def generate_mean_grasp(datasets):
smooth_wind_default = 234
means = {}
for perception in datasets:
data_list = datasets[perception]
# get the minimum model length
lens = [len(m) for m in data_list]
max_len = np.max(lens)
ret_means = []
# find the number of coordinates from the first element
num_coords = len(data_list[0][0][1])
for coord in range(num_coords):
mean_models, variance_models = [], []
for stream in data_list:
# extract only the data stream for a single coordinate (all x values)
stream_coord = np.array(zip(*zip(*stream)[1])[coord])
cur_mean_model = signal_smooth(stream_coord, smooth_wind)
mean_models += [cur_mean_model]
# find the average case over the several runs
avg_means_model = np.array([0.] * max_len)
for i in range(max_len):
n = 0
for j in range(len(mean_models)):
if i < len(mean_models[j]):
avg_means_model[i] += mean_models[j][i]
n += 1
avg_means_model[i] /= n
ret_means += [avg_means_model]
means[perception] = np.array(zip(*ret_means))
return means
def split_signals(datasets):
for perception in datasets:
data_list = datasets[perception]
# get the minimum model length
lens = [len(m) for m in data_list]
max_len = np.max(lens)
# dynamic finding of parameters
if smooth_wind_dict is None or smooth_wind_dict[perception] is None:
smooth_wind = smooth_wind_default
else:
smooth_wind = smooth_wind_dict[perception]
ret_means, ret_vars, ret_mean_models, ret_noise_vars = [], [], [], []
ret_times, noise_vars = [], []
# find the number of coordinates from the first element
num_coords = len(data_list[0][0][1])
for coord in range(num_coords):
mean_models, variance_models = [], []
times = None
for stream in data_list:
# extract only the data stream for a single coordinate (all x values)
stream_coord = np.array(zip(*zip(*stream)[1])[coord])
cur_mean_model = signal_smooth(stream_coord, smooth_wind)
mean_models += [cur_mean_model]
# sum up the squared difference over the whole model
noise_vars += [ ( sum([(x - y) ** 2 for x,y in zip(cur_mean_model,stream_coord)]) /
len(cur_mean_model) ) ]
# find the average case over the several runs
avg_means_model = np.array([0.] * max_len)
for i in range(max_len):
n = 0
for j in range(len(mean_models)):
if i < len(mean_models[j]):
avg_means_model[i] += mean_models[j][i]
n += 1
avg_means_model[i] /= n
if var_wind_dict is None or var_wind_dict[perception] is None:
var_wind = var_wind_default
else:
var_wind = var_wind_dict[perception]
# find the variance of the signal but use var_wind points around the centers
# to increase the sample size
vars_model = signal_list_variance(mean_models, avg_means_model, var_wind)
vars_model = signal_smooth(vars_model, var_smooth_wind)
vars_model = signal_smooth(vars_model, var_smooth_wind + 23)
ret_times += [times]
ret_means += [avg_means_model]
ret_vars += [vars_model]
ret_mean_models += [mean_models]
ret_noise_vars += [np.average(noise_vars)]
# TODO deal with timestamp data in some way?
# self.models[perception]["time"] = ret_times
self.models[perception]["mean"] = np.array(zip(*ret_means))
self.models[perception]["variance"] = np.array(zip(*ret_vars))
a = ret_mean_models
b = []
for stream in range(len(a[0])):
t1 = []
for val in range(len(a[0][0])):
t2 = []
for coord in range(len(a)):
if val < len(a[coord][stream]):
t2 += [a[coord][stream][val]]
t1 += [np.array(t2)]
b += [t1]
self.models[perception]["smoothed_signals"] = b
self.models[perception]["noise_variance"] = np.array(ret_noise_vars)
return self.models
