1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
|
{
"cells": [
{
"attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
"# Pre-process data for reference classification\n",
"\n",
"This script's pipeline is as follows:\n",
"\n",
"1. Recover text segments that surround certificate ID for all references in CC dataset\n",
"2. Create a DataFrame `(dgst, cert_id, label, text_segments)` out of the objects\n",
"3. Clean and dump into csv\n",
"4. Check for label noise"
]
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {},
"outputs": [],
"source": [
"from __future__ import annotations\n",
"\n",
"from dataclasses import dataclass\n",
"from sec_certs.dataset import CCDataset\n",
"from sec_certs.sample import CCCertificate\n",
"import spacy\n",
"from sec_certs.utils.parallel_processing import process_parallel\n",
"import pandas as pd\n",
"import json\n",
"\n",
"nlp = spacy.load(\"en_core_web_sm\")\n",
"from pathlib import Path\n",
"\n",
"REPO_ROOT = Path(\"../../../\").resolve()\n",
"\n",
"@dataclass\n",
"class ReferenceRecord:\n",
" \"\"\"\n",
" Intermediate object to hold references for a given certificate together with sensible attributes to be extracted\n",
" for labeling.\n",
" \"\"\"\n",
" certificate: CCCertificate | None\n",
" dgst: str\n",
" cert_id: str\n",
" location: str\n",
" label: str | None = None\n",
" sentences: set[str] | None = None\n",
"\n",
" @staticmethod\n",
" def get_reference_sentences(doc, cert_id: str) -> set[str]:\n",
" \"\"\"\n",
" Return a set of sentences corresponding to the given cert_id for the record\n",
" \"\"\"\n",
" return {sent.text for sent in doc.sents if cert_id in sent.text}\n",
"\n",
" @staticmethod\n",
" def get_cert_references_with_sentences(record: ReferenceRecord) -> set[tuple[str, str, str]]:\n",
" pth_to_read = (\n",
" record.certificate.state.st_txt_path\n",
" if record.location == \"target\"\n",
" else record.certificate.state.report_txt_path\n",
" )\n",
"\n",
" with pth_to_read.open(\"r\") as handle:\n",
" data = handle.read()\n",
"\n",
" result = ReferenceRecord.get_reference_sentences(nlp(data), record.cert_id)\n",
" record.sentences = result if result else None\n",
"\n",
" return record\n",
"\n",
" def to_pandas_tuple(self) -> tuple[str, str, str, str, set[str] | None]:\n",
" return self.dgst, self.cert_id, self.location, self.label, self.sentences\n",
"\n",
"def get_df_from_records(records: list[ReferenceRecord]):\n",
" \"\"\"\n",
" Builds dataframe with [dgst,cert_id,location,reason,sentences] with references from list of ReferenceRecords.\n",
" Reason set to None if not defined. \n",
" \"\"\"\n",
" results = process_parallel(ReferenceRecord.get_cert_references_with_sentences, records, use_threading=False, progress_bar=True)\n",
" return pd.DataFrame.from_records([x.to_pandas_tuple() for x in results], columns=[\"dgst\", \"cert_id\", \"location\", \"label\", \"sentences\"])"
]
},
{
"attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
"## Extract sentences from text files and populate dataframes"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"100%|██████████| 58/58 [00:07<00:00, 8.27it/s]\n",
"100%|██████████| 944/944 [01:06<00:00, 14.12it/s]\n",
"100%|██████████| 2259/2259 [00:32<00:00, 69.22it/s]\n"
]
}
],
"source": [
"# Load annotated references from CSV\n",
"annotations_df = pd.read_csv(REPO_ROOT / \"data/cert_id_eval/random_references.csv\")\n",
"annotations_df = annotations_df.rename(columns={\"id\": \"dgst\", \"reason\": \"label\"})\n",
"annotations_df = annotations_df.loc[annotations_df.label != \"self\"]\n",
"annotations_df.label = annotations_df.label.map(lambda x: x.upper().replace(\" \", \"_\"))\n",
"\n",
"# Load dataset\n",
"# dset = CCDataset.from_web_latest()\n",
"dset = CCDataset.from_json(REPO_ROOT / \"datasets/cc/cc_dataset.json\")\n",
"\n",
"annotated_records = [ReferenceRecord(dset[x.dgst], x.dgst, x.cert_id, x.location, x.label) for x in annotations_df.itertuples(index=False)]\n",
"\n",
"# Reference records without annotations\n",
"target_certs = [x for x in dset if x.heuristics.st_references.directly_referencing and x.state.st_txt_path]\n",
"report_certs = [x for x in dset if x.heuristics.report_references.directly_referencing and x.state.report_txt_path]\n",
"target_records = [ReferenceRecord(x, x.dgst, y, \"target\", None, None) for x in target_certs for y in x.heuristics.st_references.directly_referencing]\n",
"report_records = [ReferenceRecord(x, x.dgst, y, \"report\", None, None) for x in report_certs for y in x.heuristics.report_references.directly_referencing]\n",
"\n",
"# Filter annotated_records from report_records to avoid duplicities\n",
"annotated_keys = {(x.dgst, x.cert_id) for x in annotated_records}\n",
"report_records = [x for x in report_records if (x.dgst, x.cert_id) not in annotated_keys]\n",
"\n",
"df_labeled = get_df_from_records(annotated_records)\n",
"df_targets = get_df_from_records(target_records)\n",
"df_reports = get_df_from_records(report_records)\n",
"df = pd.concat([df_labeled, df_targets, df_reports])"
]
},
{
"attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
"## Process Dataframes and dump two versions into csv\n",
"\n",
"1. Version with `dgst, cert_id, location, single_sentence` as `*_exploded.csv`\n",
"2. Version where all sentences tied to `(dgst, cert_id)` key are merged into `sentences`. Saved as `*_grouped.csv`\n",
"\n",
"*Note*: So far don't work with test dataset"
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {},
"outputs": [],
"source": [
"# Load split labels\n",
"with (REPO_ROOT / \"data/reference_annotations_split/train.json\").open(\"r\") as handle:\n",
" train_digests = json.load(handle)\n",
"\n",
"with (REPO_ROOT / \"data/reference_annotations_split/valid.json\").open(\"r\") as handle:\n",
" valid_digests = json.load(handle)\n",
"\n",
"split_dct = {**dict.fromkeys(train_digests, \"train\"), **dict.fromkeys(valid_digests, \"valid\")}\n",
"\n",
"# Apply filtering\n",
"# TODO: We should investigate the cases when we match no sentence\n",
"df = df.loc[df.sentences.notnull()] \n",
"df[\"split\"] = df.dgst.map(split_dct)\n",
"df = df.loc[df.split.notnull()] # Discard test samples\n",
"\n",
"# TODO: Add language detection\n",
"\n",
"# Aggregate sentences from different sources (target, report) into one row\n",
"df = df.groupby([\"dgst\", \"cert_id\", \"label\", \"split\"], as_index=False)[\"sentences\"].agg({\"sentences\": lambda x: set.union(*x)})\n",
"df.to_csv(REPO_ROOT / \"datasets/reference_classification_dataset_merrged.csv\", sep=';', index=False)"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {},
"outputs": [],
"source": [
"# Check for label noise, i.e., search for instances that have different label of a reference.\n",
"duplicates_df = df[df.duplicated(subset=[\"dgst\", \"cert_id\"], keep=False)]\n",
"if not duplicates_df.empty:\n",
" print(\"Warning, label noise in dataset. I.e. tuples (dgst, cert_id) with inconsistent reason. See `duplicates_df` frame.\")"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "venv",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.8.13"
},
"orig_nbformat": 4,
"vscode": {
"interpreter": {
"hash": "a2ed43df31f510d0b358bd0625493376557b0c4d37aa99c09b398809f951b6a5"
}
}
},
"nbformat": 4,
"nbformat_minor": 2
}
|
