def run_timings_vs_num_landmarks():
np.random.seed(0)
options = simulator_options.copy()
options['num_landmarks'] = 1000
calibration = structures.Calibration.random()
trials = []
true_trajectory, measurements, spline_template, true_frame_timestamps = simulation.simulate_trajectory(
calibration, **options)
all_features = measurements.features
for n in np.linspace(10, 400, 25):
measurements.features = filter(lambda f: f.track_id < n, all_features)
try:
spline_socp.estimate_trajectory(
calibration,
measurements,
spline_template,
estimator='mixed',
feature_tolerance=options['feature_noise']*3)
trials.append((n, socp.timings['last_solve']))
except spline_socp.InsufficientObservationsError:
print 'Simulator failed to generate trajectory. Retrying...'
np.savetxt('results/timings_vs_num_landmarks.txt', trials)
def evaluate(calibration, measurements, spline_template, estimator, tol, true_trajectory):
estimated_trajectory = spline_socp.estimate_trajectory(
calibration,
measurements,
spline_template,
estimator=estimator,
feature_tolerance=tol)
pos_err = mean_position_error(true_trajectory, estimated_trajectory, measurements.frame_timestamps)
vel_err = mean_velocity_error(true_trajectory, estimated_trajectory, measurements.frame_timestamps)
bias_err = accel_bias_error(true_trajectory, estimated_trajectory)
g_err = gravity_direction_error(true_trajectory, estimated_trajectory)
return pos_err, vel_err, bias_err, g_err
def run_timings_vs_num_knots():
np.random.seed(0)
calibration = structures.Calibration.random()
trials = []
true_trajectory, measurements, spline_template, true_frame_timestamps = simulation.simulate_trajectory(
calibration, **simulator_options)
for n in np.arange(2, 21):
spline_template.knots = np.linspace(0, simulator_options['duration'], n)
try:
spline_socp.estimate_trajectory(
calibration,
measurements,
spline_template,
estimator='mixed',
feature_tolerance=simulator_options['feature_noise']*3)
trials.append((n, socp.timings['last_solve']))
except spline_socp.InsufficientObservationsError:
print 'Simulator failed to generate trajectory. Retrying...'
np.savetxt('results/timings_vs_num_knots.txt', trials)