error in ceafe metric
Opened this issue · 0 comments
yulinchen99 commented
Hi, thanks for releasing the code.
I recently found that there is a potential error in ceafe
metric.
As shown below, when the predicted and gold cluster are identical, the ceafe metric will not output 1.0 as others do.
My piece of code
def get_event2cluster(clusters):
event2cluster = {}
for cluster in clusters:
for eid in cluster:
# if eid not in even2cluster:
event2cluster[eid] = tuple(cluster)
return event2cluster
def evaluate_documents(documents, metric, beta=1):
evaluator = Evaluator(metric, beta=beta)
for document in documents:
evaluator.update(document.clusters, document.gold, document.mention_to_cluster, document.mention_to_gold)
return evaluator.get_precision(), evaluator.get_recall(), evaluator.get_f1()
class Doc:
def __init__(self, mention2cluster, mention2gold, clusters, gold):
self.mention_to_cluster = mention2cluster
self.mention_to_gold = mention2gold
self.clusters = clusters
self.gold = gold
gold = [[1,2,3,4,5], [6,7], [8,9,10, 11,12], [13]]
pred = [[1,2,3,4,5], [6,7], [8,9,10,11,12], [13]]
mention2cluster = get_event2cluster(pred)
mention2gold = get_event2cluster(gold)
doc= Doc(mention2cluster, mention2gold, pred, gold)
p, r, f = evaluate_documents([doc], muc)
print(p, r, f)
p, r, f = evaluate_documents([doc], b_cubed)
print(p, r, f)
p, r, f = evaluate_documents([doc], ceafe)
print(p, r, f)
p, r, f = evaluate_documents([doc], lea)
print(p, r, f)
the output is
1.0 1.0 1.0
1.0 1.0 1.0
1.0 0.75 0.8571428571428571
1.0 1.0 1.0
One way to fix this is to remove the line
Line 120 in 9d1ee19