elif n_points == 4 * 4:
isi = 4.8
else:
continue
labels = np.hstack([[session] * int(19.2 / isi)
for session in range(4)])
lplo2 = LeavePLabelOut(labels, p=2)
for train_id, test_id in lplo2:
betas_train, betas_test = betas_isi[train_id], betas_isi[test_id]
conditions_train, conditions_test = (conditions_isi[train_id],
conditions_isi[test_id])
# Feature selection
betas_train, betas_test = de.feature_selection(betas_train,
betas_test,
conditions_train,
k=k)
# Fit a logistic regression to score the model
accuracy = de.glm_scoring(betas_train, betas_test,
conditions_train, conditions_test)
scores.append(accuracy)
subjects.append(subject + 1)
models.append(model)
isis.append(isi)
print('finished subject ' + str(subject))
# Stack the BOLD signals and the design matrices
fmri = np.vstack(fmri)
design = np.vstack(design)
stimuli = np.vstack(stimuli)
session_id_fmri = np.hstack(session_id_fmri)
lplo = LeavePLabelOut(session_id_fmri, p=2)
for train_index, test_index in lplo:
# Split into train and test sets
fmri_train, fmri_test = fmri[train_index], fmri[test_index]
design_train, design_test = design[train_index], design[test_index]
stimuli_train, stimuli_test = stimuli[train_index], stimuli[test_index]
# Feature selection
fmri_train, fmri_test = de.feature_selection(
fmri_train, fmri_test, np.argmax(stimuli_train, axis=1))
# Fit a ridge regression to predict the design matrix
prediction_test, prediction_train, score = de.fit_ridge(
fmri_train,
fmri_test,
design_train,
design_test,
double_prediction=True,
extra=fmri_train)
# Fit a logistic regression for deconvolution
accuracy = de.logistic_deconvolution(prediction_train,
prediction_test,
stimuli_train,
stimuli_test,
logistic_window = 4
delay = 1
else:
continue
labels = np.hstack([[session] * 20 for session in range(4)])
lplo2 = LeavePLabelOut(labels, p=1)
for train_id, test_id in lplo2:
fmri_train, fmri_test = fmri_isi[train_id], fmri_isi[test_id]
design_train, design_test = (design_isi[train_id],
design_isi[test_id])
stimuli_train, stimuli_test = (stimuli_isi[train_id],
stimuli_isi[test_id])
# Feature selection
fmri_train, fmri_test = de.feature_selection(
fmri_train, fmri_test, np.argmax(stimuli_train, axis=1))
# Fit a ridge regression to predict the design matrix
prediction_test, prediction_train, score = de.fit_ridge(
fmri_train, fmri_test, design_train, design_test,
double_prediction=True, extra=fmri_train)
# Fit a logistic regression for deconvolution
accuracy = de.logistic_deconvolution(
prediction_train, prediction_test, stimuli_train,
stimuli_test, logistic_window, delay=delay)
scores.append(accuracy)
subjects.append(subject + 1)
models.append('logistic deconvolution')
isis.append(isi)
elif n_points == 6 * 4:
isi = 3.2
elif n_points == 4 * 4:
isi = 4.8
else:
continue
labels = np.hstack([[session] * int(19.2/isi) for session in range(4)])
lplo2 = LeavePLabelOut(labels, p=2)
for train_id, test_id in lplo2:
betas_train, betas_test = betas_isi[train_id], betas_isi[test_id]
conditions_train, conditions_test = (conditions_isi[train_id],
conditions_isi[test_id])
# Feature selection
betas_train, betas_test = de.feature_selection(
betas_train, betas_test, conditions_train, k=k)
# Fit a logistic regression to score the model
accuracy = de.glm_scoring(betas_train, betas_test, conditions_train,
conditions_test)
scores.append(accuracy)
subjects.append(subject + 1)
models.append(model)
isis.append(isi)
print('finished subject ' + str(subject))
junk_mask = np.where(conditions != 'ju')
conditions = conditions[junk_mask]
fmri_windows = fmri_windows[junk_mask]
session_id_onset = session_id_onset[junk_mask]
lplo = LeavePLabelOut(session_id_onset, p=1)
for train_index, test_index in lplo:
# Split into train and test sets
fmri_windows_train, fmri_windows_test = (fmri_windows[train_index],
fmri_windows[test_index])
conditions_train, conditions_test = (conditions[train_index],
conditions[test_index])
# Feature selection
fmri_windows_train, fmri_windows_test = de.feature_selection(
fmri_windows_train, fmri_windows_test, conditions_train, k=k)
# Fit a logistic regression to score the model
accuracy = de.svm_scoring(fmri_windows_train, fmri_windows_test,
conditions_train, conditions_test)
scores.append(accuracy)
subjects.append(subject + 1)
models.append(model)
print('finished subject ' + str(subject))
dict = {}
dict['subject'] = subjects
dict['model'] = models
dict['accuracy'] = scores
junk_mask = np.where(conditions != 'ju')
conditions = conditions[junk_mask]
fmri_windows = fmri_windows[junk_mask]
session_id_onset = session_id_onset[junk_mask]
lplo = LeavePLabelOut(session_id_onset, p=1)
for train_index, test_index in lplo:
# Split into train and test sets
fmri_windows_train, fmri_windows_test = (fmri_windows[train_index],
fmri_windows[test_index])
conditions_train, conditions_test = (conditions[train_index],
conditions[test_index])
# Feature selection
fmri_windows_train, fmri_windows_test = de.feature_selection(
fmri_windows_train, fmri_windows_test, conditions_train, k=k)
# Fit a logistic regression to score the model
accuracy = de.svm_scoring(fmri_windows_train, fmri_windows_test,
conditions_train, conditions_test)
scores.append(accuracy)
subjects.append(subject + 1)
models.append(model)
print('finished subject ' + str(subject))
dict = {}
dict['subject'] = subjects
dict['model'] = models
dict['accuracy'] = scores