-
Notifications
You must be signed in to change notification settings - Fork 0
/
build_topics_2.py
249 lines (211 loc) · 9.61 KB
/
build_topics_2.py
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
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
import datetime
import gc
import os
import shutil
import uuid
from contextlib import closing
import google.cloud.bigquery as bq
import google.cloud.bigquery.dbapi as bqapi
import pandas as pd
from bertopic import BERTopic
from bertopic.vectorizers import ClassTfidfTransformer
from hdbscan import HDBSCAN
from joblib import Parallel, delayed
from sentence_transformers import SentenceTransformer
from sklearn.feature_extraction.text import CountVectorizer
from tqdm import tqdm
from umap import UMAP
import config
from common import download_blob, upload_blob, BigquerySession
from summarize_topics import summarization_wrapper, create_topic_summarizer
def get_topicless_articles(client, model):
query = "SELECT CA.id, CA.title, CA.coref, CA.published " \
"FROM Articles.CleanedArticles AS CA " \
"LEFT JOIN Articles.ArticleTopic TA ON CA.id = TA.article_id AND TA.model = %s " \
"WHERE TA.article_id IS NULL AND " \
f"CA.published >= TIMESTAMP_SUB(CURRENT_TIMESTAMP(), INTERVAL {config.TOPIC_FIT_RANGE_DAY} DAY)"
results = []
with closing(bqapi.Connection(client=client)) as connection:
with closing(connection.cursor()) as cursor:
cursor.execute(query, (model,))
for r in cursor.fetchall():
results.append(list(r))
return pd.DataFrame(results,
columns=["id", "title", "body", "published"])
def get_fitting_articles(client):
query = "SELECT id, title, coref, published " \
"FROM Articles.CleanedArticles AS CA WHERE CA.published >= " \
f"TIMESTAMP_SUB(CURRENT_TIMESTAMP(), INTERVAL {config.TOPIC_FIT_RANGE_DAY} DAY)"
results = []
with closing(bqapi.Connection(client=client)) as connection:
with closing(connection.cursor()) as cursor:
cursor.execute(query)
for r in cursor.fetchall():
results.append(list(r))
return pd.DataFrame(results,
columns=["id", "title", "body", "published"])
def identify_topics(df: pd.DataFrame):
embedding_model = SentenceTransformer(config.TOPIC_EMBEDDING)
umap_model = UMAP(n_neighbors=config.TOPIC_UMAP_NEIGHBORS, n_components=config.TOPIC_UMAP_COMPONENTS,
min_dist=config.TOPIC_UMAP_MIN_DIST, metric=config.TOPIC_UMAP_METRIC)
hdbscan_model = HDBSCAN(min_cluster_size=config.TOPIC_HDBSCAN_MIN_SIZE,
metric=config.TOPIC_HDBSCAN_METRIC,
cluster_selection_method='eom',
prediction_data=True)
vectorizer_model = CountVectorizer(stop_words="english")
ctfidf_model = ClassTfidfTransformer()
# All steps together
topic_model = BERTopic(
embedding_model=embedding_model,
umap_model=umap_model,
hdbscan_model=hdbscan_model,
vectorizer_model=vectorizer_model,
ctfidf_model=ctfidf_model,
)
topics, prob = topic_model.fit_transform(df.body)
df["topic"] = topics
df["topic_prob"] = prob
return df, topic_model
def summarize_topics(df: pd.DataFrame, jobs=7):
topics = list(set(df.topic.unique()))
summarizer = create_topic_summarizer("pass")
parallel = Parallel(n_jobs=jobs, backend="threading", return_as="generator")
topic_sum = pd.DataFrame({
"topic": topics,
"summary": ["" for _ in topics]
})
with tqdm(total=len(topics)) as progress:
for i, (w, summary) in enumerate(
parallel(delayed(summarization_wrapper)(summarizer, work, df) for work in topics)):
topic_sum.loc[topic_sum.topic == w, "summary"] = summary
progress.update(1)
return topic_sum
def upload_topic(model: BERTopic):
path_name = str(uuid.uuid4())
model.save(path="./temp_model", serialization="safetensors", save_embedding_model=False, save_ctfidf=True)
shutil.make_archive(path_name, "zip", "./temp_model")
upload_blob(config.TOPIC_BUCKET, path_name + ".zip", destination_blob_name=path_name + ".zip", generation=None)
shutil.rmtree("./temp_model")
os.remove(path_name + ".zip")
return f"gs://{config.TOPIC_BUCKET}/{path_name}.zip"
def write_topics(client: bq.Client, model, sum_df):
path = upload_topic(model)
current_time = datetime.datetime.now()
model_stmt = """INSERT INTO Articles.TopicModel SELECT GENERATE_UUID(), ?, ?, False;"""
ins_stmt = """INSERT INTO Articles.TopicSummary SELECT ?, id, ? FROM Articles.TopicModel WHERE fit_date = ?"""
client = bq.Client(project=client.project)
print("Writing model at: " + str(path))
with BigquerySession(client) as session:
session.begin_transaction()
job = client.query(model_stmt, bq.QueryJobConfig(
create_session=False,
query_parameters=[
bq.ScalarQueryParameter(None, "TIMESTAMP", current_time),
bq.ScalarQueryParameter(None, "STRING", path)
],
connection_properties=[
bq.query.ConnectionProperty(
key="session_id", value=session.session_id
)
],
), location=session.location)
job.result()
for _, row in sum_df.iterrows():
job = client.query(ins_stmt, bq.QueryJobConfig(
create_session=False,
query_parameters=[
bq.ScalarQueryParameter(None, "INTEGER", row["topic"]),
bq.ScalarQueryParameter(None, "STRING", row["summary"]),
bq.ScalarQueryParameter(None, "TIMESTAMP", current_time)
],
connection_properties=[
bq.query.ConnectionProperty(
key="session_id", value=session.session_id
)
],
), location=session.location)
job.result()
session.commit()
def load_topic_model(client):
query = "SELECT id, fit_date, path FROM Articles.TopicModel AS TM " \
f"WHERE TM.fit_date >= TIMESTAMP_SUB(CURRENT_TIMESTAMP(), INTERVAL {config.TOPIC_TTL_DAY} DAY) " \
f"ORDER BY fit_date DESC LIMIT 1"
with closing(bqapi.Connection(client=client)) as connection:
with closing(connection.cursor()) as cursor:
cursor.execute(query)
result = cursor.fetchone()
if result is None:
raise FileNotFoundError("Model Unavailable")
path = result[2]
id = result[0]
download_blob(path, "./temp_topic.zip")
shutil.unpack_archive("./temp_topic.zip", "./temp_topic")
model = BERTopic.load("./temp_topic", embedding_model=SentenceTransformer(config.TOPIC_EMBEDDING))
shutil.rmtree('./temp_topic')
os.remove("./temp_topic.zip")
return id, model
def get_topic_model(client):
try:
return load_topic_model(client)
except FileNotFoundError:
articles = get_fitting_articles(client)
articles, model = identify_topics(articles)
sum_df = summarize_topics(articles)
write_topics(client, model, sum_df)
return load_topic_model(client)
def batch_insert_topic(project, batch):
query = """INSERT INTO Articles.ArticleTopic VALUES(?, ?, ?, ?)"""
while True:
try:
with closing(bq.Client(project=project)) as client:
with BigquerySession(client) as session:
session.begin_transaction()
for id, model_id, topic, topic_prob in batch:
job = client.query(query, bq.QueryJobConfig(
create_session=False,
query_parameters=[
bq.ScalarQueryParameter(None, "STRING", id),
bq.ScalarQueryParameter(None, "STRING", model_id),
bq.ScalarQueryParameter(None, "INTEGER", topic),
bq.ScalarQueryParameter(None, "FLOAT", topic_prob)
],
connection_properties=[
bq.query.ConnectionProperty(
key="session_id", value=session.session_id
)
],
), location=session.location)
job.result()
session.commit()
return
except Exception as e:
print(e)
def categorize_articles(client):
id, model = get_topic_model(client)
articles = get_topicless_articles(client, id)
if articles.shape[0] == 0:
return
topics, prob = model.transform(articles.body)
articles["topic"] = topics
articles["topic_prob"] = prob
return id, articles
def write_article_topics(articles, model_id, batch_size=16, jobs=8):
params = [(row["id"], model_id, row["topic"], row["topic_prob"]) for _, row in articles.iterrows()]
batches = []
for i in range(0, len(params), batch_size):
batches.append(params[i:i + batch_size])
parallel = Parallel(n_jobs=jobs, return_as="generator")
with tqdm(total=len(batches)) as progress:
for _ in parallel(delayed(batch_insert_topic)(client.project, b) for b in batches):
progress.update(1)
def write_article_topics_manual(articles, model_id):
result = articles[["id", "topic", "topic_prob"]].copy()
result["model"] = model_id
result.to_parquet("article_topics.parquet", index=False)
return result
if __name__ == "__main__":
with closing(bq.Client(project=config.GCP_PROJECT)) as client:
mid, articles = categorize_articles(client)
gc.collect()
# write_article_topics(articles, mid)
write_article_topics_manual(articles, mid)