lukemelas commited on
Commit
6fad1a6
1 Parent(s): e42403e

Adding datasets

Browse files
Files changed (4) hide show
  1. .gitignore +2 -1
  2. hupd.py +10 -9
  3. json-files-Jan2016.tar +0 -3
  4. tests/tests.py +179 -0
.gitignore CHANGED
@@ -1 +1,2 @@
1
- tmp
 
1
+ tmp
2
+ *.pyc
hupd.py CHANGED
@@ -119,7 +119,7 @@ class PatentsConfig(datasets.BuilderConfig):
119
  class Patents(datasets.GeneratorBasedBuilder):
120
  _DESCRIPTION
121
 
122
- VERSION = datasets.Version("1.0.1")
123
 
124
  # This is an example of a dataset with multiple configurations.
125
  # If you don't want/need to define several sub-sets in your dataset,
@@ -129,16 +129,16 @@ class Patents(datasets.GeneratorBasedBuilder):
129
  PatentsConfig(
130
  name="sample",
131
  description="Patent data from January 2016, for debugging",
132
- metadata_url="https://huggingface.co/datasets/HUPD/hupd-subset/resolve/main/metadata--Jan2016--2021-11-07.pkl",
133
- data_url="https://huggingface.co/datasets/HUPD/hupd-subset/resolve/main/json-files-Jan2016.tar",
134
- data_dir="json-files-Jan2016",
135
  ),
136
  PatentsConfig(
137
  name="all",
138
- description="Patent data from January 2016, for debugging",
139
- metadata_url="https://patentdiag.blob.core.windows.net/patent-data/metadata-2021-11-07.pkl",
140
- data_url="https://patentdiag.blob.core.windows.net/patent-data/distilled-2021-01-07.tar",
141
- data_dir="distilled",
142
  ),
143
  ]
144
 
@@ -274,8 +274,9 @@ class Patents(datasets.GeneratorBasedBuilder):
274
  for id_, x in enumerate(df.itertuples()):
275
 
276
  # JSON files are named by application number (unique)
 
277
  application_number = x.application_number
278
- filepath = os.path.join(json_dir, application_number + '.json')
279
  try:
280
  with open(filepath, 'r') as f:
281
  patent = json.load(f)
119
  class Patents(datasets.GeneratorBasedBuilder):
120
  _DESCRIPTION
121
 
122
+ VERSION = datasets.Version("1.0.2")
123
 
124
  # This is an example of a dataset with multiple configurations.
125
  # If you don't want/need to define several sub-sets in your dataset,
129
  PatentsConfig(
130
  name="sample",
131
  description="Patent data from January 2016, for debugging",
132
+ metadata_url="https://huggingface.co/datasets/HUPD/hupd/resolve/main/hupd_metadata_jan16_2022-02-22.feather",
133
+ data_url="https://huggingface.co/datasets/HUPD/hupd/resolve/main/data/sample-jan-2016.tar.gz",
134
+ data_dir="data", # this will unpack to data/sample/2016
135
  ),
136
  PatentsConfig(
137
  name="all",
138
+ description="Patent data from all years (2004-2018)",
139
+ metadata_url="https://huggingface.co/datasets/HUPD/hupd/resolve/main/hupd_metadata_2022-02-22.feather",
140
+ data_url="https://huggingface.co/datasets/HUPD/hupd/resolve/main/data/all-years.tar.gz",
141
+ data_dir="data", # this will unpack to data/{year}
142
  ),
143
  ]
144
 
274
  for id_, x in enumerate(df.itertuples()):
275
 
276
  # JSON files are named by application number (unique)
277
+ application_year = str(x.filing_date.year)
278
  application_number = x.application_number
279
+ filepath = os.path.join(json_dir, application_year, application_number + '.json')
280
  try:
281
  with open(filepath, 'r') as f:
282
  patent = json.load(f)
json-files-Jan2016.tar DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:4a7d7923941e39255112d2b40a40e8dae8579d9150459c1f0599ffe8a4cfb5a5
3
- size 2024540160
 
 
 
tests/tests.py ADDED
@@ -0,0 +1,179 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """
2
+ Dataset loading tests. Run with:
3
+ PYTHONPATH=. pytest tests/tests.py -vvrP
4
+
5
+ Additional notes about pytest:
6
+ - Skip a test with @pytest.mark.skip(reason='skipping')
7
+ - Use `-vvrP` to print stdout
8
+ """
9
+ import pdb
10
+ import os
11
+ from pathlib import Path
12
+ from pprint import pprint
13
+
14
+ import pytest
15
+ import torch
16
+ import torch.nn.functional as F
17
+ import torch.utils.data
18
+ from datasets import load_dataset
19
+
20
+
21
+ def test_dataset_sample():
22
+ """Load the sample dataset"""
23
+ root = os.getcwd()
24
+ dataset_dict = load_dataset(
25
+ 'hupd.py',
26
+ name='sample',
27
+ data_files=os.path.join(root, "hupd_metadata_jan16_2022-02-22.feather"),
28
+ data_dir=os.path.join(root, "data/sample"),
29
+ uniform_split=True
30
+ )
31
+ for name, dataset in dataset_dict.items():
32
+ print(f'Dataset {name}: {len(dataset)}')
33
+ import pdb; pdb.set_trace()
34
+
35
+
36
+ if __name__ == '__main__':
37
+ test_dataset_sample()
38
+
39
+
40
+ # # # ----- Data loading example 1 ------
41
+ # # # To load a dataset from files directly, pass in the
42
+ # # # data_files and data_dir parameters. For example:
43
+
44
+ # # # ----- Data loading example 2 ------
45
+ # # # It is simple to specify an IPCR or CPC label and
46
+ # # # a date range for training/validation. For example:
47
+ # # dataset_dict = load_dataset(
48
+ # # 'patents.py',
49
+ # # data_files="/blob/uspto/data/codebooks/data_link_new.pkl",
50
+ # # data_dir="/blob/uspto/data/distilled",
51
+ # # ipcr_label='G01T', #'G06F',
52
+ # # cpc_label=None,
53
+ # # train_filing_start_date=None,
54
+ # # train_filing_end_date=None,
55
+ # # val_filing_start_date=None,
56
+ # # val_filing_end_date=None,
57
+ # # )
58
+
59
+ # # # ----- Data loading example 3 ------
60
+ # # If you do not specify the data_files and data_dir parameters, the
61
+ # # dataset will be downloaded automatically for you. For example:
62
+ # dataset_dict = load_dataset(
63
+ # 'patents.py',
64
+ # data_dir="/blob/uspto/data/distilled",
65
+ # cache_dir='/blob/data/patents/distilled/distilled/huggingface-dataset/cache',
66
+ # ipcr_label=None, # 'G01T', #'G06F', # cpc_label='G01T',
67
+ # train_filing_start_date='2016-01-01',
68
+ # train_filing_end_date='2016-01-05',
69
+ # val_filing_start_date='2017-01-01',
70
+ # val_filing_end_date='2017-01-05',
71
+ # )
72
+
73
+
74
+ # def combine_two_sections(tokenizer, dataset, s1, s2, new_tokens):
75
+
76
+ # # Add the seperation token
77
+ # if tokenizer.sep_token != '[SEP]':
78
+ # tokenizer.add_tokens(['[SEP]'], special_tokens=True)
79
+ # tokenizer.sep_token = '[SEP]'
80
+
81
+ # print(f'[OLD] len(tokenizer.vocab) = {len(tokenizer)}')
82
+ # tokenizer.add_tokens(new_tokens + [s1.upper(), 'TITLE', 'YEAR', s2.upper()])
83
+ # print(f'[NEW] len(tokenizer.vocab) = {len(tokenizer)}')
84
+ # dataset = dataset.map(
85
+ # # lambda e: {f'{s1}_{s2}': f'[SEP] {s1.upper()} ' + e[s1 + '_label'][:4] + ' [SEP] ' + e[s2]})
86
+ # lambda e: {f'{s1}_{s2}': f'[SEP] TITLE ' + e['title'] + '. YEAR ' + e['filing_date'][:4] + f'. {s1.upper()} ' + e[s1 + '_label'][:4] + f' [SEP] {s2.upper()} ' + e[s2]})
87
+ # return tokenizer, dataset
88
+
89
+
90
+ # def convert_ids_to_string(tokenizer, input):
91
+ # return ' '.join(tokenizer.convert_ids_to_tokens(input))
92
+
93
+
94
+ # conditional = 'ipc'
95
+ # section = 'abstract'
96
+
97
+ # # Print some metadata
98
+ # print('Dataset dictionary contents:')
99
+ # pprint(dataset_dict)
100
+ # print('Dataset dictionary cached to:')
101
+ # pprint(dataset_dict.cache_files)
102
+ # print(f'Train dataset size: {dataset_dict["train"].shape}')
103
+ # print(f'Validation dataset size: {dataset_dict["validation"].shape}')
104
+
105
+ # # Example: preprocess dataset "decision" feature for classification
106
+ # decision_to_str = {
107
+ # 'REJECTED': 0,
108
+ # 'ACCEPTED': 1,
109
+ # 'PENDING': 2,
110
+ # 'CONT-REJECTED': 3,
111
+ # 'CONT-ACCEPTED': 4,
112
+ # 'CONT-PENDING': 5
113
+ # }
114
+
115
+
116
+ # def map_decision_to_string(example):
117
+ # # NOTE: returned dict updates the example
118
+ # return {'decision': decision_to_str[example['decision']]}
119
+
120
+
121
+ # # Performing the remapping means iterating over the dataset
122
+ # # NOTE: This stores the updated table in a cache file indexed
123
+ # # by the current state and the mapping function
124
+ # train_dataset = dataset_dict['train'].map(map_decision_to_string)
125
+ # print('Processed train dataset cached to: ')
126
+ # pprint(train_dataset.cache_files)
127
+
128
+ # # Example: preprocess dataset "abstract" field using huggingface
129
+ # # tokenizers for classification. We truncate at the max token length.
130
+ # from transformers import AutoTokenizer
131
+ # tokenizer = AutoTokenizer.from_pretrained('roberta-base')
132
+
133
+ # # def map_cpc_label(example):
134
+ # # # NOTE: returned dict updates the example
135
+ # # # print(tokenizer.convert_ids_to_tokens(tokenizer.convert_tokens_to_ids(example['cpc_label'][:4])))
136
+ # # return {'cpc_label': tokenizer.convert_tokens_to_ids(example['cpc_label'][:4])}
137
+ # # train_dataset = train_dataset.map(map_cpc_label)
138
+
139
+ # if conditional:
140
+ # f = open(f'{conditional}_labels.txt', 'r')
141
+ # new_tokens = f.read().split('\n')
142
+ # tokenizer, train_dataset = combine_two_sections(tokenizer, train_dataset, conditional, section, new_tokens)
143
+ # section = f'{conditional}_{section}'
144
+
145
+ # # We tokenize in batches, so it is actually quite fast
146
+ # print('Tokenizing')
147
+ # train_dataset = train_dataset.map(
148
+ # lambda e: tokenizer((e[section]), truncation=True, padding='max_length'),
149
+ # batched=True)
150
+ # print('Processed train dataset cached to: ')
151
+ # pprint(train_dataset.cache_files)
152
+ # print('Processed train dataset columns: ')
153
+ # pprint(train_dataset.column_names)
154
+
155
+ # # Convert to PyTorch Dataset
156
+ # # NOTE: If you also want to return string columns (as a list), just
157
+ # # pass `output_all_columns=True` to the dataset
158
+ # train_dataset.set_format(type='torch',
159
+ # columns=['input_ids', 'attention_mask', 'decision'])
160
+
161
+ # # Standard PyTorch DataLoader
162
+ # from torch.utils.data import DataLoader
163
+ # train_dataloader = DataLoader(train_dataset, batch_size=16)
164
+ # print('Shapes of items in batch from standard PyTorch DataLoader:')
165
+ # pprint({k: v.shape for k, v in next(iter(train_dataloader)).items()})
166
+ # print('Batch from standard PyTorch DataLoader:')
167
+ # batch = next(iter(train_dataloader))
168
+ # pprint(batch['input_ids'])
169
+ # pprint(batch['decision'])
170
+
171
+ # # Print examples
172
+ # print(convert_ids_to_string(tokenizer, batch['input_ids'][0]))
173
+ # pprint(batch['input_ids'][0][:20])
174
+ # # vocab = batch['input_ids'][0][:20]
175
+ # # for elt in vocab:
176
+ # # print(f'{elt}: {convert_ids_to_string(tokenizer, [elt])}')
177
+ # print(tokenizer.decode(batch['input_ids'][0]))
178
+
179
+ # print('All done')