Update disrpt.py
Browse files
disrpt.py
CHANGED
@@ -1,10 +1,44 @@
|
|
1 |
import datasets
|
2 |
import csv
|
3 |
import requests
|
|
|
|
|
|
|
|
|
4 |
|
5 |
|
6 |
-
|
7 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
8 |
|
9 |
class Config(datasets.BuilderConfig):
|
10 |
citation=citation
|
@@ -38,59 +72,80 @@ files = [
|
|
38 |
"zho.rst.sctb",
|
39 |
]
|
40 |
|
41 |
-
files_2021 = [
|
42 |
-
"deu.rst.pcc",
|
43 |
-
"eng.pdtb.pdtb",
|
44 |
-
"eng.rst.gum",
|
45 |
-
"eng.rst.rstdt",
|
46 |
-
"eng.sdrt.stac",
|
47 |
-
"eus.rst.ert",
|
48 |
-
"fas.rst.prstc",
|
49 |
-
"fra.sdrt.annodis",
|
50 |
-
"fra.sdrt.annodis bug fixes",
|
51 |
-
"nld.rst.nldt",
|
52 |
-
"por.rst.cstn",
|
53 |
-
"rus.rst.rrt",
|
54 |
-
"spa.rst.rststb",
|
55 |
-
"spa.rst.sctb",
|
56 |
-
"tur.pdtb.tdb",
|
57 |
-
"zho.pdtb.cdtb",
|
58 |
-
"zho.rst.sctb",
|
59 |
-
]
|
60 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
61 |
|
62 |
-
|
63 |
-
|
64 |
-
|
|
|
|
|
|
|
65 |
|
66 |
class Dataset(datasets.GeneratorBasedBuilder):
|
67 |
|
68 |
BUILDER_CONFIGS = [
|
69 |
Config(
|
70 |
-
name=n,
|
71 |
-
data_dir=n
|
72 |
-
) for n in files
|
73 |
]
|
74 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
75 |
def _split_generators(self, dl_manager: datasets.DownloadManager):
|
76 |
-
|
77 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
78 |
if train_key in data_file:
|
79 |
train=[datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": data_file[train_key]})]
|
80 |
else:
|
81 |
train=[]
|
82 |
return train+[
|
83 |
-
datasets.SplitGenerator(name=datasets.Split.VALIDATION, gen_kwargs={"filepath": data_file[
|
84 |
-
datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"filepath": data_file[
|
85 |
-
|
86 |
]
|
87 |
|
88 |
-
|
89 |
-
|
|
|
90 |
|
91 |
def _generate_examples(self, filepath):
|
|
|
92 |
with open(filepath, encoding="utf-8") as f:
|
93 |
-
|
|
|
|
|
|
|
|
|
94 |
for id_, row in enumerate(reader):
|
95 |
if id_ == 0:
|
96 |
continue
|
|
|
1 |
import datasets
|
2 |
import csv
|
3 |
import requests
|
4 |
+
import pandas as pd
|
5 |
+
import inspect
|
6 |
+
import copy
|
7 |
+
import ftfy
|
8 |
|
9 |
|
10 |
+
key_to_entry = requests.get('https://www.dropbox.com/scl/fi/85pnc7n6e4puoureavtzo/filtered_disrpt.json?rlkey=6cbgbe9vn2549eths7ah8gm7u&dl=1').json()
|
11 |
+
citation="\n".join(key_to_entry.values())
|
12 |
+
|
13 |
+
datasets_and_citations = {
|
14 |
+
"deu.rst.pcc": "stede-neumann-2014-potsdam",
|
15 |
+
"eng.dep.covdtb": "nishida-matsumoto-2022-domain",
|
16 |
+
"eng.dep.scidtb": "yang-li-2018-scidtb",
|
17 |
+
"eng.rst.gum": "Zeldes2017",
|
18 |
+
"eng.rst.rstdt": "carlson-etal-2001-building",
|
19 |
+
"eng.sdrt.stac": "asher-etal-2016-discourse",
|
20 |
+
"eus.rst.ert": "IruskietaAranzabeIlarrazaEtAl2013",
|
21 |
+
"fas.rst.prstc": "shahmohammadi2021persian",
|
22 |
+
"fra.sdrt.annodis": "afantenos-etal-2012-empirical",
|
23 |
+
"nld.rst.nldt": "redeker-etal-2012-multi",
|
24 |
+
"por.rst.cstn": "CardosoMazieroRosarioCastroJorgeEtAl2011",
|
25 |
+
"rus.rst.rrt": "toldova-etal-2017-rhetorical",
|
26 |
+
"spa.rst.rststb": "da-cunha-etal-2011-development",
|
27 |
+
"spa.rst.sctb": "cao-etal-2018-rst",
|
28 |
+
"zho.dep.scidtb": "yi-etal-2021-unifying,cheng-li-2019-zero",
|
29 |
+
"zho.rst.gcdt": "peng_gcdt_2022,peng_chinese_2022",
|
30 |
+
"zho.rst.sctb": "cao-etal-2018-rst",
|
31 |
+
"eng.pdtb.pdtb": "prasad-etal-2014-reflections",
|
32 |
+
"eng.pdtb.tedm": "zeyrek-etal-2018-multilingual,zeyrek2019ted",
|
33 |
+
"ita.pdtb.luna": "tonelli-etal-2010-annotation,RiccardiStepanovChowdhury2016",
|
34 |
+
"por.pdtb.crpc": "CRPC-DB-Portuguese,genereux-etal-2012-introducing",
|
35 |
+
"por.pdtb.tedm": "zeyrek-etal-2018-multilingual,zeyrek2019ted",
|
36 |
+
"tha.pdtb.tdtb": "",
|
37 |
+
"tur.pdtb.tdb": "zeyrek-webber-2008-discourse,zeyrek-kurfali-2017-tdb",
|
38 |
+
"tur.pdtb.tedm": "zeyrek-etal-2018-multilingual,zeyrek2019ted",
|
39 |
+
"zho.pdtb.cdtb": "Zhou2014"
|
40 |
+
}
|
41 |
+
|
42 |
|
43 |
class Config(datasets.BuilderConfig):
|
44 |
citation=citation
|
|
|
72 |
"zho.rst.sctb",
|
73 |
]
|
74 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
75 |
|
76 |
+
def parse_conll_stream(file_stream):
|
77 |
+
names = ['id', 'form', 'lemma', 'upos', 'xpos', 'feats', 'head', 'deprel', 'deps', 'misc']
|
78 |
+
sentence = {name: [] for name in names}
|
79 |
+
|
80 |
+
for line in file_stream:
|
81 |
+
line = line.strip()
|
82 |
+
if line.startswith("#"):
|
83 |
+
continue
|
84 |
+
if not line:
|
85 |
+
if sentence['id']:
|
86 |
+
yield sentence
|
87 |
+
sentence = {name: [] for name in names}
|
88 |
+
continue
|
89 |
+
token_data = line.split('\t')
|
90 |
+
for name, value in zip(names, token_data):
|
91 |
+
sentence[name].append(value)
|
92 |
|
93 |
+
def get_kwarg_names(func):
|
94 |
+
return [k for k, v in inspect.signature(func).parameters.items() if v.default != v.empty]
|
95 |
+
|
96 |
+
_URLs = {f'{task}-{split}.{type}':f"https://raw.githubusercontent.com/disrpt/sharedtask2023/main/data/{task}/{task}_{split}.{type}" \
|
97 |
+
for task in files for split in 'train dev test'.split() for type in ['rels','conllu']}
|
98 |
+
#_URLs = {k:v for k,v in _URLs.items() if requests.get(v).status_code!=404}
|
99 |
|
100 |
class Dataset(datasets.GeneratorBasedBuilder):
|
101 |
|
102 |
BUILDER_CONFIGS = [
|
103 |
Config(
|
104 |
+
name=f"{n}.{type}",
|
105 |
+
data_dir=f"{n}.{type}"
|
106 |
+
) for n in files for type in ["rels","conllu"]
|
107 |
]
|
108 |
+
def __init__(self,*args,**kwargs):
|
109 |
+
self.BUILDER_CONFIG_CLASS.__post_init__=lambda x:x
|
110 |
+
base_kwargs_names=get_kwarg_names(super().__init__)
|
111 |
+
gen_kwargs={}
|
112 |
+
for k,v in copy.deepcopy(kwargs).items():
|
113 |
+
if k not in base_kwargs_names:
|
114 |
+
gen_kwargs[k]=v
|
115 |
+
del kwargs[k]
|
116 |
+
self.gen_kwargs=gen_kwargs
|
117 |
+
return super().__init__(*args,**kwargs)
|
118 |
+
|
119 |
def _split_generators(self, dl_manager: datasets.DownloadManager):
|
120 |
+
cfg_name = self.config.name.rsplit('.', 1)[0]
|
121 |
+
data_dir = self.config.data_dir.replace(".rels","").replace(".conllu","")
|
122 |
+
print("datadir:",data_dir)
|
123 |
+
type = self.config.name.split('.')[-1]
|
124 |
+
urls={k:v for (k,v) in _URLs.items() if cfg_name in k and requests.get(v).status_code!=404}
|
125 |
+
data_file = dl_manager.download(urls)
|
126 |
+
train_key = data_dir+'-train'
|
127 |
+
print("datafile:",data_file, self.config.data_dir)
|
128 |
if train_key in data_file:
|
129 |
train=[datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": data_file[train_key]})]
|
130 |
else:
|
131 |
train=[]
|
132 |
return train+[
|
133 |
+
datasets.SplitGenerator(name=datasets.Split.VALIDATION, gen_kwargs={"filepath": data_file[data_dir+'-dev.'+type]}),
|
134 |
+
datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"filepath": data_file[data_dir+'-test.'+type]}),
|
|
|
135 |
]
|
136 |
|
137 |
+
def _info(self): return datasets.DatasetInfo(
|
138 |
+
citation=key_to_entry.get(datasets_and_citations.get(self.config.name),None)
|
139 |
+
)
|
140 |
|
141 |
def _generate_examples(self, filepath):
|
142 |
+
print("GEN",self.config.name)
|
143 |
with open(filepath, encoding="utf-8") as f:
|
144 |
+
if "conllu" in self.config.name:
|
145 |
+
stream=parse_conll_stream(f)
|
146 |
+
for i, row in enumerate(stream):
|
147 |
+
yield i,row
|
148 |
+
reader = csv.DictReader(f,delimiter='\t',quoting=csv.QUOTE_NONE)
|
149 |
for id_, row in enumerate(reader):
|
150 |
if id_ == 0:
|
151 |
continue
|