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
|
"""
Simple script to perform hyperparameter search over various parameters of SentenceTransformer trained for reference
meaning classification.
"""
from __future__ import annotations
import os
from functools import partial
from pathlib import Path
import click
import optuna
import pandas as pd
import torch
from rapidfuzz import fuzz
from sec_certs.model.references.annotator_trainer import ReferenceAnnotatorTrainer
from sec_certs.model.references.segment_extractor import ReferenceSegmentExtractor
from sklearn.metrics import f1_score
from sec_certs.dataset import CCDataset
from sec_certs.utils.helpers import compute_heuristics_version
from sec_certs.utils.nlp import prec_recall_metric
def replace_all(text: str, to_replce: set[str]) -> str:
for i in to_replce:
text = text.replace(i, "")
return text
def load_dataset(dataset_path: Path, annotations_dir: Path) -> tuple[CCDataset, pd.DataFrame]:
"""
Load dataset from dataset_path and annotations_dir
:param dataset_path: path to dataset
:param annotations_dir: path to annotations
:return: pd.DataFrame
"""
train_annotations = pd.read_csv(annotations_dir / "train.csv")
valid_annotations = pd.read_csv(annotations_dir / "valid.csv")
all_annotations = pd.concat([train_annotations, valid_annotations])
all_annotations = all_annotations[all_annotations.label != "None"].assign(label=lambda df: df.label.str.upper())
dset = CCDataset.from_json(dataset_path)
all_certs = {x.dgst: x for x in dset.certs.values()}
dset.certs = {x.dgst: x for x in dset.certs.values() if x.dgst in all_annotations.dgst.unique()}
cert_id_to_name_mapping = {x.heuristics.cert_id: x.name for x in all_certs.values()}
all_annotations["referenced_cert_name"] = all_annotations["referenced_cert_id"].map(cert_id_to_name_mapping)
all_annotations["cert_name"] = all_annotations["dgst"].map(lambda x: dset[x].name)
all_annotations["cert_versions"] = all_annotations["cert_name"].map(compute_heuristics_version)
all_annotations = all_annotations.loc[all_annotations["referenced_cert_name"].notnull()].copy()
all_annotations["referenced_cert_versions"] = all_annotations["referenced_cert_name"].map(
compute_heuristics_version
)
all_annotations["cert_name_stripped_version"] = all_annotations.apply(
lambda x: replace_all(x["cert_name"], x["cert_versions"]), axis=1
)
all_annotations["referenced_cert_name_stripped_version"] = all_annotations.apply(
lambda x: replace_all(x["referenced_cert_name"], x["referenced_cert_versions"]), axis=1
)
all_annotations["name_similarity"] = all_annotations.apply(
lambda x: fuzz.token_set_ratio(x["cert_name"], x["referenced_cert_name"]), axis=1
)
all_annotations["name_similarity_stripped_version"] = all_annotations.apply(
lambda x: fuzz.token_set_ratio(x["cert_name_stripped_version"], x["referenced_cert_name_stripped_version"]),
axis=1,
)
all_annotations["name_len_diff"] = all_annotations.apply(
lambda x: abs(len(x["cert_name_stripped_version"]) - len(x["referenced_cert_name_stripped_version"])), axis=1
)
return dset, all_annotations
def preprocess_data(dset: CCDataset, df: pd.DataFrame) -> pd.DataFrame:
"""
Preprocess data
:param df: pd.DataFrame
:return: pd.DataFrame
"""
def process_segment(segment: str, referenced_cert_id: str) -> str:
segment = segment.replace(referenced_cert_id, "the referenced product")
return segment
new_df = ReferenceSegmentExtractor()(dset.certs.values())
new_df = new_df.loc[new_df.label.notnull()].copy()
new_df = new_df.merge(
df.loc[
:,
[
"dgst",
"referenced_cert_id",
"name_similarity_stripped_version",
"name_len_diff",
"cert_name",
"referenced_cert_name",
],
],
on=["dgst", "referenced_cert_id"],
)
new_df.segments = new_df.apply(
lambda row: [process_segment(x, row.referenced_cert_id) for x in row.segments], axis=1
)
return new_df
def define_trainer(trial: optuna.trial.Trial, df: pd.DataFrame) -> ReferenceAnnotatorTrainer:
use_analytical_rule_name_similarity = trial.suggest_categorical(
"use_analytical_rule_name_similarity", [True, False]
)
n_iterations = trial.suggest_int("n_iterations", 1, 50)
n_epochs = trial.suggest_int("n_epochs", 1, 5)
batch_size = trial.suggest_int("batch_size", 8, 32)
segmenter_metric = trial.suggest_categorical("segmenter_metric", ["accuracy", "f1"])
ensemble_soft_voting_power = trial.suggest_int("ensemble_soft_voting_power", 1, 5)
return ReferenceAnnotatorTrainer.from_df(
df,
prec_recall_metric,
mode="training",
use_analytical_rule_name_similarity=use_analytical_rule_name_similarity,
n_iterations=n_iterations,
n_epochs=n_epochs,
batch_size=batch_size,
segmenter_metric=segmenter_metric,
ensemble_soft_voting_power=ensemble_soft_voting_power,
)
def objective(trial: optuna.trial.Trial, df: pd.DataFrame):
trainer = define_trainer(trial, df)
trainer.train()
annotator = trainer.clf
df_predicted = annotator.predict_df(df)
return f1_score(
df_predicted.loc[df_predicted.split == "valid", ["y_pred"]],
df_predicted.loc[df_predicted.split == "valid", ["label"]],
zero_division="warn",
average="weighted",
)
@click.command()
@click.option("-n", "--trials", "trials", type=int, required=True, help="Number of optimization trials to run.")
@click.option(
"-d",
"--dataset",
"dataset_path",
type=click.Path(exists=True, dir_okay=False, file_okay=True, readable=True),
required=True,
help="Path to CCDataset json.",
)
@click.option(
"-a",
"--annotations",
"annotations_dir",
type=click.Path(exists=True, dir_okay=True, file_okay=False, readable=True),
required=True,
help="Path to annotations directory.",
)
@click.option(
"-o",
"--output",
"output_dir",
type=click.Path(exists=True, dir_okay=True, file_okay=False, readable=True),
required=True,
help="Path to output directory.",
)
@click.option("-t", "--timeout", "timeout", type=int, default=24, help="Timeout in hours", show_default=True)
def main(trials: int, dataset_path: Path, annotations_dir: Path, output_dir: Path, timeout: int):
if not torch.cuda.is_available():
print("GPU is not available, exiting. Did you set `CUDA_VISIBLE_DEVICES` environment variable properly?")
return -1
if os.environ.get("TOKENIZERS_PARALLELISM", True) != "FALSE":
print(
"Tokenizers parallelism not disabled for spacy, exiting. Did you set `TOKENIZERS_PARALLELISM` environment variable to `FALSE`?"
)
return -1
# os.environ["CUDA_VISIBLE_DEVICES"] = "MIG-56c53afb-6f08-5e5b-83fa-32fc6f09eeb0"
# os.environ["TOKENIZERS_PARALLELISM"] = "FALSE"
dataset_path = Path(dataset_path)
annotations_dir = Path(annotations_dir)
output_dir = Path(output_dir)
print("Loading dataset...")
cc_dset, df = load_dataset(dataset_path, annotations_dir)
print("Preprocessing data...")
df_processed = preprocess_data(cc_dset, df)
partial_objective = partial(objective, df=df_processed)
print("Starting hyperparameter search...")
study = optuna.create_study(direction="maximize")
study.optimize(partial_objective, n_trials=trials, timeout=60 * 60 * timeout)
study.trials_dataframe().to_csv(output_dir / "hyperparameter_search.csv")
ax = optuna.visualization.matplotlib.plot_optimization_history(study)
ax.figure.savefig(output_dir / "optimization_history.pdf", bbox_inches="tight")
ax = optuna.visualization.matplotlib.plot_param_importances(study)
ax.figure.savefig(output_dir / "param_importances.pdf", bbox_inches="tight")
ax = optuna.visualization.matplotlib.plot_timeline(study)
ax.figure.savefig(output_dir / "timeline.pdf", bbox_inches="tight")
print("Done.")
if __name__ == "__main__":
main()
|
