asahi417 commited on
Commit
f07159f
1 Parent(s): 592d989
Files changed (3) hide show
  1. conceptnet.py +72 -0
  2. get_stats.py +30 -0
  3. process.py +30 -20
conceptnet.py CHANGED
@@ -0,0 +1,72 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import json
2
+ import datasets
3
+
4
+ logger = datasets.logging.get_logger(__name__)
5
+ _DESCRIPTION = """[ConceptNet5](https://ojs.aaai.org/index.php/AAAI/article/view/11164)"""
6
+ _NAME = "conceptnet"
7
+ _VERSION = "1.0.0"
8
+ _CITATION = """
9
+ @inproceedings{speer2017conceptnet,
10
+ title={Conceptnet 5.5: An open multilingual graph of general knowledge},
11
+ author={Speer, Robyn and Chin, Joshua and Havasi, Catherine},
12
+ booktitle={Thirty-first AAAI conference on artificial intelligence},
13
+ year={2017}
14
+ }
15
+ """
16
+
17
+ _HOME_PAGE = "https://github.com/asahi417/relbert"
18
+ _URL = f'https://huggingface.co/datasets/relbert/{_NAME}/raw/main/dataset'
19
+ _URLS = {
20
+ str(datasets.Split.TRAIN): [f'{_URL}/train.jsonl'],
21
+ str(datasets.Split.VALIDATION): [f'{_URL}/valid.jsonl'],
22
+ }
23
+
24
+
25
+ class ConceptNetHighConfidenceConfig(datasets.BuilderConfig):
26
+ """BuilderConfig"""
27
+
28
+ def __init__(self, **kwargs):
29
+ """BuilderConfig.
30
+ Args:
31
+ **kwargs: keyword arguments forwarded to super.
32
+ """
33
+ super(ConceptNetHighConfidenceConfig, self).__init__(**kwargs)
34
+
35
+
36
+ class ConceptNetHighConfidence(datasets.GeneratorBasedBuilder):
37
+ """Dataset."""
38
+
39
+ BUILDER_CONFIGS = [
40
+ ConceptNetHighConfidenceConfig(name=_NAME, version=datasets.Version(_VERSION), description=_DESCRIPTION),
41
+ ]
42
+
43
+ def _split_generators(self, dl_manager):
44
+ downloaded_file = dl_manager.download_and_extract(_URLS)
45
+ return [datasets.SplitGenerator(name=i, gen_kwargs={"filepaths": downloaded_file[str(i)]})
46
+ for i in [datasets.Split.TRAIN, datasets.Split.VALIDATION]]
47
+
48
+ def _generate_examples(self, filepaths):
49
+ _key = 0
50
+ for filepath in filepaths:
51
+ logger.info(f"generating examples from = {filepath}")
52
+ with open(filepath, encoding="utf-8") as f:
53
+ _list = [i for i in f.read().split('\n') if len(i) > 0]
54
+ for i in _list:
55
+ data = json.loads(i)
56
+ yield _key, data
57
+ _key += 1
58
+
59
+ def _info(self):
60
+ return datasets.DatasetInfo(
61
+ description=_DESCRIPTION,
62
+ features=datasets.Features(
63
+ {
64
+ "relation_type": datasets.Value("string"),
65
+ "positives": datasets.Sequence(datasets.Sequence(datasets.Value("string"))),
66
+ "negatives": datasets.Sequence(datasets.Sequence(datasets.Value("string"))),
67
+ }
68
+ ),
69
+ supervised_keys=None,
70
+ homepage=_HOME_PAGE,
71
+ citation=_CITATION,
72
+ )
get_stats.py ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import pandas as pd
2
+ from datasets import load_dataset
3
+
4
+ data = load_dataset('relbert/conceptnet')
5
+ stats = []
6
+ for k in data.keys():
7
+ for i in data[k]:
8
+ stats.append({'relation_type': i['relation_type'], 'split': k, 'positives': len(i['positives']), 'negatives': len(i['negatives'])})
9
+ df = pd.DataFrame(stats)
10
+ df_train = df[df['split'] == 'train']
11
+ df_valid = df[df['split'] == 'validation']
12
+ stats = []
13
+ for r in df['relation_type'].unique():
14
+ _df_t = df_train[df_train['relation_type'] == r]
15
+ _df_v = df_valid[df_valid['relation_type'] == r]
16
+ stats.append({
17
+ 'relation_type': r,
18
+ 'positive (train)': 0 if len(_df_t) == 0 else _df_t['positives'].values[0],
19
+ 'negative (train)': 0 if len(_df_t) == 0 else _df_t['negatives'].values[0],
20
+ 'positive (validation)': 0 if len(_df_v) == 0 else _df_v['positives'].values[0],
21
+ 'negative (validation)': 0 if len(_df_v) == 0 else _df_v['negatives'].values[0],
22
+ })
23
+
24
+ df = pd.DataFrame(stats).sort_values(by=['relation_type'])
25
+ df.index = df.pop('relation_type')
26
+ df.to_csv('stats.csv')
27
+ with open('stats.md', 'w') as f:
28
+ f.write(df.to_markdown())
29
+
30
+
process.py CHANGED
@@ -1,7 +1,7 @@
1
  import os
2
  import json
3
  from tqdm import tqdm
4
-
5
  from datasets import load_dataset
6
 
7
  export_dir = 'dataset'
@@ -30,23 +30,33 @@ def check(example):
30
 
31
  dataset = dataset.filter(lambda example: check(example))
32
  relations = list(set(dataset["rel"]))
 
 
 
 
 
 
 
 
 
 
 
 
33
 
34
- for r in tqdm(relations):
35
-
36
- _dataset = dataset.filter(lambda example: example['rel'] == r)
37
- _dataset = _dataset.shuffle(0)
38
- train_size = int(len(_dataset) * 0.7)
39
-
40
- with open(f"{export_dir}/train.{os.path.basename(r)}.jsonl", 'w') as f:
41
- f.write(json.dumps({
42
- 'relation_type': os.path.basename(r),
43
- 'positives': [[i['arg1'], i['arg2']] for i in _dataset[:train_size]],
44
- 'negatives': []
45
- }))
46
-
47
- with open(f"{export_dir}/valid.{os.path.basename(r)}.jsonl", 'w') as f:
48
- f.write(json.dumps({
49
- 'relation_type': os.path.basename(r),
50
- 'positives': [[i['arg1'], i['arg2']] for i in _dataset[train_size:]],
51
- 'negatives': []
52
- }))
 
1
  import os
2
  import json
3
  from tqdm import tqdm
4
+ import numpy as np
5
  from datasets import load_dataset
6
 
7
  export_dir = 'dataset'
 
30
 
31
  dataset = dataset.filter(lambda example: check(example))
32
  relations = list(set(dataset["rel"]))
33
+ all_word = [os.path.basename(i) for i in dataset['arg1'] + dataset['arg2']]
34
+ t, c = np.unique(all_word, return_counts=True)
35
+ freq = {_t: _c for _t, _c in zip(t, c)}
36
+
37
+
38
+ def freq_filter(example): # filter by entity frequency
39
+ if freq[os.path.basename(example['arg1'])] < 5:
40
+ return False
41
+ if freq[os.path.basename(example['arg2'])] < 5:
42
+ return False
43
+ return True
44
+
45
 
46
+ with open(f"{export_dir}/train.jsonl", 'w') as f_train:
47
+ with open(f"{export_dir}/valid.jsonl", 'w') as f_valid:
48
+ for r in tqdm(relations):
49
+ _dataset = dataset.filter(lambda example: example['rel'] == r).shuffle(0)
50
+ pairs = [[os.path.basename(i['arg1']), os.path.basename(i['arg2'])] for i in _dataset if freq_filter(i)]
51
+ train_size = int(len(_dataset) * 0.7)
52
+ f_train.write(json.dumps({
53
+ 'relation_type': os.path.basename(r),
54
+ 'positives': pairs[:train_size],
55
+ 'negatives': []
56
+ }))
57
+ if len(pairs[train_size:]) > 0:
58
+ f_valid.write(json.dumps({
59
+ 'relation_type': os.path.basename(r),
60
+ 'positives': pairs[train_size:],
61
+ 'negatives': []
62
+ }))