n_iter = 1000
n_folds = 10
clf = SVC(kernel='linear', C=1, class_weight='balanced')
mvp = joblib.load(op.join(base_dir, 'mvp_other.jl'))
pipe = Pipeline([('scaler', StandardScaler()),
('ufs', MeanEuclidean(cutoff=2.3)),
('perm', ArrayPermuter()),
('svm', clf)])
for i in range(n_iter):
print('iter %i' % (i+1))
out_path = op.join(base_dir, 'other',
'perm_%i' % (i+1))
mvp_results = MvpResultsClassification(mvp, n_folds,
out_path=out_path,
feature_scoring='coef', verbose=False)
folds = StratifiedKFold(mvp.y, n_folds=n_folds, shuffle=True)
for train_idx, test_idx in folds:
train, test = mvp.X[train_idx, :], mvp.X[test_idx, :]
y_train, y_test = mvp.y[train_idx], mvp.y[test_idx]
pipe.fit(train, y_train)
pred = pipe.predict(ArrayPermuter().fit_transform(test))
mvp_results.update(test_idx, pred, pipeline=pipe)
mvp_results.compute_scores()
mvp_results.write(feature_viz=True, confmat=True, to_tstat=True)
clf = SVC(kernel='linear', C=1, class_weight='balanced')
mvp = joblib.load(op.join(base_dir, 'mvp_other.jl'))
mvp_cross = joblib.load(op.join(base_dir, 'mvp_self.jl'))
pipe = Pipeline([('scaler', StandardScaler()),
('ufs', MeanEuclidean(cutoff=2.3)),
('perm', ArrayPermuter()),
('svm', clf)])
for i in range(n_iter):
print('iter %i' % (i+1))
out_path = op.join(base_dir, 'other>>self',
'perm_%i' % (i+1))
mvp_results = MvpResultsClassification(mvp_cross, n_folds,
out_path=out_path,
feature_scoring='fwm', verbose=False)
folds = StratifiedKFold(mvp.y, n_folds=n_folds, shuffle=True)
for train_idx, test_idx in folds:
train, test = mvp.X[train_idx, :], mvp.X[test_idx, :]
y_train, y_test = mvp.y[train_idx], mvp.y[test_idx]
pipe.fit(train, y_train)
pred = pipe.predict(ArrayPermuter().fit_transform(mvp_cross.X))
mvp_results.update(range(mvp_cross.y.size), pred, pipeline=pipe)
mvp_results.compute_scores()
print(mvp_results.confmat)
mvp_results.write(feature_viz=True, confmat=True, to_tstat=True)
base_dir = '/media/lukas/data/SharedStates/RESULTS/MVPA/Validation/Condition_average_mvpa'
### ARGUMENTS ###
n_folds = 10
out_path = op.join(base_dir, 'other>>self')
clf = SVC(kernel='linear', C=1, class_weight='balanced')
mvp = joblib.load(op.join(base_dir, 'mvp_other.jl'))
mvp_cross = joblib.load(op.join(base_dir, 'mvp_self.jl'))
pipe = Pipeline([('scaler', StandardScaler()),
('ufs', MeanEuclidean(cutoff=2.3)),
('svm', clf)])
mvp_results = MvpResultsClassification(mvp_cross, n_folds, out_path=out_path,
feature_scoring='coef', verbose=True)
folds = StratifiedKFold(mvp.y, n_folds=n_folds)
for train_idx, test_idx in folds:
train, test = mvp.X[train_idx, :], mvp.X[test_idx, :]
y_train, y_test = mvp.y[train_idx], mvp.y[test_idx]
pipe.fit(train, y_train)
pred = pipe.predict(mvp_cross.X)
mvp_results.update(range(mvp_cross.y.size), pred, pipeline=pipe)
mvp_results.compute_scores()
mvp_results.write()
mvp_results.save_model(pipe)
from sklearn.svm import SVC
base_dir = '/media/lukas/data/SharedStates/RESULTS/MVPA/Validation/Condition_average_mvpa'
### ARGUMENTS ###
n_folds = 10
out_path = op.join(base_dir, 'self') # change 'self' to e.g. 'other' to do within other-decoding
clf = SVC(kernel='linear', C=1, class_weight='balanced')
mvp = joblib.load(op.join(base_dir, 'mvp_self.jl'))
pipe = Pipeline([('scaler', StandardScaler()),
('ufs', MeanEuclidean(cutoff=2.3)),
('svm', clf)])
mvp_results = MvpResultsClassification(mvp, n_folds, out_path=out_path,
feature_scoring='fwm', verbose=True)
folds = StratifiedKFold(mvp.y, n_folds=n_folds)
for train_idx, test_idx in folds:
train, test = mvp.X[train_idx, :], mvp.X[test_idx, :]
y_train, y_test = mvp.y[train_idx], mvp.y[test_idx]
pipe.fit(train, y_train)
pred = pipe.predict(test)
mvp_results.update(test_idx, pred, pipeline=pipe)
mvp_results.compute_scores()
mvp_results.write(multiclass='ovo')
mvp_results.save_model(pipe)
