turk_label = turk[vid][turk_lid]
if len(truth_sorted_lidlist) <= turk_to_truth[index]:
pass
# print "[fp]", turk_label["time"]
# false_positive += 1
# for t in turk_label["points_turk"]:
# print " Turk:", t["time"], t["desc"].encode("utf-8")
else:
true_lid = truth_sorted_lidlist[turk_to_truth[index]]
true_label = truth[vid][true_lid]
distance = math.fabs(
float(turk_label["time"]) - float(true_label["time"]))
signed_distance = float(turk_label["time"]) - float(
true_label["time"])
if found is not 1 and "matched_new" not in true_label and distance <= window_size:
match_result = is_same_label(true_label["desc"],
turk_label["label"])
if match_result[0]:
# if True:
true_label["matched_new"] = True
# print "[m ]", '{:3.2f}'.format(float(match_result[3])), '{:3.2f}'.format(float(turk_label["time"])), true_label["time"], distance, signed_distance""
print "[m ]", '{:3.2f}'.format(
float(turk_label["time"])
), true_label["time"], distance, signed_distance
# print " True:", true_label["desc"]
# print " Turk:", turk_label["label"].encode("utf-8")
found += 1
# else:
# print match_result[3]
# print " True:", true_label["desc"]
# print " Turk:", turk_label["label"].encode("utf-8")
l3 = cluster[2]
final_label = ""
# print l1["vid"], cid
if l1["answer"] == l2["answer"] == l3["answer"]:
# print "[unanimous]", getLabel(l1)
final_label = getLabel(l1)
count_unanimous += 1
if l1["answer"] == "noop":
count_unanimous_noop_included += 1
if l1["answer"] == "@":
count_unanimous_custom_included += 1
elif (l1["answer"] == l2["answer"]) or (l2["answer"] == l3["answer"]) or (l3["answer"] == l1["answer"]):
if l1["answer"] == l2["answer"]:
if l1["answer"] == "@":
count_majority_custom_included += 1
if is_same_label(getLabel(l1), getLabel(l3))[0]:
count_majority_dup_string += 1
answer_label = l1
if l1["answer"] == "noop":
count_majority_noop_included += 1
elif l2["answer"] == l3["answer"]:
if l2["answer"] == "@":
count_majority_custom_included += 1
if is_same_label(getLabel(l2), getLabel(l1))[0]:
count_majority_dup_string += 1
answer_label = l2
if l2["answer"] == "noop":
count_majority_noop_included += 1
elif l3["answer"] == l1["answer"]:
if l3["answer"] == "@":
count_majority_custom_included += 1
def merge_neighbors(clusters, label_turk_list, label_turk_list_desc, sorted_turk):
threshold = 10 # only consider neighbor labels within 10 second distance
clone_clusters = list(clusters)
current_cluster = "no cluster yet"
for (i, label) in enumerate(label_turk_list):
if i == 0: # starting from index 1, because it's about merging i and i-1
continue
key = str(clone_clusters[i])
if current_cluster != key: # new cluster beginning
new_cluster_count = 0
current_cluster = key
# if key != "-1.0": # looking at only valid clusters (this will now include _R and _T)
try:
# skip if i-1 is in the same cluster as i.
# an existing cluster doesn't need merging.
if key == str(clone_clusters[i-1]):
continue
import math
# print "\n", i, (i-1), "in cluster", clone_clusters[i-1], label_turk_list[i], label_turk_list[i-1]
dist = math.fabs(label_turk_list[i] - label_turk_list[i-1])
if i == len(label_turk_list) - 1:
next_dist = 10000
else:
next_dist = math.fabs(label_turk_list[i] - label_turk_list[i+1])
if dist > threshold:
continue
similarity = is_same_label(label_turk_list_desc[i], label_turk_list_desc[i-1])
# to make sure i-1 is more similar to i than to i-2, both in terms of time and label
if i == len(label_turk_list) - 1:
next_similarity = [False, "", "", 0]
else:
next_similarity = is_same_label(label_turk_list_desc[i], label_turk_list_desc[i+1])
# check distance between i and i-1, and between i and i+1
# in order for merge between i and i-1 to happen, i and i-1 should be closer
# print "[testing...]", similarity[3], next_similarity[3], dist, next_dist
similarity_cond = similarity[0] and similarity[3] > next_similarity[3] and dist < next_dist
if not similarity_cond:
# print "[not similar]", similarity[1]
continue
# check if i-1's cluster contains labels by i's Turker, which violates reclustering
tid = sorted_turk[i]["workerid"]
tid_exists = False
for (j, label_inner) in enumerate(label_turk_list):
# only checking i-1's cluster
if i >= j and clone_clusters[j] == clone_clusters[i-1]:
if sorted_turk[j]["workerid"] == tid:
tid_exists = True
break
if tid_exists:
# print "[cannot merge]", "multiple labels from", tid
continue
print "[merge]", i, (i-1), label_turk_list_desc[i], "===", label_turk_list_desc[i-1], dist, next_dist
# case by case based on whether items are singletons or not
prev_key = str(clone_clusters[i-1])
is_singleton = clone_clusters.count(key) == 1
prev_is_singleton = clone_clusters.count(prev_key) == 1
if is_singleton and prev_is_singleton:
clone_clusters[i] = clone_clusters[i-1] # either way works
elif not is_singleton and prev_is_singleton:
clone_clusters[i-1] = clone_clusters[i]
elif is_singleton and not prev_is_singleton:
clone_clusters[i] = clone_clusters[i-1]
else:
# dealing with clusters on both sides.
# merge to the cluster whose centroid is closer to i
print "[clusters both sides]", clone_clusters[i], clone_clusters[i-1]
cent = get_cluster_centroid(key, clone_clusters, label_turk_list)
prev_cent = get_cluster_centroid(prev_key, clone_clusters, label_turk_list)
if math.fabs(label_turk_list[i] - cent) >= math.fabs(label_turk_list[i] - prev_cent):
clone_clusters[i] = clone_clusters[i-1]
else:
clone_clusters[i-1] = clone_clusters[i]
except IndexError: # ignore
print "index error", i
return clone_clusters
valid_turk_count += 1
turk_lid = turk_sorted_lidlist[index]
turk_label = turk[vid][turk_lid]
if len(truth_sorted_lidlist) <= turk_to_truth[index]:
pass
# print "[fp]", turk_label["time"]
# false_positive += 1
# for t in turk_label["points_turk"]:
# print " Turk:", t["time"], t["desc"].encode("utf-8")
else:
true_lid = truth_sorted_lidlist[turk_to_truth[index]]
true_label = truth[vid][true_lid]
distance = math.fabs(float(turk_label["time"]) - float(true_label["time"]))
signed_distance = float(turk_label["time"]) - float(true_label["time"])
if found is not 1 and "matched_new" not in true_label and distance <= window_size:
match_result = is_same_label(true_label["desc"], turk_label["label"])
if match_result[0]:
# if True:
true_label["matched_new"] = True
# print "[m ]", '{:3.2f}'.format(float(match_result[3])), '{:3.2f}'.format(float(turk_label["time"])), true_label["time"], distance, signed_distance""
print "[m ]", '{:3.2f}'.format(float(turk_label["time"])), true_label["time"], distance, signed_distance
# print " True:", true_label["desc"]
# print " Turk:", turk_label["label"].encode("utf-8")
found += 1
# else:
# print match_result[3]
# print " True:", true_label["desc"]
# print " Turk:", turk_label["label"].encode("utf-8")
if found == 1:
match += 1
def merge_neighbors(clusters, label_turk_list, label_turk_list_desc,
sorted_turk):
threshold = 10 # only consider neighbor labels within 10 second distance
clone_clusters = list(clusters)
current_cluster = "no cluster yet"
for (i, label) in enumerate(label_turk_list):
if i == 0: # starting from index 1, because it's about merging i and i-1
continue
key = str(clone_clusters[i])
if current_cluster != key: # new cluster beginning
new_cluster_count = 0
current_cluster = key
# if key != "-1.0": # looking at only valid clusters (this will now include _R and _T)
try:
# skip if i-1 is in the same cluster as i.
# an existing cluster doesn't need merging.
if key == str(clone_clusters[i - 1]):
continue
import math
# print "\n", i, (i-1), "in cluster", clone_clusters[i-1], label_turk_list[i], label_turk_list[i-1]
dist = math.fabs(label_turk_list[i] - label_turk_list[i - 1])
if i == len(label_turk_list) - 1:
next_dist = 10000
else:
next_dist = math.fabs(label_turk_list[i] -
label_turk_list[i + 1])
if dist > threshold:
continue
similarity = is_same_label(label_turk_list_desc[i],
label_turk_list_desc[i - 1])
# to make sure i-1 is more similar to i than to i-2, both in terms of time and label
if i == len(label_turk_list) - 1:
next_similarity = [False, "", "", 0]
else:
next_similarity = is_same_label(label_turk_list_desc[i],
label_turk_list_desc[i + 1])
# check distance between i and i-1, and between i and i+1
# in order for merge between i and i-1 to happen, i and i-1 should be closer
# print "[testing...]", similarity[3], next_similarity[3], dist, next_dist
similarity_cond = similarity[
0] and similarity[3] > next_similarity[3] and dist < next_dist
if not similarity_cond:
# print "[not similar]", similarity[1]
continue
# check if i-1's cluster contains labels by i's Turker, which violates reclustering
tid = sorted_turk[i]["workerid"]
tid_exists = False
for (j, label_inner) in enumerate(label_turk_list):
# only checking i-1's cluster
if i >= j and clone_clusters[j] == clone_clusters[i - 1]:
if sorted_turk[j]["workerid"] == tid:
tid_exists = True
break
if tid_exists:
# print "[cannot merge]", "multiple labels from", tid
continue
print "[merge]", i, (
i - 1), label_turk_list_desc[i], "===", label_turk_list_desc[
i - 1], dist, next_dist
# case by case based on whether items are singletons or not
prev_key = str(clone_clusters[i - 1])
is_singleton = clone_clusters.count(key) == 1
prev_is_singleton = clone_clusters.count(prev_key) == 1
if is_singleton and prev_is_singleton:
clone_clusters[i] = clone_clusters[i - 1] # either way works
elif not is_singleton and prev_is_singleton:
clone_clusters[i - 1] = clone_clusters[i]
elif is_singleton and not prev_is_singleton:
clone_clusters[i] = clone_clusters[i - 1]
else:
# dealing with clusters on both sides.
# merge to the cluster whose centroid is closer to i
print "[clusters both sides]", clone_clusters[
i], clone_clusters[i - 1]
cent = get_cluster_centroid(key, clone_clusters,
label_turk_list)
prev_cent = get_cluster_centroid(prev_key, clone_clusters,
label_turk_list)
if math.fabs(label_turk_list[i] -
cent) >= math.fabs(label_turk_list[i] -
prev_cent):
clone_clusters[i] = clone_clusters[i - 1]
else:
clone_clusters[i - 1] = clone_clusters[i]
except IndexError: # ignore
print "index error", i
return clone_clusters
