def get_sub_result_df(base_model, input_names, output_names, result_column,
i_sub):
total_sub_df = pd.DataFrame()
start_time = datetime.now()
print('Subject ' + str(i_sub) + ', start at ' +
start_time.strftime('%H:%M:%S'))
all_sub_data = pd.read_csv(ALL_SUB_FILE, index_col=False)
other_sub_data = all_sub_data[all_sub_data['subject_id'] != i_sub]
x_train = other_sub_data[input_names]
y_train = other_sub_data[output_names]
my_evaluator = Evaluation(output_names, base_model)
my_evaluator.train_sklearn(x_train, y_train) # very time consuming
test_sub_data = all_sub_data[all_sub_data['subject_id'] == i_sub]
del all_sub_data
subject_data = SubjectData(PROCESSED_DATA_PATH, i_sub)
# the range have to be defined after subject data to get the diameter
shank_diameter = SubjectData.get_cylinder_diameter(test_sub_data,
'l_shank')
thigh_diameter = SubjectData.get_cylinder_diameter(test_sub_data,
'l_thigh')
for segment_moved in SEGMENT_NAMES:
if i_sub == 0: # for test
print('subject 0, ' + segment_moved)
multi_offset = MultiAxisOffset.get_segment_multi_translation(
segment_moved, shank_diameter, thigh_diameter)
offset_combo_list = multi_offset.get_combos()
offset_combo_len = len(offset_combo_list)
all_offsets_df = multi_offset.get_offset_df()
for speed in SPEEDS:
test_speed_data = test_sub_data[test_sub_data['speed'] == float(
speed)]
height = test_speed_data['height'].as_matrix()[0]
x_test = test_speed_data[input_names]
y_test = test_speed_data[output_names].as_matrix()
R2 = np.zeros([offset_combo_len, len(output_names)])
RMSEs = np.zeros([offset_combo_len, len(output_names)])
NRMSEs = np.zeros([offset_combo_len, len(output_names)])
my_xy_generator = XYGeneratorUni(subject_data, speed, output_names,
input_names)
for i_combo in range(len(offset_combo_list)):
x_test_modified = my_xy_generator.modify_x_segment(
x_test, offset_combo_list[i_combo], height)
y_pred = my_evaluator.evaluate_sklearn(x_test_modified, y_test)
R2[i_combo, :], RMSEs[i_combo, :], NRMSEs[
i_combo, :] = my_evaluator.get_all_scores(y_test, y_pred)
evaluation_result = np.column_stack([R2, RMSEs, NRMSEs])
scores_df = pd.DataFrame(evaluation_result, columns=result_column)
scores_df = scores_df.reset_index(drop=True)
speed_df = pd.concat([all_offsets_df, scores_df], axis=1)
speed_df.insert(loc=0, column='segment', value=segment_moved)
speed_df.insert(loc=0, column='speed', value=str(speed))
speed_df.insert(loc=0, column='subject_id', value=i_sub)
total_sub_df = pd.concat([total_sub_df, speed_df], axis=0)
end_time = datetime.now()
print('Subject ' + str(i_sub) + ', finished at ' +
end_time.strftime('%H:%M:%S'))
return total_sub_df
def get_all_rotation_result(input_names,
output_names,
result_column,
i_sub,
model_name,
error_value=25):
start_time = datetime.now()
print('Subject ' + str(i_sub) + ', start at ' +
start_time.strftime('%H:%M:%S'))
file_path = RESULT_PATH + 'evaluator\\' + model_name + '\\' + model_name + '_subject_' + str(
i_sub) + '.pkl'
with open(file_path, 'rb') as fp:
my_evaluator = pickle.load(fp)
all_sub_data = pd.read_csv(ALL_SUB_FILE, index_col=False)
total_sub_df = pd.DataFrame()
test_sub_data = all_sub_data[all_sub_data['subject_id'] == i_sub]
del all_sub_data
subject_data = SubjectData(PROCESSED_DATA_PATH, i_sub)
segment_combo_class = RotationCombos(error_value=error_value)
combos_list = segment_combo_class.get_segment_combos()
combo_len = len(combos_list)
all_offsets_df = segment_combo_class.get_offset_df()
for speed in SPEEDS:
test_speed_data = test_sub_data[test_sub_data['speed'] == float(speed)]
x_test = test_speed_data[input_names]
y_test = test_speed_data[output_names].as_matrix()
height = test_speed_data['height'].as_matrix()[0]
R2 = np.zeros([combo_len + 1, len(output_names)])
RMSEs = np.zeros([combo_len + 1, len(output_names)])
NRMSEs = np.zeros([combo_len + 1, len(output_names)])
y_pred = my_evaluator.evaluate_sklearn(x_test, y_test)
R2[0, :], RMSEs[0, :], NRMSEs[0, :] = my_evaluator.get_all_scores(
y_test, y_pred)
my_xy_generator = XYGeneratorUni(subject_data, speed, output_names,
input_names)
for i_combo in range(len(combos_list)):
x_test_modified = my_xy_generator.modify_x_all_combined(
x_test, combos_list[i_combo], height)
y_pred = my_evaluator.evaluate_sklearn(x_test_modified, y_test)
R2[i_combo +
1, :], RMSEs[i_combo +
1, :], NRMSEs[i_combo +
1, :] = my_evaluator.get_all_scores(
y_test, y_pred)
evaluation_result = np.column_stack([R2, RMSEs, NRMSEs])
scores_df = pd.DataFrame(evaluation_result, columns=result_column)
scores_df = scores_df.reset_index(drop=True)
speed_df = pd.concat([all_offsets_df, scores_df], axis=1)
speed_df.insert(loc=0, column='speed', value=str(speed))
speed_df.insert(loc=0, column='subject_id', value=i_sub)
total_sub_df = pd.concat([total_sub_df, speed_df], axis=0)
end_time = datetime.now()
print('Subject ' + str(i_sub) + ', finished at ' +
end_time.strftime('%H:%M:%S'))
return total_sub_df
def get_segment_rotation_result(input_names, output_names, result_column,
i_sub, model_name):
start_time = datetime.now()
print('Subject ' + str(i_sub) + ', start at ' +
start_time.strftime('%H:%M:%S'))
file_path = RESULT_PATH + 'evaluator\\' + model_name + '\\' + model_name + '_subject_' + str(
i_sub) + '.pkl'
with open(file_path, 'rb') as fp:
my_evaluator = pickle.load(fp)
all_sub_data = pd.read_csv(ALL_SUB_FILE, index_col=False)
total_sub_df = pd.DataFrame()
test_sub_data = all_sub_data[all_sub_data['subject_id'] == i_sub]
del all_sub_data
subject_data = SubjectData(PROCESSED_DATA_PATH, i_sub)
for segment_moved in SEGMENT_NAMES:
if i_sub == 0: # for test
print('subject 0, ' + segment_moved)
rotation_offsets = OneAxisRotation.get_one_axis_rotation(segment_moved)
all_offsets_df = rotation_offsets.get_offset_df()
rotation_len = len(rotation_offsets)
for speed in SPEEDS:
test_speed_data = test_sub_data[test_sub_data['speed'] == float(
speed)]
height = test_speed_data['height'].as_matrix()[0]
x_test = test_speed_data[input_names]
y_test = test_speed_data[output_names].as_matrix()
R2 = np.zeros([rotation_len, len(output_names)])
RMSEs = np.zeros([rotation_len, len(output_names)])
NRMSEs = np.zeros([rotation_len, len(output_names)])
my_xy_generator = XYGeneratorUni(subject_data, speed, output_names,
input_names)
for i_rotation in range(rotation_len):
x_test_modified = my_xy_generator.modify_x_segment(
x_test, rotation_offsets[i_rotation], height)
y_pred = my_evaluator.evaluate_sklearn(x_test_modified, y_test)
R2[i_rotation, :], RMSEs[i_rotation, :], NRMSEs[
i_rotation, :] = my_evaluator.get_all_scores(
y_test, y_pred)
evaluation_result = np.column_stack([R2, RMSEs, NRMSEs])
scores_df = pd.DataFrame(evaluation_result, columns=result_column)
scores_df = scores_df.reset_index(drop=True)
speed_df = pd.concat([all_offsets_df, scores_df], axis=1)
speed_df.insert(loc=0, column='segment', value=segment_moved)
speed_df.insert(loc=0, column='speed', value=str(speed))
speed_df.insert(loc=0, column='subject_id', value=i_sub)
total_sub_df = pd.concat([total_sub_df, speed_df], axis=0)
end_time = datetime.now()
print('Subject ' + str(i_sub) + ', finished at ' +
end_time.strftime('%H:%M:%S'))
return total_sub_df
# # for test
# SUB_NUM = 2
X_NORM_ALL = True # if true, normalize along all the subjects, if false, normalize each subject independently
Y_NORM = False # if true, normalize along all the subjects, if false, do not do normalize
for X_NORM_ALL in [False]:
for Y_NORM in [True]:
print('X_NORM_ALL: ' + str(X_NORM_ALL) + '\tY_NORM: ' + str(Y_NORM))
x, y = {}, {}
for i_sub in range(SUB_NUM):
subject_data = SubjectData(PROCESSED_DATA_PATH, i_sub)
x_sub, y_sub = {}, {}
for speed in SPEEDS:
my_xy_generator = XYGeneratorUni(subject_data,
SEGMENT_NAMES[0], speed,
output_names, input_names)
x_trial, y_trial = my_xy_generator.get_xy()
x_sub[speed] = x_trial.as_matrix()
y_sub[speed] = y_trial.as_matrix()
if not X_NORM_ALL:
x_scalar.fit(
np.row_stack(
[x_sub[SPEEDS[0]], x_sub[SPEEDS[1]],
x_sub[SPEEDS[2]]]))
for speed in SPEEDS:
x_sub[speed] = x_scalar.transform(x_sub[speed])
x[i_sub] = x_sub
y[i_sub] = y_sub
for i_sub_test in range(1):
my_evaluator.train_sklearn() # very time consuming
for speed in SPEEDS:
test_sub_data = all_sub_data[all_sub_data['subject_id'] == i_sub_test]
test_speed_data = test_sub_data[test_sub_data['speed'] == float(speed)]
height = test_speed_data['height'].as_matrix()[0]
x_test = test_speed_data[input_names]
y_test = test_speed_data[output_names]
my_evaluator.set_x_test(x_test)
my_evaluator.set_y_test(y_test)
my_evaluator.evaluate_sklearn()
scores = np.zeros([combo_len, len(output_names)])
RMSEs = np.zeros([combo_len, len(output_names)])
NRMSEs = np.zeros([combo_len, len(output_names)])
my_xy_generator = XYGeneratorUni(subject_data, speed, output_names,
input_names)
for i_combo in range(len(combos_list)):
x_test_modified = my_xy_generator.modify_x_all_combined(
x_test, combos_list[i_combo], height)
my_evaluator.set_x_test(x_test_modified)
my_evaluator.evaluate_sklearn()
scores[i_combo, :] = my_evaluator.get_scores()
RMSEs[i_combo, :] = my_evaluator.get_RMSE()
NRMSEs[i_combo, :] = my_evaluator.get_NRMSE()
evaluation_result = np.column_stack([scores, RMSEs, NRMSEs])
scores_df = pd.DataFrame(evaluation_result, columns=result_column)
scores_df = scores_df.reset_index(drop=True)
speed_df = pd.concat([all_offsets_df, scores_df], axis=1)
speed_df.insert(loc=0, column='speed', value=str(speed))
speed_df.insert(loc=0, column='subject_id', value=i_sub_test)
total_result_df = pd.concat([total_result_df, speed_df], axis=0)
test_size=0.3,
random_state=2)
test_index = x_test.index
my_evaluator = CrossValidation(output_names, model, n_fold=3)
my_evaluator.set_train(x_train, y_train)
my_evaluator.set_y_test(y_test)
my_evaluator.train_sklearn() # very time consuming
scores = np.zeros([combo_len, len(output_names)])
for i_combo in range(len(offset_list)):
x_modified = np.zeros([0, len(input_names)])
for speed in SPEEDS:
speed_df = sub_data_df[sub_data_df['speed'] == float(speed)]
x_speed = speed_df[input_names]
my_xy_generator = XYGeneratorUni(subject_data, speed, output_names,
input_names)
x_modified_speed = my_xy_generator.modify_x_all_combined(
x_speed, offset_list[i_combo])
x_modified = np.row_stack([x_modified, x_modified_speed])
x_test_modified = pd.DataFrame(x_modified).loc[test_index]
my_evaluator.set_x_test(x_test_modified)
y_pred = my_evaluator.evaluate_sklearn()
scores[i_combo, :] = my_evaluator.get_scores()
scores_df = pd.DataFrame(scores, columns=output_names)
scores_df = scores_df.reset_index(drop=True)
subject_df = pd.concat([all_offsets_df, scores_df], axis=1)
subject_df.insert(loc=0, column='subject_id', value=i_sub)
total_result_df = pd.concat([total_result_df, subject_df], axis=0)
EvaluationUni.save_result(total_result_df, model, input_names, output_names,
def get_segment_translation_result(input_names, output_names, result_column,
i_sub, model_names):
print('one trans')
start_time = datetime.now()
print('Subject ' + str(i_sub) + ', start at ' +
start_time.strftime('%H:%M:%S'))
my_evaluators = []
for model_name in model_names:
file_path = RESULT_PATH + 'evaluator\\' + model_name + '\\' + model_name + '_subject_' + str(
i_sub) + '.pkl'
with open(file_path, 'rb') as fp:
my_evaluators.append(pickle.load(fp))
all_sub_data = pd.read_csv(ALL_SUB_FILE, index_col=False)
total_sub_df = pd.DataFrame()
test_sub_data = all_sub_data[all_sub_data['subject_id'] == i_sub]
del all_sub_data
subject_data = SubjectData(PROCESSED_DATA_PATH, i_sub)
# the range have to be defined after subject data to get the diameter
shank_diameter = SubjectData.get_cylinder_diameter(test_sub_data,
'l_shank')
thigh_diameter = SubjectData.get_cylinder_diameter(test_sub_data,
'l_thigh')
for segment_moved in SEGMENT_NAMES:
# if i_sub == 0: # for test
# print('subject 0, ' + segment_moved)
multi_offset = MultiAxisOffset.get_segment_multi_translation(
segment_moved, shank_diameter, thigh_diameter)
offset_combo_list = multi_offset.get_combos()
offset_combo_len = len(offset_combo_list)
all_offsets_df = multi_offset.get_offset_df()
for speed in SPEEDS:
test_speed_data = test_sub_data[test_sub_data['speed'] == float(
speed)]
height = test_speed_data['height'].as_matrix()[0]
x_test = test_speed_data[input_names]
y_test = test_speed_data[output_names].as_matrix()
R2 = np.zeros(
[offset_combo_len,
len(output_names) * len(my_evaluators)])
RMSEs = np.zeros(
[offset_combo_len,
len(output_names) * len(my_evaluators)])
NRMSEs = np.zeros(
[offset_combo_len,
len(output_names) * len(my_evaluators)])
my_xy_generator = XYGeneratorUni(subject_data, speed, output_names,
input_names)
for i_combo in range(len(offset_combo_list)):
x_test_modified = my_xy_generator.modify_x_segment(
x_test, offset_combo_list[i_combo], height)
for i_eval in range(len(my_evaluators)):
y_pred = my_evaluators[i_eval].evaluate_sklearn(
x_test_modified, y_test)
R2[i_combo, i_eval*len(output_names):(i_eval+1)*len(output_names)], \
RMSEs[i_combo, i_eval*len(output_names):(i_eval+1)*len(output_names)],\
NRMSEs[i_combo, i_eval*len(output_names):(i_eval+1)*len(output_names)] =\
my_evaluators[i_eval].get_all_scores(y_test, y_pred)
evaluation_result = np.column_stack([R2, RMSEs, NRMSEs])
scores_df = pd.DataFrame(evaluation_result, columns=result_column)
scores_df = scores_df.reset_index(drop=True)
speed_df = pd.concat([all_offsets_df, scores_df], axis=1)
speed_df.insert(loc=0, column='segment', value=segment_moved)
speed_df.insert(loc=0, column='speed', value=str(speed))
speed_df.insert(loc=0, column='subject_id', value=i_sub)
speed_df.insert(loc=0, column='experiment', value='one_trans')
total_sub_df = pd.concat([total_sub_df, speed_df], axis=0)
end_time = datetime.now()
print('Subject ' + str(i_sub) + ', finished at ' +
end_time.strftime('%H:%M:%S'))
return total_sub_df
def get_all_trans_rota(input_names,
output_names,
result_column,
i_sub,
model_names,
rotation_value=25):
print('all trans rota')
start_time = datetime.now()
print('Subject ' + str(i_sub) + ', start at ' +
start_time.strftime('%H:%M:%S'))
my_evaluators = []
for model_name in model_names:
file_path = RESULT_PATH + 'evaluator\\' + model_name + '\\' + model_name + '_subject_' + str(
i_sub) + '.pkl'
with open(file_path, 'rb') as fp:
my_evaluators.append(pickle.load(fp))
all_sub_data = pd.read_csv(ALL_SUB_FILE, index_col=False)
total_sub_df = pd.DataFrame()
test_sub_data = all_sub_data[all_sub_data['subject_id'] == i_sub]
del all_sub_data
subject_data = SubjectData(PROCESSED_DATA_PATH, i_sub)
thigh_diameter = SubjectData.get_cylinder_diameter(test_sub_data,
'l_thigh')
shank_diameter = SubjectData.get_cylinder_diameter(test_sub_data,
'l_shank')
combo_value = np.array([[100, 100], [100, 100], [25, 100], [25, 100],
[25, 100], [25, 100], [50, 50], [50, 50]])
segment_combo_class = TransRotaCombos(combo_value,
thigh_diameter,
shank_diameter,
rotation_value=rotation_value)
combos_list = segment_combo_class.get_segment_combos()
combo_len = len(combos_list)
all_offsets_df = segment_combo_class.get_offset_df()
for speed in SPEEDS:
test_speed_data = test_sub_data[test_sub_data['speed'] == float(speed)]
x_test = test_speed_data[input_names]
y_test = test_speed_data[output_names].as_matrix()
height = test_speed_data['height'].as_matrix()[0]
R2 = np.zeros([combo_len + 1, len(output_names) * len(my_evaluators)])
RMSEs = np.zeros(
[combo_len + 1,
len(output_names) * len(my_evaluators)])
NRMSEs = np.zeros(
[combo_len + 1,
len(output_names) * len(my_evaluators)])
# get the first data (no placement error)
for i_eval in range(len(my_evaluators)):
y_pred = my_evaluators[i_eval].evaluate_sklearn(x_test, y_test)
R2[0, i_eval * len(output_names):(i_eval + 1) * len(output_names)], \
RMSEs[0, i_eval * len(output_names):(i_eval + 1) * len(output_names)], \
NRMSEs[0, i_eval * len(output_names):(i_eval + 1) * len(output_names)] = \
my_evaluators[i_eval].get_all_scores(y_test, y_pred)
my_xy_generator = XYGeneratorUni(subject_data, speed, output_names,
input_names)
for i_combo in range(len(combos_list)):
x_test_modified = my_xy_generator.modify_x_all_combined(
x_test, combos_list[i_combo], height)
# get all other data
for i_eval in range(len(my_evaluators)):
y_pred = my_evaluators[i_eval].evaluate_sklearn(
x_test_modified, y_test)
R2[i_combo + 1, i_eval * len(output_names):(i_eval + 1) * len(output_names)], \
RMSEs[i_combo + 1, i_eval * len(output_names):(i_eval + 1) * len(output_names)], \
NRMSEs[i_combo + 1, i_eval * len(output_names):(i_eval + 1) * len(output_names)] = \
my_evaluators[i_eval].get_all_scores(y_test, y_pred)
evaluation_result = np.column_stack([R2, RMSEs, NRMSEs])
scores_df = pd.DataFrame(evaluation_result, columns=result_column)
scores_df = scores_df.reset_index(drop=True)
speed_df = pd.concat([all_offsets_df, scores_df], axis=1)
speed_df.insert(loc=0, column='speed', value=str(speed))
speed_df.insert(loc=0, column='subject_id', value=i_sub)
speed_df.insert(loc=0, column='experiment', value='all_trans_rota')
total_sub_df = pd.concat([total_sub_df, speed_df], axis=0)
end_time = datetime.now()
print('Subject ' + str(i_sub) + ', finished at ' +
end_time.strftime('%H:%M:%S'))
return total_sub_df
def get_combined_result_df(input_names, output_names, result_column,
base_model, i_sub):
start_time = datetime.now()
print('Subject ' + str(i_sub) + ', start at ' +
start_time.strftime('%H:%M:%S'))
all_sub_data = pd.read_csv(ALL_SUB_FILE, index_col=False)
other_sub_data = all_sub_data[all_sub_data['subject_id'] != i_sub]
x_train = other_sub_data[input_names]
y_train = other_sub_data[output_names]
my_evaluator = Evaluation(output_names, base_model)
my_evaluator.train_sklearn(x_train, y_train) # very time consuming
total_sub_df = pd.DataFrame()
test_sub_data = all_sub_data[all_sub_data['subject_id'] == i_sub]
del all_sub_data
subject_data = SubjectData(PROCESSED_DATA_PATH, i_sub)
# the range have to be defined after subject data to get the diameter
thigh_diameter = SubjectData.get_cylinder_diameter(test_sub_data,
'l_thigh')
shank_diameter = SubjectData.get_cylinder_diameter(test_sub_data,
'l_shank')
combo_value = np.array([[100, 100], [100, 100], [25, 100], [25, 100],
[25, 100], [25, 100], [50, 50], [50, 50]])
segment_combo_class = TranslationCombos(combo_value, thigh_diameter,
shank_diameter)
combos_list = segment_combo_class.get_segment_combos()
combo_len = len(combos_list)
all_offsets_df = segment_combo_class.get_offset_df()
for speed in SPEEDS:
test_speed_data = test_sub_data[test_sub_data['speed'] == float(speed)]
x_test = test_speed_data[input_names]
y_test = test_speed_data[output_names].as_matrix()
height = test_speed_data['height'].as_matrix()[0]
R2 = np.zeros([combo_len + 1, len(output_names)])
RMSEs = np.zeros([combo_len + 1, len(output_names)])
NRMSEs = np.zeros([combo_len + 1, len(output_names)])
y_pred = my_evaluator.evaluate_sklearn(x_test, y_test)
R2[0, :], RMSEs[0, :], NRMSEs[0, :] = my_evaluator.get_all_scores(
y_test, y_pred)
my_xy_generator = XYGeneratorUni(subject_data, speed, output_names,
input_names)
for i_combo in range(len(combos_list)):
x_test_modified = my_xy_generator.modify_x_all_combined(
x_test, combos_list[i_combo], height)
y_pred = my_evaluator.evaluate_sklearn(x_test_modified, y_test)
R2[i_combo +
1, :], RMSEs[i_combo +
1, :], NRMSEs[i_combo +
1, :] = my_evaluator.get_all_scores(
y_test, y_pred)
evaluation_result = np.column_stack([R2, RMSEs, NRMSEs])
scores_df = pd.DataFrame(evaluation_result, columns=result_column)
scores_df = scores_df.reset_index(drop=True)
speed_df = pd.concat([all_offsets_df, scores_df], axis=1)
speed_df.insert(loc=0, column='speed', value=str(speed))
speed_df.insert(loc=0, column='subject_id', value=i_sub)
total_sub_df = pd.concat([total_sub_df, speed_df], axis=0)
end_time = datetime.now()
print('Subject ' + str(i_sub) + ', finished at ' +
end_time.strftime('%H:%M:%S'))
return total_sub_df