def calculate_and_save_correlation(subj_1_num, subj_2_num):
"""
Calculates correlation using smoothed 2-D data with 8 full width half
maximum mm, and saves values into a designated correlation_path. If a file
with calculated correlations already exists, uses that cached version
instead.
Parameters
----------
subj_1_num : int
subj_2_num : int
Returns
-------
None
"""
correlation_path = dp.get_correlation_path(subj_1_num, subj_2_num)
if not exists(correlation_path) or not USE_CACHED_DATA:
subj_1_data = np.load(dp.get_smoothed_2d_path(subj_1_num, 8))
subj_2_data = np.load(dp.get_smoothed_2d_path(subj_2_num, 8))
correlations = correlation(subj_1_data, subj_2_data)
np.save(correlation_path, correlations)
print('Saved {0}'.format(correlation_path))
else:
print('Using cached version of {0}'.format(correlation_path))
def main():
subj_num, fwhm_mm = 1, 4
voxels_sorted_by_t_statistic = lm.VoxelExtractor(subj_num,
'int-ext').t_stat()
design_matrix = dm.DesignMatrix(dp.get_smoothed_2d_path(subj_num, fwhm_mm))
train_volume_indices = pv.get_train_indices()
cv_values = []
for num_features in NUM_FEATURES:
voxel_feature_indices = voxels_sorted_by_t_statistic[:num_features]
X_train = design_matrix.get_design_matrix(train_volume_indices,
voxel_feature_indices)
y_train = np.array(design_matrix.get_labels(train_volume_indices))
cv_accuracies = []
for train, test in KFold(len(X_train), 5):
X_cv_train = X_train[train, :]
y_cv_train = y_train[train]
X_cv_test = X_train[test, :]
y_cv_test = y_train[test]
model = rf.Classifier(X_cv_train, y_cv_train)
model.train()
y_predicted = model.predict(X_cv_test)
cv_accuracies.append(accuracy_score(y_predicted, y_cv_test))
print(np.mean(cv_accuracies))
cv_values.append(np.mean(cv_accuracies))
output_path = '{0}/figures/rf_cross_validated_accuracies.txt'.format(
REPO_HOME_PATH)
np.savetxt(output_path, cv_values)
print('Saved {0}'.format(output_path))
def main():
subj_num, fwhm_mm = 1, 4
voxels_sorted_by_t_statistic = lm.VoxelExtractor(subj_num,
'int-ext').t_stat()
design_matrix = dm.DesignMatrix(dp.get_smoothed_2d_path(subj_num, fwhm_mm))
train_volume_indices = pv.get_train_indices()
cv_values = []
for num_features in NUM_FEATURES:
voxel_feature_indices = voxels_sorted_by_t_statistic[:num_features]
X_train = design_matrix.get_design_matrix(train_volume_indices,
voxel_feature_indices)
y_train = np.array(design_matrix.get_labels(train_volume_indices))
cv_accuracies = []
for train, test in KFold(len(X_train), 5):
X_cv_train = X_train[train, :]
y_cv_train = y_train[train]
X_cv_test = X_train[test, :]
y_cv_test = y_train[test]
model = rf.Classifier(X_cv_train, y_cv_train)
model.train()
y_predicted = model.predict(X_cv_test)
cv_accuracies.append(accuracy_score(y_predicted, y_cv_test))
print(np.mean(cv_accuracies))
cv_values.append(np.mean(cv_accuracies))
output_path = '{0}/figures/rf_cross_validated_accuracies.txt'.format(
REPO_HOME_PATH)
np.savetxt(output_path, cv_values)
print('Saved {0}'.format(output_path))
def __init__(self, subject, interest_col_str, data=None):
"""
VoxelExtractor generates the t-statistic for each voxel based on a
given interest column. For example, to see the voxel that can best
predict interior or exterior, you would use the following code.
ve = VoxelExtractor(1, 'int-ext')
a = ve.t_stat()[0]
ve.plot_single_voxel(0)
interest_col_str takes in either 'day-night' or 'int-ext'.
data can also be overrided if needed, else the smoothed 2d version
will be used.
"""
self.subject = subject
if interest_col_str == "int-ext":
self.interest_col_ind = 0
elif interest_col_str == "day-night":
self.interest_col_ind = 1
else:
raise ValueError("Incorrect interest column name: please use either 'int-ext' or 'day-night'")
self.interest_col_str = interest_col_str
if data is None:
data_path = dp.get_smoothed_2d_path(self.subject, 4)
data = np.load(data_path)
data = data[:, NUM_OFFSET_VOLUMES:]
self.data = data
self.design = None
self.B = None
self.t_values = None
def __init__(self, subject, interest_col_str, data=None):
"""
VoxelExtractor generates the t-statistic for each voxel based on a
given interest column. For example, to see the voxel that can best
predict interior or exterior, you would use the following code.
ve = VoxelExtractor(1, 'int-ext')
a = ve.t_stat()[0]
ve.plot_single_voxel(0)
interest_col_str takes in either 'day-night' or 'int-ext'.
data can also be overrided if needed, else the smoothed 2d version
will be used.
"""
self.subject = subject
if interest_col_str == "int-ext":
self.interest_col_ind = 0
elif interest_col_str == "day-night":
self.interest_col_ind = 1
else:
raise ValueError(
"Incorrect interest column name: please use either 'int-ext' or 'day-night'"
)
self.interest_col_str = interest_col_str
if data is None:
data_path = dp.get_smoothed_2d_path(self.subject, 4)
data = np.load(data_path)
data = data[:, NUM_OFFSET_VOLUMES:]
self.data = data
self.design = None
self.B = None
self.t_values = None
def test_get_smoothed_2d_path():
paths = [dp.get_smoothed_2d_path(subject, FWHM_MM) for subject in SUBJECTS]
assert paths[0] == '{0}/data/processed/sub1_rcds_smoothed_8_mm_2d.npy'.format(
REPO_HOME_PATH)
assert paths[1] == '{0}/data/processed/sub2_rcds_smoothed_8_mm_2d.npy'.format(
REPO_HOME_PATH)
assert paths[2] == '{0}/data/processed/sub3_rcds_smoothed_8_mm_2d.npy'.format(
REPO_HOME_PATH)
assert paths[3] == '{0}/data/processed/sub4_rcds_smoothed_8_mm_2d.npy'.format(
REPO_HOME_PATH)
assert paths[4] == '{0}/data/processed/sub5_rcds_smoothed_8_mm_2d.npy'.format(
REPO_HOME_PATH)
def test_get_smoothed_2d_path():
paths = [dp.get_smoothed_2d_path(subject, FWHM_MM) for subject in SUBJECTS]
assert paths[
0] == '{0}/data/processed/sub1_rcds_smoothed_8_mm_2d.npy'.format(
REPO_HOME_PATH)
assert paths[
1] == '{0}/data/processed/sub2_rcds_smoothed_8_mm_2d.npy'.format(
REPO_HOME_PATH)
assert paths[
2] == '{0}/data/processed/sub3_rcds_smoothed_8_mm_2d.npy'.format(
REPO_HOME_PATH)
assert paths[
3] == '{0}/data/processed/sub4_rcds_smoothed_8_mm_2d.npy'.format(
REPO_HOME_PATH)
assert paths[
4] == '{0}/data/processed/sub5_rcds_smoothed_8_mm_2d.npy'.format(
REPO_HOME_PATH)
def main():
subj_num, fwhm_mm = 1, 4
voxels_sorted_by_t_statistic = lm.VoxelExtractor(subj_num,
'int-ext').t_stat()
design_matrix = dm.DesignMatrix(dp.get_smoothed_2d_path(subj_num, fwhm_mm))
train_volume_indices = pv.get_train_indices()
test_volume_indices = pv.get_test_indices()
voxel_feature_indices = voxels_sorted_by_t_statistic[:NUM_FEATURES]
X_train = design_matrix.get_design_matrix(train_volume_indices,
voxel_feature_indices)
y_train = np.array(design_matrix.get_labels(train_volume_indices))
X_test = design_matrix.get_design_matrix(test_volume_indices,
voxel_feature_indices)
y_test = np.array(design_matrix.get_labels(test_volume_indices))
model = rf.Classifier(X_train, y_train)
model.train()
y_predicted = model.predict(X_test)
accuracy = accuracy_score(y_test, y_predicted)
print('Validation set accuracy: {0}'.format(accuracy))
output_path = '{0}/figures/validation_accuracy.txt'.format(REPO_HOME_PATH)
np.savetxt(output_path, np.array([accuracy]))
print('Saved {0}'.format(output_path))
def main():
subj_num, fwhm_mm = 1, 4
voxels_sorted_by_t_statistic = lm.VoxelExtractor(subj_num,
'int-ext').t_stat()
design_matrix = dm.DesignMatrix(dp.get_smoothed_2d_path(subj_num, fwhm_mm))
train_volume_indices = pv.get_train_indices()
test_volume_indices = pv.get_test_indices()
voxel_feature_indices = voxels_sorted_by_t_statistic[:NUM_FEATURES]
X_train = design_matrix.get_design_matrix(train_volume_indices,
voxel_feature_indices)
y_train = np.array(design_matrix.get_labels(train_volume_indices))
X_test = design_matrix.get_design_matrix(test_volume_indices,
voxel_feature_indices)
y_test = np.array(design_matrix.get_labels(test_volume_indices))
model = rf.Classifier(X_train, y_train)
model.train()
y_predicted = model.predict(X_test)
accuracy = accuracy_score(y_test, y_predicted)
print('Validation set accuracy: {0}'.format(accuracy))
output_path = '{0}/figures/validation_accuracy.txt'.format(
REPO_HOME_PATH)
np.savetxt(output_path, np.array([accuracy]))
print('Saved {0}'.format(output_path))
vol_rms_diff = np.sqrt(np.mean(diff_data**2, axis=0))
return vol_rms_diff[NUM_OFFSET_VOLUMES:]
def save_plot(vol_rms_diff, subj_num):
"""
Plots the root mean square differences for a particular subject and saves
that plot into the figures folder
Parameters
----------
vol_rms_diff : array
subj_num : int
Returns
-------
None
"""
plt.plot(vol_rms_diff)
plot_path = '{0}/figures/subj{1}_vol_rms_diff.png'.format(REPO_HOME_PATH,
subj_num)
plt.savefig(plot_path)
print('Saved {0}'.format(plot_path))
if __name__ == '__main__':
subj_num, fwhm_mm = 1, 8
vol_rms_diff = calc_vol_rms_diff(dp.get_smoothed_2d_path(subj_num,
fwhm_mm))
save_plot(vol_rms_diff, subj_num)
vol_rms_diff = np.sqrt(np.mean(diff_data**2, axis=0))
return vol_rms_diff[NUM_OFFSET_VOLUMES:]
def save_plot(vol_rms_diff, subj_num):
"""
Plots the root mean square differences for a particular subject and saves
that plot into the figures folder
Parameters
----------
vol_rms_diff : array
subj_num : int
Returns
-------
None
"""
plt.plot(vol_rms_diff)
plot_path = '{0}/figures/subj{1}_vol_rms_diff.png'.format(
REPO_HOME_PATH, subj_num)
plt.savefig(plot_path)
print('Saved {0}'.format(plot_path))
if __name__ == '__main__':
subj_num, fwhm_mm = 1, 8
vol_rms_diff = calc_vol_rms_diff(dp.get_smoothed_2d_path(
subj_num, fwhm_mm))
save_plot(vol_rms_diff, subj_num)
