Jiann commited on
Commit
d19772c
·
verified ·
1 Parent(s): f5a0329

Upload 6 files

Browse files
plotcom/.DS_Store ADDED
Binary file (6.15 kB). View file
 
plotcom/README.md ADDED
@@ -0,0 +1,36 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Plot Completion Dataset
2
+
3
+ ### Data Example
4
+
5
+ ```
6
+ {
7
+ "story":
8
+ "在一个金碧辉煌的天国王宫里住着两位公主,她们都非常的美丽善良。<MASK>大王就跟国王说,你要是不把你的女儿交出来我就杀了你,正当国王在想办法的时候,一个白马王子出现了,他直接把山寨的大王杀了,跟国王说,我要娶两位公主,国王也同意让女儿嫁了白马王子。从此以后,两位天国的公主与白马王子过上了幸福的日子。",
9
+ "plot":
10
+ "有一次,山寨的大王要来天国王宫里提亲,可是国王并不同意把自己的女儿嫁给山寨的大王做山寨夫人。"
11
+ }
12
+ ```
13
+
14
+ - "story" (`str`):input story,`<MASK>` means the position of the removed sentence.
15
+ - "plot" (`str`):the removed sentence.
16
+
17
+
18
+
19
+ ### Evaluation
20
+
21
+ The prediction result should have the same format with `test.jsonl`
22
+
23
+ ```shell
24
+ python eval.py prediction_file test.jsonl
25
+ ```
26
+
27
+
28
+
29
+ We use bleu and distinct as the evaluation metrics. The output of the script `eval.py` is a dictionary as follows:
30
+
31
+ ```python
32
+ {'bleu-1': '_', 'bleu-2': '_', 'bleu-3': '_', 'bleu-4': '_', 'distinct-1': '_', 'distinct-2': '_', 'distinct-3': '_', 'distinct-4': '_'}
33
+ ```
34
+
35
+ - Dependencies: jieba=0.42.1, nltk=3.6.2, numpy=1.20.3
36
+
plotcom/eval.py ADDED
@@ -0,0 +1,91 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import json
2
+ import argparse
3
+ import sys
4
+ import numpy as np
5
+ import jieba
6
+ import nltk
7
+ from nltk.translate.bleu_score import sentence_bleu, SmoothingFunction
8
+ from nltk import ngrams
9
+
10
+ def bleu(data):
11
+ """
12
+ compute rouge score
13
+ Args:
14
+ data (list of dict including reference and candidate):
15
+ Returns:
16
+ res (dict of list of scores): rouge score
17
+ """
18
+
19
+ res = {}
20
+ for i in range(1, 5):
21
+ res["sentence-bleu-%d"%i] = []
22
+ res["corpus-bleu-%d"%i] = nltk.translate.bleu_score.corpus_bleu([[d["reference"].strip().split()] for d in data], [d["candidate"].strip().split() for d in data], weights=tuple([1./i for j in range(i)]))
23
+ for tmp_data in data:
24
+ origin_candidate = tmp_data['candidate']
25
+ origin_reference = tmp_data['reference']
26
+ assert isinstance(origin_candidate, str)
27
+ if not isinstance(origin_reference, list):
28
+ origin_reference = [origin_reference]
29
+
30
+ for i in range(1, 5):
31
+ res["sentence-bleu-%d"%i].append(sentence_bleu(references=[r.strip().split() for r in origin_reference], hypothesis=origin_candidate.strip().split(), weights=tuple([1./i for j in range(i)])))
32
+
33
+ for key in res:
34
+ if "sentence" in key:
35
+ res[key] = np.mean(res[key])
36
+
37
+ return res
38
+
39
+
40
+
41
+ def distinct(eval_data):
42
+ result = {}
43
+ for i in range(1, 5):
44
+ all_ngram, all_ngram_num = {}, 0.
45
+ for k, tmp_data in enumerate(eval_data):
46
+ ngs = ["_".join(c) for c in ngrams(tmp_data["candidate"].strip().split(), i)]
47
+ all_ngram_num += len(ngs)
48
+ for s in ngs:
49
+ if s in all_ngram:
50
+ all_ngram[s] += 1
51
+ else:
52
+ all_ngram[s] = 1
53
+ result["distinct-%d"%i] = len(all_ngram) / float(all_ngram_num)
54
+ return result
55
+
56
+
57
+
58
+ def load_file(filename):
59
+ data = []
60
+ with open(filename, "r") as f:
61
+ for line in f.readlines():
62
+ data.append(json.loads(line))
63
+ f.close()
64
+ return data
65
+
66
+ def proline(line):
67
+ return " ".join([w for w in jieba.cut("".join(line.strip().split()))])
68
+
69
+
70
+ def compute(golden_file, pred_file, return_dict=True):
71
+ golden_data = load_file(golden_file)
72
+ pred_data = load_file(pred_file)
73
+
74
+ if len(golden_data) != len(pred_data):
75
+ raise RuntimeError("Wrong Predictions")
76
+
77
+ eval_data = [{"reference": proline(g["plot"]), "candidate": proline(p["plot"])} for g, p in zip(golden_data, pred_data)]
78
+ res = bleu(eval_data)
79
+ res.update(distinct(eval_data))
80
+ for key in res:
81
+ res[key] = "_"
82
+ return res
83
+
84
+ def main():
85
+ argv = sys.argv
86
+ print("预测结果:{}, 测试集: {}".format(argv[1], argv[2]))
87
+ print(compute(argv[2], argv[1]))
88
+
89
+
90
+ if __name__ == '__main__':
91
+ main()
plotcom/test.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
plotcom/train.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
plotcom/val.jsonl ADDED
The diff for this file is too large to render. See raw diff