def calculate_label_counts(self, distances, records, train_labels, k, list_of_all_labels):
"""
Calculates the label counts c and c_prim.
"""
# init c and c' per a class:
label_counts = {}
for label in list_of_all_labels:
label_counts[label] = {"c": defaultdict(lambda: 0), "c_prim": defaultdict(lambda: 0)}
# find c and c' per a class:
for ind in records:
distance_vector = distances[ind]
# print "ind, distance_vector, train_labels[ind], k, train_labels:", ind, distance_vector, train_labels[ind], k, train_labels
labels = knn_repetitions_multilabel(
distance_vector, records, train_labels, k + 1
) # k+1 because we will consider the sample itself
neigh_counts = count_distinct_objects(labels)
# substract all the labels of the sample being classified
for self_label in train_labels[ind]:
neigh_counts[self_label] -= 1
oracle_ans = set(train_labels[ind])
for label in list_of_all_labels:
# label in oracle labelling:
if label in oracle_ans:
label_counts[label]["c"][neigh_counts.get(label, 0)] += 1
# label NOT in oracle labelling:
else:
label_counts[label]["c_prim"][neigh_counts.get(label, 0)] += 1
return label_counts
def classify(self, sample_distances):
nearest_labels = knn_repetitions_multilabel(sample_distances,
self.records,
self.train_labels, self.k)
neigh_counts = count_distinct_objects(nearest_labels)
result = []
#for each label check the threshold:
for label in self.list_of_all_labels:
PRINTER(
'[classify]: condidering label, neigh_counts[label], self.fraction_knn_thresholds[label] '
+ str(label) + ", " + str(neigh_counts[label]) + ", " +
str(self.fraction_knn_thresholds[label]))
if neigh_counts[label] >= self.fraction_knn_thresholds[label]:
result.append(label)
PRINTER('[classify]: result ' + str(result))
return result
def classify(self, sample_distances):
nearest_labels = knn_repetitions_multilabel(sample_distances, self.records, self.train_labels, self.k)
neigh_counts = count_distinct_objects(nearest_labels)
result = []
# for each label check the threshold:
for label in self.list_of_all_labels:
PRINTER(
"[classify]: condidering label, neigh_counts[label], self.fraction_knn_thresholds[label] "
+ str(label)
+ ", "
+ str(neigh_counts[label])
+ ", "
+ str(self.fraction_knn_thresholds[label])
)
if neigh_counts[label] >= self.fraction_knn_thresholds[label]:
result.append(label)
PRINTER("[classify]: result " + str(result))
return result
def calculate_label_counts(self, distances, records, train_labels, k,
list_of_all_labels):
'''
Calculates the label counts c and c_prim.
'''
#init c and c' per a class:
label_counts = {}
for label in list_of_all_labels:
label_counts[label] = {
'c': defaultdict(lambda: 0),
'c_prim': defaultdict(lambda: 0)
}
#find c and c' per a class:
for ind in records:
distance_vector = distances[ind]
#print "ind, distance_vector, train_labels[ind], k, train_labels:", ind, distance_vector, train_labels[ind], k, train_labels
labels = knn_repetitions_multilabel(
distance_vector, records, train_labels,
k + 1) #k+1 because we will consider the sample itself
neigh_counts = count_distinct_objects(labels)
#substract all the labels of the sample being classified
for self_label in train_labels[ind]:
neigh_counts[self_label] -= 1
oracle_ans = set(train_labels[ind])
for label in list_of_all_labels:
#label in oracle labelling:
if label in oracle_ans:
label_counts[label]['c'][neigh_counts.get(label, 0)] += 1
#label NOT in oracle labelling:
else:
label_counts[label]['c_prim'][neigh_counts.get(label,
0)] += 1
return label_counts