy: {average_results['Accuracy (%)']:.2f}%")
print(f"Precision: {average_results['Precision (%)']:.2f}%")
print(f"Recall: {average_results['Recall (%)']:.2f}%")
print(f"F1-score: {average_results['F1-score (%)']:.2f}%")
# %% [markdown]
# # Multiband filtering
# %%
def process_bands(data, bands, fs, order, n_components, labels):
all_features = []
for band in bands:
lowcut, highcut = band
# Étape 1 : Filtrage des données pour la bande actuelle
filtered_data = my_filter_function(data, lowcut, highcut, fs, order)
# Étape 2 : Extraction des caractéristiques CSP
csp_features, _ = CSP_OVR_optim(filtered_data, labels, n_components)
# Ajouter les caractéristiques de cette bande
all_features.append(csp_features)
# Étape 3 : Combiner les caractéristiques de toutes les bandes
combined_features = np.concatenate(all_features, axis=1)
return combined_features
# %%
# %%capture
all_person_featuresT = []
all_person_featuresE = []
order_filter=6
lowcut =8.0
highcut =30.0
fs=250
n_components_csp=22
bands = [(8, 12), (12, 30), (30, 45)]
i=0
# Loop on each person for Train data
for person_index, (x_data_T, y_labels_T) in enumerate(zip(all_x_data_T, all_y_labels_T)):
i=i+1
print(i)
# CSPPP
CSP_T= process_bands(x_data_T, bands, fs, order_filter, n_components_csp, y_labels_T)
all_person_featuresT.append(CSP_T)
i=0
# Loop on each person for test data
for person_index, (x_data_E, y_labels_E) in enumerate(zip(all_x_data_E, all_y_labels_E)):
i=i+1
print(i)
# CSPPP
CSP_E=process_bands(x_data_E, bands, fs, order_filter, n_components_csp, y_labels_E)
all_person_featuresE.append(CSP_E)
# Convertissez les listes en tableaux numpy
all_person_featuresT = np.array(all_person_featuresT)
all_person_featuresE = np.array(all_person_featuresE)
# %% [markdown]
# # LDA classification
# %%
from sklearn.model_selection import KFold
from sklearn.preprocessing import StandardScaler
from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score
from sklearn.pipeline import make_pipeline
import numpy as np
# Initialisation des listes
cv_train_accuracies, cv_val_accuracies, cv_test_accuracies = [], [], []
cv_train_precisions, cv_val_precisions, cv_test_precisions = [], [], []
cv_train_recalls, cv_val_recalls, cv_test_recalls = [], [], []
cv_train_f1s, cv_val_f1s, cv_test_f1s = [], [], []
kfold = KFold(n_splits=5, shuffle=True, random_state=42)
for i, (features_train, y_labels_T, features_test, y_labels_E) in enumerate(zip(all_person_featuresT, all_y_labels_T,
all_person_featuresE, all_y_labels_E)):
print(f"\nProcessing subject {i + 1}")
# Pour accumuler les scores par pli
fold_train_accs, fold_val_accs = [], []
fold_train_prec, fold_val_prec = [], []
fold_train_rec, fold_val_rec = [], []
fold_train_f1, fold_val_f1 = [], []
model = make_pipeline(StandardScaler(),
LinearDiscriminantAnalysis(solver='lsqr', shrinkage='auto',
n_components=3, tol=1e-4))
# Validation croisée
for train_idx, val_idx in kfold.split(features_train):
X_train_fold, X_val_fold = features_train[train_idx], features_train[val_idx]
y_train_fold, y_val_fold = y_labels_T[train_idx], y_labels_T[val_idx]
model.fit(X_train_fold, y_train_fold)
# Prédictions
y_train_pred = model.predict(X_train_fold)
y_val_pred = model.predict(X_val_fold)
# Accuracy
fold_train_accs.append(accuracy_score(y_train_fold, y_train_pred))
fold_val_accs.append(accuracy_score(y_val_fold, y_val_pred))
# Precision, Recall, F1
fold_train_prec.append(precision_score(y_train_fold, y_train_pred, average='weighted', zero_division=0))
fold_val_prec.append(precision_score(y_val_fold, y_val_pred, average='weighted', zero_division=0))
fold_train_rec.append(recall_score(y_train_fold, y_train_pred, average='weighted', zero_division=0))
fold_val_rec.append(recall_score(y_val_fold, y_val_pred, average='weighted', zero_division=0))
fold_train_f1.append(f1_score(y_train_fold, y_train_pred, average='weighted', zero_division=0))
fold_val_f1.append(f1_score(y_val_fold, y_val_pred, average='weighted', zero_division=0))
# Réentraînement sur tout l'ensemble d'apprentissage
model.fit(features_train, y_labels_T)
y_test_pred = model.predict(features_test)
# Test scores
test_acc = accuracy_score(y_labels_E, y_test_pred)
test_prec = precision_score(y_labels_E, y_test_pred, average='weighted', zero_division=0)
test_rec = recall_score(y_labels_E, y_test_pred, average='weighted', zero_division=0)
test_f1 = f1_score(y_labels_E, y_test_pred, average='weighted', zero_division=0)
# Moyennes sur les plis
cv_train_accuracies.append(np.mean(fold_train_accs))
cv_val_accuracies.append(np.mean(fold_val_accs))
cv_test_accuracies.append(test_acc)
cv_train_precisions.append(np.mean(fold_train_prec))
cv_val_precisions.append(np.mean(fold_val_prec))
cv_test_precisions.append(test_prec)
cv_train_recalls.append(np.mean(fold_train_rec))
cv_val_recalls.append(np.mean(fold_val_rec))
cv_test_recalls.append(test_rec)
cv_train_f1s.append(np.mean(fold_train_f1))
cv_val_f1s.append(np.mean(fold_val_f1))
cv_test_f1s.append(test_f1)
# %%
import pandas as pd
# Conversion des scores en pourcentages
test_accuracies_percent = [acc * 100 for acc in cv_test_accuracies]
test_precisions_percent = [prec * 100 for prec in cv_test_precisions]
test_recalls_percent = [rec * 100 for rec in cv_test_recalls]
test_f1_scores_percent = [f1 * 100 for f1 in cv_test_f1s]
# Création d'un DataFrame avec les scores
results_df = pd.DataFrame({
"Subject": [f"Person {i+1}" for i in range(len(test_accuracies_percent))],
"Accuracy (%)": test_accuracies_percent,
"Precision (%)": test_precisions_percent,
"Recall (%)": test_recalls_percent,
"F1-score (%)": test_f1_scores_percent
})
# Arrondir les valeurs à 2 décimales
results_df = results_df.round(2)
# Affichage du tableau
print("\nTest Metrics per Subject:\n")
print(results_df.to_string(index=False))
# Moyennes globales
average_results = results_df.mean(numeric_only=True)
print("\nAverage Test Metrics Across All Subjects:")
print(f"Accuracy: {average_results['Accuracy (%)']:.2f}%")
print(f"Precision: {average_results['Precision (%)']:.2f}%")
print(f"Recall: {average_results['Recall (%)']:.2f}%")
print(f"F1-score: {average_results['F1-score (%)']:.2f}%")
# %% [markdown]
# # RFC classification
# %%
from sklearn.model_selection import KFold
from sklearn.preprocessing import StandardScaler
from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score
from sklearn.pipeline import make_pipeline
import numpy as np
from sklearn.ensemble import RandomForestClassifier
# Initialisation des listes
cv_train_accuracies, cv_val_accuracies, cv_test_accuracies = [], [], []
cv_train_precisions, cv_val_precisions, cv_test_precisions = [], [], []
cv_train_recalls, cv_val_recalls, cv_test_recalls = [], [], []
cv_train_f1s, cv_val_f1s, cv_test_f1s = [], [], []
kfold = KFold(n_splits=5, shuffle=True, random_state=42)
for i, (features_train, y_labels_T, features_test, y_labels_E) in enumerate(zip(all_person_featuresT, all_y_labels_T,
all_person_featuresE, all_y_labels_E)):
print(f"\nProcessing subject {i + 1}")
# Pour accumuler les scores par pli
fold_train_accs, fold_val_accs = [], []
fold_train_prec, fold_val_prec = [], []
fold_train_rec, fold_val_rec = [], []
fold_train_f1, fold_val_f1 = [], []
model = make_pipeline(StandardScaler(),
RandomForestClassifier(n_estimators=100, random_state=42))
# Validation croisée
for train_idx, val_idx in kfold.split(features_train):
X_train_fold, X_val_fold = features_train[train_idx], features_train[val_idx]
y_train_fold, y_val_fold = y_labels_T[train_idx], y_labels_T[val_idx]
model.fit(X_train_fold, y_train_fold)
# Prédictions
y_train_pred = model.predict(X_train_fold)
y_val_pred = model.predict(X_val_fold)
# Accuracy
fold_train_accs.append(accuracy_score(y_train_fold, y_train_pred))
fold_val_accs.append(accuracy_score(y_val_fold, y_val_pred))
# Precision, Recall, F1
fold_train_prec.append(precision_score(y_train_fold, y_train_pred, average='weighted', zero_division=0))
fold_val_prec.append(precision_score(y_val_fold, y_val_pred, average='weighted', zero_division=0))
fold_train_rec.append(recall_score(y_train_fold, y_train_pred, average='weighted', zero_division=0))
fold_val_rec.append(recall_score(y_val_fold, y_val_pred, average='weighted', zero_division=0))
fold_train_f1.append(f1_score(y_train_fold, y_train_pred, average='weighted', zero_division=0))
fold_val_f1.append(f1_score(y_val_fold, y_val_pred, average='weighted', zero_division=0))
# Réentraînement sur tout l'ensemble d'apprentissage
model.fit(features_train, y_labels_T)
y_test_pred = model.predict(features_test)
# Test scores
test_acc = accuracy_score(y_labels_E, y_test_pred)
test_prec = precision_score(y_labels_E, y_test_pred, average='weighted', zero_division=0)
test_rec = recall_score(y_labels_E, y_test_pred, average='weighted', zero_division=0)
test_f1 = f1_score(y_labels_E, y_test_pred, average='weighted', zero_division=0)
# Moyennes sur les plis
cv_train_accuracies.append(np.mean(fold_train_accs))
cv_val_accuracies.append(np.mean(fold_val_accs))
cv_test_accuracies.append(test_acc)
cv_train_precisions.append(np.mean(fold_train_prec))
cv_val_precisions.append(np.mean(fold_val_prec))
cv_test_precisions.append(test_prec)
cv_train_recalls.append(np.mean(fold_train_rec))
cv_val_recalls.append(np.mean(fold_val_rec))
cv_test_recalls.append(test_rec)
cv_train_f1s.append(np.mean(fold_train_f1))
cv_val_f1s.append(np.mean(fold_val_f1))
cv_test_f1s.append(test_f1)
# %%
import pandas as pd
# Conversion des scores en pourcentages
test_accuracies_percent = [acc * 100 for acc in cv_test_accuracies]
test_precisions_percent = [prec * 100 for prec in cv_test_precisions]
test_recalls_percent = [rec * 100 for rec in cv_test_recalls]
test_f1_scores_percent = [f1 * 100 for f1 in cv_test_f1s]
# Création d'un DataFrame avec les scores
results_df = pd.DataFrame({
"Subject": [f"Person {i+1}" for i in range(len(test_accuracies_percent))],
"Accuracy (%)": test_accuracies_percent,
"Precision (%)": test_precisions_percent,
"Recall (%)": test_recalls_percent,
"F1-score (%)": test_f1_scores_percent
})
# Arrondir les valeurs à 2 décimales
results_df = results_df.round(2)
# Affichage du tableau
print("\nTest Metrics per Subject:\n")
print(results_df.to_string(index=False))
# Moyennes globales
average_results = results_df.mean(numeric_only=True)
print("\nAverage Test Metrics Across All Subjects:")
print(f"Accuracy: {average_results['Accuracy (%)']:.2f}%")
print(f"Precision: {average_results['Precision (%)']:.2f}%")
print(f"Recall: {average_results['Recall (%)']:.2f}%")
print(f"F1-score: {average_results['F1-score (%)']:.2f}%")
# %% [markdown]
# # MLP classification
# %%
from sklearn.model_selection import KFold
from sklearn.preprocessing import StandardScaler
from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score
from sklearn.pipeline import make_pipeline
import numpy as np
from sklearn.ensemble import RandomForestClassifier
from sklearn.neural_network import MLPClassifier
# Initialisation des listes
cv_train_accuracies, cv_val_accuracies, cv_test_accuracies = [], [], []
cv_train_precisions, cv_val_precisions, cv_test_precisions = [], [], []
cv_train_recalls, cv_val_recalls, cv_test_recalls = [], [], []
cv_train_f1s, cv_val_f1s, cv_test_f1s = [], [], []
kfold = KFold(n_splits=5, shuffle=True, random_state=42)
for i, (features_train, y_labels_T, features_test, y_labels_E) in enumerate(zip(all_person_featuresT, all_y_labels_T,
all_person_featuresE, all_y_labels_E)):
print(f"\nProcessing subject {i + 1}")
# Pour accumuler les scores par pli
fold_train_accs, fold_val_accs = [], []
fold_train_prec, fold_val_prec = [], []
fold_train_rec, fold_val_rec = [], []
fold_train_f1, fold_val_f1 = [], []
model = make_pipeline(StandardScaler(),
MLPClassifier(hidden_layer_sizes=(100,), max_iter=300, activation='tanh', solver='adam',
learning_rate_init=0.001, random_state=42 ))
# Validation croisée
for train_idx, val_idx in kfold.split(features_train):
X_train_fold, X_val_fold = features_train[train_idx], features_train[val_idx]
y_train_fold, y_val_fold = y_labels_T[train_idx], y_labels_T[val_idx]
model.fit(X_train_fold, y_train_fold)
# Prédictions
y_train_pred = model.predict(X_train_fold)
y_val_pred = model.predict(X_val_fold)
# Accuracy
fold_train_accs.append(accuracy_score(y_train_fold, y_train_pred))
fold_val_accs.append(accuracy_score(y_val_fold, y_val_pred))
# Precision, Recall, F1
fold_train_prec.append(precision_score(y_train_fold, y_train_pred, average='weighted', zero_division=0))
fold_val_prec.append(precision_score(y_val_fold, y_val_pred, average='weighted', zero_division=0))
fold_train_rec.append(recall_score(y_train_fold, y_train_pred, average='weighted', zero_division=0))
fold_val_rec.append(recall_score(y_val_fold, y_val_pred, average='weighted', zero_division=0))
fold_train_f1.append(f1_score(y_train_fold, y_train_pred, average='weighted', zero_division=0))
fold_val_f1.append(f1_score(y_val_fold, y_val_pred, average='weighted', zero_division=0))
# Réentraînement sur tout l'ensemble d'apprentissage
model.fit(features_train, y_labels_T)
y_test_pred = model.predict(features_test)
# Test scores
test_acc = accuracy_score(y_labels_E, y_test_pred)
test_prec = precision_score(y_labels_E, y_test_pred, average='weighted', zero_division=0)
test_rec = recall_score(y_labels_E, y_test_pred, average='weighted', zero_division=0)
test_f1 = f1_score(y_labels_E, y_test_pred, average='weighted', zero_division=0)
# Moyennes sur les plis
cv_train_accuracies.append(np.mean(fold_train_accs))
cv_val_accuracies.append(np.mean(fold_val_accs))
cv_test_accuracies.append(test_acc)
cv_train_precisions.append(np.mean(fold_train_prec))
cv_val_precisions.append(np.mean(fold_val_prec))
cv_test_precisions.append(test_prec)
cv_train_recalls.append(np.mean(fold_train_rec))
cv_val_recalls.append(np.mean(fold_val_rec))
cv_test_recalls.append(test_rec)
cv_train_f1s.append(np.mean(fold_train_f1))
cv_val_f1s.append(np.mean(fold_val_f1))
cv_test_f1s.append(test_f1)
# %%
import pandas as pd
# Conversion des scores en pourcentages
test_accuracies_percent = [acc * 100 for acc in cv_test_accuracies]
test_precisions_percent = [prec * 100 for prec in cv_test_precisions]
test_recalls_percent = [rec * 100 for rec in cv_test_recalls]
test_f1_scores_percent = [f1 * 100 for f1 in cv_test_f1s]
# Création d'un DataFrame avec les scores
results_df = pd.DataFrame({
"Subject": [f"Person {i+1}" for i in range(len(test_accuracies_percent))],
"Accuracy (%)": test_accuracies_percent,
"Precision (%)": test_precisions_percent,
"Recall (%)": test_recalls_percent,
"F1-score (%)": test_f1_scores_percent
})
# Arrondir les valeurs à 2 décimales
results_df = results_df.round(2)
# Affichage du tableau
print("\nTest Metrics per Subject:\n")
print(results_df.to_string(index=False))
# Moyennes globales
average_results = results_df.mean(numeric_only=True)
print("\nAverage Test Metrics Across All Subjects:")
print(f"Accuracy: {average_results['Accuracy (%)']:.2f}%")
print(f"Precision: {average_results['Precision (%)']:.2f}%")
print(f"Recall: {average_results['Recall (%)']:.2f}%")
print(f"F1-score: {average_results['F1-score (%)']:.2f}%")
# %% [markdown]
# # votting simple
# %%
from sklearn.ensemble import VotingClassifier
from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
from sklearn.neural_network import MLPClassifier
from sklearn.ensemble import RandomForestClassifier
from sklearn.pipeline import make_pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score
# Initialisation des listes pour stocker les performances
voting_accuracies, voting_precisions, voting_recalls, voting_f1_scores = [], [], [], []
# Boucle sur chaque personne
for i, (X_train, y_train, X_test, y_test) in enumerate(zip(all_person_featuresT, all_y_labels_T,
all_person_featuresE, all_y_labels_E)):
print(f"\n>>> Voting classifier - Subject {i + 1}")
# Définition des 3 classificateurs de base
lda = make_pipeline(StandardScaler(), LinearDiscriminantAnalysis(solver='lsqr', shrinkage='auto', n_components=3, tol=1e-4))
rf = make_pipeline(StandardScaler(), RandomForestClassifier(n_estimators=100, random_state=42))
mlp = make_pipeline(StandardScaler(),MLPClassifier(hidden_layer_sizes=(100,), max_iter=300, activation='tanh', solver='adam',
learning_rate_init=0.001, random_state=42 ))
# Voting classifier (majority vote)
voting_clf = VotingClassifier(estimators=[
('lda', lda),
('mlp', mlp),
('rf', rf)
], voting='hard')
# Entraînement
voting_clf.fit(X_train, y_train)
# Prédiction
y_pred = voting_clf.predict(X_test)
# Évaluation
acc = accuracy_score(y_test, y_pred)
prec = precision_score(y_test, y_pred, average='weighted', zero_division=0)
rec = recall_score(y_test, y_pred, average='weighted', zero_division=0)
f1 = f1_score(y_test, y_pred, average='weighted', zero_division=0)
# Stockage
voting_accuracies.append(acc)
voting_precisions.append(prec)
voting_recalls.append(rec)
voting_f1_scores.append(f1)
# Résultats moyens
print("\n>>> Voting Classifier - Average Test Metrics:")
print(f"Accuracy: {np.mean(voting_accuracies) * 100:.2f}%")
print(f"Precision: {np.mean(voting_precisions) * 100:.2f}%")
print(f"Recall: {np.mean(voting_recalls) * 100:.2f}%")
print(f"F1-score: {np.mean(voting_f1_scores) * 100:.2f}%")
# %% [markdown]
# # votting+ Cross validation
# %%
from sklearn.ensemble import VotingClassifier
from sklearn.model_selection import cross_val_score
from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score
# Initialisation des métriques
metrics = {
'subject': [], 'acc': [], 'prec': [], 'rec': [], 'f1': [],
'cv_acc_mean': [], 'cv_acc_std': []
}
for i, (X_train, y_train, X_test, y_test) in enumerate(zip(all_person_featuresT, all_y_labels_T,
all_person_featuresE, all_y_labels_E)):
print(f"\n>>> Subject {i + 1}")
# Définition des classifieurs
lda = make_pipeline(StandardScaler(), LinearDiscriminantAnalysis(solver='lsqr', shrinkage='auto'))
rf = make_pipeline(StandardScaler(), RandomForestClassifier(n_estimators=100, class_weight='balanced', random_state=42))
mlp = make_pipeline(StandardScaler(), MLPClassifier(hidden_layer_sizes=(100,), max_iter=300, early_stopping=True, random_state=42))
# Voting Classifier
voting_clf = VotingClassifier(
estimators=[('lda', lda), ('rf', rf), ('mlp', mlp)],
voting='soft' # ou 'hard'
)
# Validation croisée
cv_scores = cross_val_score(voting_clf, X_train, y_train, cv=5, scoring='accuracy')
metrics['cv_acc_mean'].append(np.mean(cv_scores))
metrics['cv_acc_std'].append(np.std(cv_scores))
# Entraînement final
voting_clf.fit(X_train, y_train)
y_pred = voting_clf.predict(X_test)
# Stockage des résultats
metrics['subject'].append(i+1)
metrics['acc'].append(accuracy_score(y_test, y_pred))
metrics['prec'].append(precision_score(y_test, y_pred, average='weighted', zero_division=0))
metrics['rec'].append(recall_score(y_test, y_pred, average='weighted', zero_division=0))
metrics['f1'].append(f1_score(y_test, y_pred, average='weighted', zero_division=0))
# Affichage détaillé
print(f"CV Accuracy: {metrics['cv_acc_mean'][-1]:.3f} ± {metrics['cv_acc_std'][-1]:.3f}")
print(f"Test Accuracy: {metrics['acc'][-1]:.3f}")
# Résultats finaux
print("\n>>> Final Metrics <<<")
print(f"Mean Test Accuracy: {np.mean(metrics['acc']) * 100:.2f}%")
print(f"Mean Test Precision: {np.mean(metrics['prec']) * 100:.2f}%")
print(f"Mean Test Recall: {np.mean(metrics['rec']) * 100:.2f}%")
print(f"Mean Test F1-score: {np.mean(metrics['f1']) * 100:.2f}%")
# %%
from sklearn.ensemble import VotingClassifier
from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
from sklearn.neural_network import MLPClassifier
from sklearn.ensemble import RandomForestClassifier
from sklearn.pipeline import make_pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score
from collections import Counter
import numpy as np
# Initialisation des listes pour stocker les performances
voting_accuracies, voting_precisions, voting_recalls, voting_f1_scores = [], [], [], []
# Boucle sur chaque personne
for i, (X_train, y_train, X_test, y_test) in enumerate(zip(all_person_featuresT, all_y_labels_T,
all_person_featuresE, all_y_labels_E)):
print(f"\n>>> Voting classifier - Subject {i + 1}")
# Définition des 3 classificateurs de base
lda = make_pipeline(StandardScaler(), LinearDiscriminantAnalysis(solver='lsqr', shrinkage='auto'))
rf = make_pipeline(StandardScaler(), RandomForestClassifier(n_estimators=100, random_state=42))
mlp = make_pipeline(StandardScaler(), MLPClassifier(hidden_layer_sizes=(100,), max_iter=300,
activation='tanh', solver='adam',
learning_rate_init=0.001, random_state=42))
# Voting classifier (majority vote)
voting_clf = VotingClassifier(estimators=[
('lda', lda),
('mlp', mlp),
('rf', rf)
], voting='hard') # Pour vote majoritaire
# Entraînement
voting_clf.fit(X_train, y_train)
# Prédiction finale
y_pred = voting_clf.predict(X_test)
# Évaluation
acc = accuracy_score(y_test, y_pred)
prec = precision_score(y_test, y_pred, average='weighted', zero_division=0)
rec = recall_score(y_test, y_pred, average='weighted', zero_division=0)
f1 = f1_score(y_test, y_pred, average='weighted', zero_division=0)
# Stockage des métriques
voting_accuracies.append(acc)
voting_precisions.append(prec)
voting_recalls.append(rec)
voting_f1_scores.append(f1)
# Prédictions individuelles
lda_pred = voting_clf.named_estimators_['lda'].predict(X_test)
mlp_pred = voting_clf.named_estimators_['mlp'].predict(X_test)
rf_pred = voting_clf.named_estimators_['rf'].predict(X_test)
# Affichage des votes pour les 10 premiers échantillons
print("\nIndividual classifier decisions (first 10 samples) :")
print("Index | LDA | MLP | RF | Final Decision | Majority vote")
print("------------------------------------------------------------")
for j in range(min(10, len(X_test))):
votes = [lda_pred[j], mlp_pred[j], rf_pred[j]]
vote_count = Counter(votes)
majority_class, count = vote_count.most_common(1)[0]
print(f"{j:5d} | {lda_pred[j]:3} | {mlp_pred[j]:3} | {rf_pred[j]:3} | {y_pred[j]:14} | {majority_class} ({count} votes)")
# Résultats moyens sur tous les sujets
print("\n>>> Voting Classifier - Average Test Metrics:")
print(f"Accuracy: {np.mean(voting_accuracies) * 100:.2f}%")
print(f"Precision: {np.mean(voting_precisions) * 100:.2f}%")
print(f"Recall: {np.mean(voting_recalls) * 100:.2f}%")
print(f"F1-score: {np.mean(voting_f1_scores) * 100:.2f}%")
# %%

BCI_R-FBCSP.ipynb, under CC-BY-4.0 · at the source

Overview

Authors: Souissi Jihen¹, Sourour Karmani^2,3, Kais Belwafi⁴, Mahdi Jemmali⁵, Ridha Djemal¹

Institutions: University of Sfax (Tunisia); University of Monastir (Tunisia); University of Sousse (Tunisia); University of Sharjah (United Arab Emirates)

Journal: Brain informatics, volume 13, issue 1, article 12

Dates: received 26 September 2025; accepted 28 February 2026; published online 26 March 2026

Type: Research article · Language: English

License: CC BY-NC-ND

Identifiers: DOI 10.1186/s40708-026-00295-0 · PMID 41886201 · PMCID PMC13133326 · OpenAlex W7140680856

Open access: gold, a free copy (OpenAlex)

Status: code verified (how the code is found and verified)

Categories: EEG (modality), methods / tools (subfield)

Methods: Spectral & time-frequency, Preprocessing, Statistics, Machine learning

Keywords: Brain-computer interface, Motor imagery, ElectroEncephaloGram, Feature extraction, Riemannian geometry, Multi-classification

Topic: EEG and Brain-Computer Interfaces (Cognitive Neuroscience, Neuroscience), according to OpenAlex

Funding: University of Sharjah (2502150165)

Citations: not cited yet (Europe PMC); 47 references in the paper

Research resources: Scikit-learn 0.23.2 RRID:SCR_002577, MNE-Python RRID:SCR_005972, Python 3.9 RRID:SCR_008394, NumPy 1.19 RRID:SCR_008633, EEG Competition IV Dataset 2a RRID:SCR_014177, with statsmodels RRID:SCR_015794

Abstract

The abstract is not reproduced here: the paper's license (CC BY-NC-ND) does not allow it. Read it in the paper, at the publisher or on Europe PMC.

Repositories

Its files are read in the Code ↔ Paper reader above, with 1 match between paragraphs and lines of code.

Zenodo 15825805

License: CC-BY-4.0

State: the link answers, verified on 30 September 2026

Evidence: files inventoried

Size: 3 files, 0 scripts

Software Heritage: not checked

Found in: the text, “Key resources”

Not found: README, license file, CITATION.cff, environment file, tests, continuous integration, documentation

Availability: 1 check, the latest on 30 September 2026: the link answers (HTTP 200)

30 September 2026: the link answers (HTTP 200)

At the source:

souissijihen/BCI_RW_FBCSP_GigaDB

License: none: the authors keep all their rights

State: the link is dead, verified on 30 September 2026

Evidence: found in the paper

Software Heritage: not archived

Found in: the text, “Key resources”

Not found: README, license file, CITATION.cff, environment file, tests, continuous integration, documentation

Availability: 1 check, the latest on 30 September 2026: the link is dead

30 September 2026: the link is dead

At the source: github.com/souissijihen/BCI_RW_FBCSP_GigaDB

Zenodo 15828534

License: CC-BY-4.0

State: the link answers, verified on 30 September 2026

Evidence: files inventoried

Size: 1 file

Software Heritage: not checked

Found in: the text, “Key resources”

Not found: README, license file, CITATION.cff, environment file, tests, continuous integration, documentation

Tools: MNE-Python (3 files), NumPy (3 files), pyRiemann (3 files), scikit-learn (3 files), SciPy (3 files), pandas (2 files), Matplotlib (1 file)

Availability: 1 check, the latest on 30 September 2026: the link answers (HTTP 200)

30 September 2026: the link answers (HTTP 200)

4 files

BCI_R-FBCSP.ipynb, Jupyter, 1,259 lines, 1 match
BCI_RW-FBCSP.ipynb, Jupyter, 1,212 lines
CSP_jrnl_affichage .ipynb, Jupyter, 638 lines
README.md, Text, 19 lines

At the source:

souissijihen/bci_rw_fbcsp-

License: none: the authors keep all their rights

State: the link is dead, verified on 30 September 2026

Evidence: found in the paper

Software Heritage: not archived

Found in: the Zenodo archive record

Not found: README, license file, CITATION.cff, environment file, tests, continuous integration, documentation

Availability: 1 check, the latest on 30 September 2026: the link is dead

30 September 2026: the link is dead

At the source: github.com/souissijihen/BCI_RW_FBCSP-/tree/v0.1

Tracing map

Proposed by the machine: these links were found in the paper and verified at the source, without human review. The map will receive a Zenodo DOI once one of the paper's authors has validated it with their ORCID.

What the map holds:

4 repositories of the authors' code, each at its verified commit, with its license and how the link was found in the paper;
3 scripts, each with its path and the digest of its content;
1 match between paragraphs of the paper and lines of the code (method lexical-v1);
neither the text of the paper nor the code itself.

Its JSON (tracing-map.json) is deposited on Zenodo with its DOI once the map is validated.

Data

Datasets cited

bbci.de/competition/iv, at bbci.de; found in the references

Data availability statement

The paper has a data availability statement. Its license (CC BY-NC-ND) does not allow reproducing it here; in short, from what the harvester recognized in it:

no repository, dataset or request procedure was recognized in it

Read it in the paper: doi.org/10.1186/s40708-026-00295-0.

Versions

The history of this record: each version stored by the harvester or made by a correction of its authors or of the maintainers of its code, and what changed in its facts. The texts of the paper (its abstract, its availability statements) are not part of it; versions that changed only those are not listed.

Version 1, 30 September 2026: the first record

Recorded: type, language, journal, volume, issue, pages, dates, 5 authors, 6 keywords, 1 funder, 36 references, 6 RRIDs.

Cite

This paper

Jihen, S., Karmani, S., Belwafi, K., Jemmali, M., & Djemal, R. (2026). Filter bank CSP with Riemannian weighting for disability-centric motor imagery brain computer interface. Brain informatics, 13(1), 12. https://doi.org/10.1186/s40708-026-00295-0

BibTeX

@article{jihen2026filter, author = {Jihen, Souissi and Karmani, Sourour and Belwafi, Kais and Jemmali, Mahdi and Djemal, Ridha}, title = {{Filter bank CSP with Riemannian weighting for disability-centric motor imagery brain computer interface}}, journal = {Brain informatics}, year = {2026}, month = mar, volume = {13}, number = {1}, pages = {12}, publisher = {Springer}, issn = {2198-4018}, doi = {10.1186/s40708-026-00295-0}, url = {https://doi.org/10.1186/s40708-026-00295-0}, pmid = {41886201}, pmcid = {PMC13133326} }

RIS

TY - JOUR AU - Jihen, Souissi AU - Karmani, Sourour AU - Belwafi, Kais AU - Jemmali, Mahdi AU - Djemal, Ridha TI - Filter bank CSP with Riemannian weighting for disability-centric motor imagery brain computer interface T2 - Brain informatics J2 - Brain Inform PY - 2026 DA - 2026/03/26 VL - 13 IS - 1 SP - 12 SN - 2198-4018 PB - Springer DO - 10.1186/s40708-026-00295-0 UR - https://doi.org/10.1186/s40708-026-00295-0 LA - en ER -

CSL-JSON

{ "id": "10.1186/s40708-026-00295-0", "type": "article-journal", "title": "Filter bank CSP with Riemannian weighting for disability-centric motor imagery brain computer interface", "container-title": "Brain informatics", "author": [ { "family": "Jihen", "given": "Souissi" }, { "family": "Karmani", "given": "Sourour" }, { "family": "Belwafi", "given": "Kais" }, { "family": "Jemmali", "given": "Mahdi" }, { "family": "Djemal", "given": "Ridha" } ], "container-title-short": "Brain Inform", "volume": "13", "issue": "1", "page": "12", "DOI": "10.1186/s40708-026-00295-0", "PMID": "41886201", "PMCID": "PMC13133326", "ISSN": "2198-4018", "publisher": "Springer", "URL": "https://doi.org/10.1186/s40708-026-00295-0", "language": "en", "issued": { "date-parts": [ [ 2026, 3, 26 ] ] } }

The tracing map gets a citation of its own once an author has validated it and it has a DOI.

Contribute

The authors of this paper can claim it, correct its record and validate its tracing map, and the maintainers of its code (its owner, or a public member of its organization) correct what it says of their repository; anyone signed in can ask for its removal. Every request goes to OSCR's own machine, which answers it; your account page follows them.

Sign in with ORCID to claim this paper as one of its authors, correct its record or validate its tracing map: when the paper's metadata lists your ORCID iD, you are recognized at once. Maintainers of its code: sign in with GitHub, then claim the repository on your account page.

Add the badge to the code's README

If you are an author or a maintainer of this paper's code, you can add this badge. It says what OSCR knows of the paper (its code, its validated tracing map) and links the code to this page. Paste one of these snippets under the README's title. Keep the two marks around it, so that the badge can be replaced in place and is never added twice. OSCR changes nothing in your repository.

Markdown

<!-- oscr-badge:start -->
[![OSCR](https://openscicode.org/badge/doi/10.1186/s40708-026-00295-0.svg)](https://openscicode.org/badge/doi/10.1186/s40708-026-00295-0)
<!-- oscr-badge:end -->

reStructuredText

.. oscr-badge:start

.. image:: https://openscicode.org/badge/doi/10.1186/s40708-026-00295-0.svg
   :target: https://openscicode.org/badge/doi/10.1186/s40708-026-00295-0
   :alt: OSCR

.. oscr-badge:end

HTML

<!-- oscr-badge:start -->
<a href="https://openscicode.org/badge/doi/10.1186/s40708-026-00295-0"><img src="https://openscicode.org/badge/doi/10.1186/s40708-026-00295-0.svg" alt="OSCR"></a>
<!-- oscr-badge:end -->

On GitHub, open the README, choose Edit (the pencil), paste the snippet under the title, then "Commit changes".

If you choose to, you can have the badge kept up to date for you, with oscr badge --write from the command line or with the GitHub Action in your repository's workflows. Every other form and option (a card, AsciiDoc, LaTeX, shields.io) is on the badge page.

Request its removal

To ask OSCR to remove this record, the copies of its authors' scripts or its tracing map, use the removal request page: signed in, you say who you are, what to remove and why, then review and confirm the request. Published rules decide every request (how).

Discussion, reproductions, activity

Discussion: code errors and mismatches between the code and the paper about this paper and its code, from readers and its authors, beside the paper and its tracing map.

Report a mismatch between the code and the paper · Report a code error (signed in)

Reproductions: reports from readers who ran the authors' code and did not get the paper's result, what they ran, with which environment, commit and data.

Report a reproduction failure (signed in)

Versions of its code: the releases of its code tied to a version of this paper (the preprint, the accepted manuscript, the version of record…), each with the tracing map versioned with it.

Activity: what happens around this paper: new versions of its record, its map's validation, discussions and reproductions. It opens with sign-in.

Filter bank CSP with Riemannian weighting for disability-centric motor imagery brain computer interface.

Code ↔ Paper

Paper

The authors' code

Overview

Abstract

Repositories

Zenodo 15825805

souissijihen/BCI_RW_FBCSP_GigaDB

Zenodo 15828534

souissijihen/bci_rw_fbcsp-

Tracing map

Data

Datasets cited

Data availability statement

Versions

Version 1, 30 September 2026: the first record

Cite

This paper

Similar papers

Contribute

Claim this paper

Correct its record

Validate its tracing map

Add the badge to the code's README

Request its removal

Discussion, reproductions, activity