def label_emission_data(file_to_label, **kwargs):
radius = kwargs['radius'] if 'radius' in kwargs else RADIUS
filename = kwargs['filename'] if 'filename' in kwargs else None
window = kwargs['window'] if 'window' in kwargs else WINDOW
n = kwargs['n'] if 'n' in kwargs else N
distance_only = kwargs[
'distance_only'] if 'distance_only' in kwargs else False
tmpfile = 'matched_files/Rental2Youssef_matched.csv'
#viterbi.run_viterbi(file_to_label,filename=tmpfile, radius=radius, window=window,n=n)
observations = read_observations(file_to_label)
results = read_resulting_path(tmpfile)
labeled = []
for t, obs in enumerate(observations):
possible_segments, prob, point = emission_probability.compute_emission_probabilities(
obs, radius, n)
for i, seg in enumerate(possible_segments):
start_node = get_node_id(seg['way_osm_id'], seg['index_in_way'])
end_node = get_node_id(seg['way_osm_id'], seg['index_in_way'] + 1)
node_ids = (start_node[0], end_node[0])
node_ids2 = (end_node[0], start_node[0])
if ((node_ids == results[t]) or (node_ids2 == results[t])):
if (distance_only):
labeled.append(seg['distance'])
else:
labeled.append((node_ids, seg['distance_score'],
seg['tangent_score'], 1))
else:
if (distance_only == False):
labeled.append((node_ids, seg['distance_score'],
seg['tangent_score'], -1))
if filename is not None:
with open(filename, 'w') as f:
f.write('Label, Distance Score, Tangent Score \n')
for l in labeled:
if (distance_only):
f.write(str(l) + '\n')
else:
f.write(
str(l[3]) + ', ' + str(l[1]) + ', ' + str(l[2]) + '\n')
return
return labeled
def label_emission_data(file_to_label, **kwargs):
radius = kwargs['radius'] if 'radius' in kwargs else RADIUS
filename = kwargs['filename'] if 'filename' in kwargs else None
window = kwargs['window'] if 'window' in kwargs else WINDOW
n = kwargs['n'] if 'n' in kwargs else N
distance_only = kwargs['distance_only'] if 'distance_only' in kwargs else False
tmpfile = 'matched_files/Rental2Youssef_matched.csv'
#viterbi.run_viterbi(file_to_label,filename=tmpfile, radius=radius, window=window,n=n)
observations = read_observations(file_to_label)
results = read_resulting_path(tmpfile)
labeled =[]
for t, obs in enumerate(observations):
possible_segments, prob, point = emission_probability.compute_emission_probabilities(obs,radius,n)
for i,seg in enumerate(possible_segments):
start_node = get_node_id(seg['way_osm_id'],seg['index_in_way'])
end_node = get_node_id(seg['way_osm_id'],seg['index_in_way']+1)
node_ids=(start_node[0],end_node[0])
node_ids2 = (end_node[0],start_node[0])
if ((node_ids == results[t]) or (node_ids2 == results[t])):
if (distance_only):
labeled.append(seg['distance'])
else:
labeled.append((node_ids,seg['distance_score'],seg['tangent_score'], 1))
else:
if (distance_only == False):
labeled.append((node_ids,seg['distance_score'],seg['tangent_score'], -1))
if filename is not None:
with open(filename, 'w') as f:
f.write('Label, Distance Score, Tangent Score \n')
for l in labeled:
if (distance_only):
f.write(str(l) + '\n')
else:
f.write(str(l[3]) + ', ' + str(l[1]) +', '+str(l[2]) + '\n')
return
return labeled
def viterbi(observations, **kwargs):
TRANSITION_PROBS = []
radius = kwargs['radius'] if 'radius' in kwargs else RADIUS
filename = kwargs['filename'] if 'filename' in kwargs else None
window = kwargs['window'] if 'window' in kwargs else WINDOW
n = kwargs['n'] if 'n' in kwargs else N
return_gps = kwargs['return_gps'] if 'return_gps' in kwargs else False
print 'Running viterbi. window size: {0}, max states {1}, max radius {2}'.format(window,n,radius)
result_sequence = []
segments_table = []
probabilities_table = []
segments, emission_probabilities, point = compute_emission_probabilities(observations[0],radius, n)
for i, segment in enumerate(segments):
segments[i]['previous'] = None
segments[i]['direction'] = None
segments_table.append(segments)
probabilities_table.append(emission_probabilities)
for window_idx in range(len(observations) / window + 1):
current_obs = observations[window_idx*window:(window_idx+1)*window]
if (len(current_obs) == 0):
break
for t, obs in enumerate(current_obs):
TRANSITION_PROBS.append([])
if t == 0:
continue
previous_point = point
segments, emission_probabilities, point = compute_emission_probabilities(obs, radius, n)
transition_probabilities = compute_transition_probabilities_training(previous_point,
point,
segments_table[t-1],
segments,
window_idx*window+t,
TRANSITION_PROBS)
segments_table.append([])
probabilities_table.append([])
for i, emission_probability in enumerate(emission_probabilities):
candidates = []
for j, previous_probability in enumerate(probabilities_table[t-1]):
candidates.append(previous_probability * transition_probabilities[j][i] * emission_probability)
idx, highest_probability = max(enumerate(candidates), key=lambda x: x[1])
probabilities_table[t].append(highest_probability)
segments[i]['previous'] = idx
segments[i]['direction'] = utils.calculate_direction(segments_table[t-1][idx], segments[i])
segments_table[t].append(segments[i])
last_idx, last_val = max(enumerate(probabilities_table[t]), key=lambda x: x[1])
idx = last_idx
intermediate_result = []
for _t in range(len(current_obs))[::-1]:
cur = segments_table[_t][idx]
intermediate_result.append(cur)
if _t != 0:
idx = cur['previous']
probabilities_table = [[1]]
segments_table = [[segments_table[t][last_idx]]]
result_sequence = result_sequence + intermediate_result[::-1]
for t, cur in enumerate(result_sequence):
if t == 0 or not TRANSITION_PROBS[t]:
continue
prev = result_sequence[t-1]
prev_str = '{0},{1}'.format(prev['way_osm_id'], prev['index_in_way'])
cur_str = '{0},{1}'.format(cur['way_osm_id'], cur['index_in_way'])
TRANSITION_PROBS[t] = {prev_str : TRANSITION_PROBS[t][prev_str]}
TRANSITION_PROBS[t][prev_str][cur_str][2] = 1
return TRANSITION_PROBS
def viterbi(observations, **kwargs):
radius = kwargs['radius'] if 'radius' in kwargs else RADIUS
filename = kwargs['filename'] if 'filename' in kwargs else None
window = kwargs['window'] if 'window' in kwargs else WINDOW
n = kwargs['n'] if 'n' in kwargs else N
return_gps = kwargs['return_gps'] if 'return_gps' in kwargs else False
print 'Running viterbi. window size: {0}, max states {1}, max radius {2}'.format(window,n,radius)
result_sequence = []
segments_table = []
probabilities_table = []
segments, emission_probabilities, point = compute_emission_probabilities(observations[0],radius, n)
for i, segment in enumerate(segments):
segments[i]['previous'] = None
segments[i]['direction'] = None
segments_table.append(segments)
probabilities_table.append(emission_probabilities)
for window_idx in range(len(observations) / window + 1):
current_obs = observations[window_idx*window:(window_idx+1)*window]
if (len(current_obs) == 0):
break
for t, obs in enumerate(current_obs):
if t == 0:
continue
previous_point = point
segments, emission_probabilities, point = compute_emission_probabilities(obs, radius, n)
transition_probabilities = compute_transition_probabilities(previous_point,
point,
segments_table[t-1],
segments)
segments_table.append([])
probabilities_table.append([])
for i, emission_probability in enumerate(emission_probabilities):
candidates = []
for j, previous_probability in enumerate(probabilities_table[t-1]):
candidates.append(previous_probability * transition_probabilities[j][i] * emission_probability)
idx, highest_probability = max(enumerate(candidates), key=lambda x: x[1])
probabilities_table[t].append(highest_probability)
segments[i]['previous'] = idx
segments[i]['direction'] = utils.calculate_direction(segments_table[t-1][idx], segments[i])
segments_table[t].append(segments[i])
last_idx, last_val = max(enumerate(probabilities_table[t]), key=lambda x: x[1])
idx = last_idx
intermediate_result = []
for _t in range(len(current_obs))[::-1]:
cur = segments_table[_t][idx]
intermediate_result.append(cur)
if _t != 0:
idx = cur['previous']
probabilities_table = [[1]]
segments_table = [[segments_table[t][last_idx]]]
result_sequence = result_sequence + intermediate_result[::-1]
if return_gps:
node_gps = utils.get_node_gps_points(result_sequence)
start_points = ['{0},{1}'.format(point[0][0], point[0][1]) for point in node_gps]
end_points = ['{0},{1}'.format(point[1][0], point[1][1]) for point in node_gps]
with open('result_nodes.csv', 'w') as resf:
for i, point in enumerate(start_points):
resf.write(point+'\n')
resf.write(end_points[i]+'\n')
return
node_ids = utils.get_node_ids(result_sequence)
if filename is not None:
utils.write_to_file(node_ids, filename)
return
return node_ids
