inner_train_fMRI_data = train_fMRI_data[inner_train_index, :]
inner_test_fMRI_data = train_fMRI_data[inner_test_index, :]
inner_train_sMRI_data = train_sMRI_data[inner_train_index, :]
inner_test_sMRI_data = train_sMRI_data[inner_test_index, :]
inner_train_targets = train_targets[inner_train_index]
# correct inner training data
#inner_train_data, LS_dict, v_pool = fit_transform_neuroCombat(inner_train_data, train_metadata.iloc[inner_train_index, :], 'Site', continuous_cols=[timepoint + '|total_panss'])
inner_train_fMRI_data, model = fit_transform_neuroCombat(
inner_train_fMRI_data,
train_metadata.iloc[inner_train_index, :],
'Site',
continuous_cols=[timepoint + '|total_panss'])
# correct testing data
inner_test_fMRI_data = apply_neuroCombat_model(
inner_test_fMRI_data, train_metadata.iloc[inner_test_index, :],
model, 'Site')
inner_train_sMRI_data, model = fit_transform_neuroCombat(
inner_train_sMRI_data,
train_metadata.iloc[inner_train_index, :],
'Site',
continuous_cols=[timepoint + '|total_panss'])
# correct testing data
inner_test_sMRI_data = apply_neuroCombat_model(
inner_test_sMRI_data, train_metadata.iloc[inner_test_index, :],
model, 'Site')
# make inner predictions
rgr_lin.fit(inner_train_fMRI_data, inner_train_targets)
inner_fMRI_pred = rgr_lin.predict(inner_test_fMRI_data)
train_data = connectivity_data[train_index, :]
test_data = connectivity_data[test_index, :]
# do supervised site correction?
if site_correction == 'comBat_supervised':
# correct training data
train_data, model = fit_transform_neuroCombat(
train_data,
metadata.iloc[train_index, :],
'Site',
continuous_cols=[timepoint + '|total_panss'])
# correct testing dara
test_data = apply_neuroCombat_model(test_data,
metadata.iloc[test_index, :],
model, 'Site')
#recreate a stack of matrices for train and test
# add extra singleton dimension for channels
train_matrices = np.reshape(train_data, (train_size, n_regions, n_regions))
test_matrices = np.reshape(test_data, (test_size, n_regions, n_regions))
train_matrices = train_matrices[:, :, :, np.newaxis]
test_matrices = test_matrices[:, :, :, np.newaxis]
# create and compile the model
model = brainnetCNN_model_2((n_regions, n_regions, 1),
n_filters,
use_bias,
n_outputs=n_PANSS_items)
# my custom
train_fMRI_data = fMRI_data[train_index, :]
test_fMRI_data = fMRI_data[test_index, :]
train_sMRI_data = sMRI_data[train_index, :]
test_sMRI_data = sMRI_data[test_index, :]
# do supervised site correction?
if site_correction == 'comBat_supervised' :
# correct training data
train_fMRI_data, fMRI_model = fit_transform_neuroCombat(train_fMRI_data, metadata.iloc[train_index, :], 'Site', continuous_cols=[timepoint + '|total_panss'])
train_sMRI_data, sMRI_model = fit_transform_neuroCombat(train_sMRI_data, metadata.iloc[train_index, :], 'Site', continuous_cols=[timepoint + '|total_panss'])
# correct testing dara
test_fMRI_data = apply_neuroCombat_model(test_fMRI_data,
metadata.iloc[test_index, :], fMRI_model,
'Site')
test_sMRI_data = apply_neuroCombat_model(test_sMRI_data,
metadata.iloc[test_index, :], sMRI_model,
'Site')
#recreate a stack of matrices for train and test
# add extra singleton dimension for channels
train_fMRI_matrices = np.reshape(train_fMRI_data, (train_size, n_fMRI_regions, n_fMRI_regions))
test_fMRI_matrices = np.reshape(test_fMRI_data, (test_size, n_fMRI_regions, n_fMRI_regions))
train_fMRI_matrices = train_fMRI_matrices[:, :, :, np.newaxis]
test_fMRI_matrices = test_fMRI_matrices[:, :, :, np.newaxis]
train_sMRI_matrices = np.reshape(train_sMRI_data, (train_size, n_sMRI_regions, n_sMRI_regions))
test_sMRI_matrices = np.reshape(test_sMRI_data, (test_size, n_sMRI_regions, n_sMRI_regions))
train_sMRI_matrices = train_sMRI_matrices[:, :, :, np.newaxis]
test_sMRI_matrices = test_sMRI_matrices[:, :, :, np.newaxis]
