From c0e80d5ad689a131270a42b4170d79561ddbdd45 Mon Sep 17 00:00:00 2001
From: zz-jacob <zhangz.goal@gmail.com>
Date: Thu, 21 Apr 2022 17:36:46 +0800
Subject: [PATCH] use sacrebleu.corpus_bleu()
---
convlab2/nlg/evaluate_unified_datasets.py | 8 +++++---
1 file changed, 5 insertions(+), 3 deletions(-)
diff --git a/convlab2/nlg/evaluate_unified_datasets.py b/convlab2/nlg/evaluate_unified_datasets.py
index cb36a82e..68c48932 100644
--- a/convlab2/nlg/evaluate_unified_datasets.py
+++ b/convlab2/nlg/evaluate_unified_datasets.py
@@ -1,5 +1,6 @@
import sys
from nltk.translate.bleu_score import corpus_bleu
+import sacrebleu
from nltk.tokenize import word_tokenize
sys.path.append('../..')
import json
@@ -32,9 +33,10 @@ def evaluate(predict_result, ontology):
references = []
candidates = []
for i in range(len(predict_result)):
- references.append([word_tokenize(predict_result[i]['utterance'])])
- candidates.append(word_tokenize(predict_result[i]['prediction']))
- metrics['bleu'] = corpus_bleu(references, candidates)
+ references.append(predict_result[i]['utterance'])
+ candidates.append(predict_result[i]['prediction'])
+ # metrics['bleu'] = corpus_bleu(references, candidates)
+ metrics['bleu'] = sacrebleu.corpus_bleu(candidates, [references], lowercase=True).score
# ERROR Rate
## get all values in ontology
--
GitLab