def test_cwt():
# Load/format data
src = os.path.abspath(__file__ + '/../../')+'/demo/demo_data.csv'
raw_data = pd.read_csv(src, skiprows=99, names=['timestamps', 'x', 'y', 'z'], usecols=[0, 1, 2, 3])
raw_data = raw_data.iloc[3500:3700,:]
raw_data['y'] = raw_data['y'] * 9.80665
# Run function being tested
obtained_ic_peaks, obtained_fc_peaks = util._cwt(raw_data.y, 50, 5, 10)
# Confirm expected results
np.testing.assert_array_equal(obtained_ic_peaks, [10, 43, 75, 110, 141, 171])
np.testing.assert_array_equal(obtained_fc_peaks, [19, 50, 84, 117, 148, 179])
def extract_features(self,
subject_height,
subject_height_units='centimeter',
sensor_height_ratio=0.53,
result_file=None,
classified_gait=None,
ic_prom=5,
fc_prom=25):
''' Continuous wavelet transform based method of gait feature detection optimization methods
Parameters
----------
subject_height : int or float
Height of the subject. Accepts centimeters by default.
subject_height_units : str
Units of provided subject height. Centimeters by default.
- options: 'centimeter', 'inches', 'meter'
sensor_height_ratio : float
Height of the sensor relative to subject height. Calculated: sensor height / subject height
result_file : str
Optional argument that accepts .csv filepath string to save resulting gait feature dataframe to.
None by default. (ie. myfolder/myfile.csv)
classified_gait : str or pandas.core.frame.DataFrame
Pandas dataframe containing results of gait bout classification procedure (classify_bouts)
OR
File path of .h5 file containing results of gait bout classification procedure (classify_bouts)
ic_prom : int
Prominance of initial contact peak detection
fc_prom : int
Prominance of final contact peak detection
'''
import pandas as pd
import gaitpy.util as util
import warnings
print('\tExtracting features...')
# Load data
y_accel, timestamps = util._load_data(self, self.down_sample)
# If classified gait is provided, load pandas dataframe or h5 file
if classified_gait is not None:
if type(classified_gait) is str:
gait_predictions = pd.read_hdf(classified_gait)
elif type(classified_gait) is pd.core.frame.DataFrame:
gait_predictions = classified_gait
else:
print(
'Unable to load classified gait: Please make sure the data is in the correct format, aborting...'
)
return
# Isolate gait bouts
gait_windows = gait_predictions[gait_predictions['prediction'] ==
1]
if gait_windows.empty:
print(
'The classified_gait data indicates no bouts of gait were detected, aborting...'
)
return
# Concatenate concurrent bouts
gait_bouts = util._concatenate_windows(gait_windows,
window_length=3)
else:
# if classified_gait is not provided, assume entire timeseries is 1 bout of gait
start_time = timestamps[0].astype('datetime64[ms]')
end_time = timestamps.iloc[-1].astype('datetime64[ms]')
gait_bouts = pd.DataFrame(
data={
'start_time': [start_time],
'end_time': [end_time],
'bout_length': [(end_time -
start_time).item().total_seconds()]
})
all_bout_gait_features = pd.DataFrame()
bout_n = 1
# Loop through gait bouts
for row_n, bout in gait_bouts.iterrows():
bout_indices = (
timestamps.astype('datetime64[ms]') >= bout.start_time) & (
timestamps.astype('datetime64[ms]') <= bout.end_time)
bout_data = pd.DataFrame([])
bout_data['y'] = pd.DataFrame(
y_accel.loc[bout_indices].reset_index(drop=True))
bout_data['ts'] = timestamps.loc[bout_indices].reset_index(
drop=True)
if len(bout_data.y) < 15:
warnings.warn('There are too few data points between ' +
str(bout.start_time) + ' and ' +
str(bout.end_time) + ', skipping bout...')
continue
# Run CWT Gait Model IC and FC detection
ic_peaks, fc_peaks = util._cwt(bout_data.y, self.down_sample,
ic_prom, fc_prom)
# Run gait cycle optimization procedure
pd.options.mode.chained_assignment = None
optimized_gait = util._optimization(bout_data['ts'], ic_peaks,
fc_peaks)
if optimized_gait.empty or 1 not in list(
optimized_gait.Gait_Cycle):
continue
# Calculate changes in height of the center of mass
optimized_gait = util._height_change_com(optimized_gait,
bout_data['ts'],
bout_data['y'],
self.down_sample)
# Calculate gait features
sensor_height = util._calculate_sensor_height(
subject_height, subject_height_units, sensor_height_ratio)
gait_features = util._cwt_feature_extraction(
optimized_gait, sensor_height)
# remove center of mass height and gait cycle boolean columns, remove rows with NAs
gait_features.dropna(inplace=True)
gait_features.drop(['CoM_height', 'Gait_Cycle', 'FC_opp_foot'],
axis=1,
inplace=True)
gait_features.insert(0, 'bout_number', bout_n)
gait_features.insert(1, 'bout_length_sec', bout.bout_length)
gait_features.insert(2, 'bout_start_time', bout.start_time)
gait_features.insert(5, 'gait_cycles', len(gait_features))
all_bout_gait_features = all_bout_gait_features.append(
gait_features)
bout_n += 1
all_bout_gait_features.reset_index(drop=True, inplace=True)
all_bout_gait_features.iloc[:,
7:] = all_bout_gait_features.iloc[:,
7:].round(
3)
# Save results
if result_file:
try:
if not result_file.endswith('.csv'):
result_file += '.csv'
all_bout_gait_features.to_csv(result_file,
index=False,
float_format='%.3f')
except:
print(
'Unable to save data: Please make sure your results directory exists, aborting...'
)
return
if all_bout_gait_features.empty:
print(
'\tFeature extraction complete. No gait cycles detected...\n')
else:
print('\tFeature extraction complete!\n')
return all_bout_gait_features