-
Notifications
You must be signed in to change notification settings - Fork 0
/
measure_small.py
152 lines (135 loc) · 5.31 KB
/
measure_small.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
import argparse
import json
from pathlib import Path
import numpy as np
import pandas as pd
from qdrant_client import models
from loguru import logger
from tqdm import tqdm
from db_utils import init_qdrant_client, init_sql_session
from model_utils import init_model
from constants import PREFIXES
def parse_arguments() -> argparse.Namespace:
parser = argparse.ArgumentParser()
parser.add_argument("model_name", type=str, default="word2vec")
parser.add_argument("--test_file", type=str, default="defs.tsv")
parser.add_argument("--prefix", choices=["query", "passage"], default=None)
parser.add_argument(
"--collection_name", type=str, default="sonaveeb-semantic-search"
)
parser.add_argument("--target_col", type=str, default="word_ee")
parser.add_argument("--target_id_col", type=str, default="word_ee_id")
parser.add_argument("--syn_col", type=str, default="synonym_ids")
parser.add_argument("--eng_col", type=str, default="def_en")
parser.add_argument("--est_col", type=str, default="def_ee")
parser.add_argument("--rus_col", type=str, default="def_ru")
parser.add_argument("--word2vec_baseline", action="store_true")
group = parser.add_mutually_exclusive_group(required=True)
group.add_argument(
"--est", action="store_true", help="Use definitions in Estonian for search"
)
group.add_argument(
"--eng", action="store_true", help="Use definitions in English for search"
)
group.add_argument(
"--rus", action="store_true", help="Use definitions in Russian for search"
)
return parser.parse_args()
if __name__ == "__main__":
args = parse_arguments()
collection_name = args.collection_name
target_col = args.target_col
target_id_col = args.target_id_col
syn_col = args.syn_col
data = pd.read_csv(args.test_file, sep="\t")
session_maker = init_sql_session()
_, model = init_model(args.model_name)
if args.est:
logger.info("Using definitions in Estonian")
definition_lang = "est"
definitions = data[args.est_col].tolist()
elif args.eng:
logger.info("Using definitions in English")
definition_lang = "eng"
definitions = data[args.eng_col].tolist()
elif args.rus:
logger.info("Using definitions in Russian")
definition_lang = "rus"
definitions = data[args.rus_col].tolist()
else:
raise ValueError("No language option selected")
if args.prefix:
logger.info(f"Prepending definitions with prefix `{args.prefix}`")
definitions = [
f"{PREFIXES[args.prefix]}{definition}" for definition in definitions
]
logger.info("Encoding definitions")
vectors = model.encode(definitions).tolist()
target_words = data[target_col].tolist()
target_ids = data[target_id_col].tolist()
synonym_ids = [
[int(i) for i in el.split(",")] if isinstance(el, str) else None
for el in data[syn_col].tolist()
]
if not args.word2vec_baseline:
qd_client = init_qdrant_client()
search_queries = [
models.SearchRequest(
vector=vector,
limit=100,
with_payload=True,
offset=0, # not skipping the first item because the query definitions are different from what we have
)
for vector in vectors
]
logger.info("Performing vector search")
matches = qd_client.search_batch(
collection_name=collection_name,
requests=search_queries,
)
matched_ids: list[list[int]] = [
[point.payload["word_id"] for point in match] for match in matches
]
elif args.model_name == "word2vec":
logger.info("Performing baseline word2vec search")
matched_ids = []
for vector in tqdm(vectors):
matched_ids.append(model.search_by_vector(np.array(vector)))
else:
raise ValueError("word2vec baseline is only usable with word2vec model")
assert len(matched_ids) == len(target_words)
out_prefix = f"{args.model_name}-{definition_lang}"
if args.prefix:
query_prefix_name = f"-{args.prefix}"
else:
query_prefix_name = ""
if "passage" in args.collection_name:
collection_postfix_name = "-passage"
elif "query" in args.collection_name:
collection_postfix_name = "-query"
else:
collection_postfix_name = ""
if args.word2vec_baseline:
baseline_str = "-baseline"
else:
baseline_str = ""
raw_output_dir = "./raw_output/small"
raw_output_path = Path(raw_output_dir).resolve()
raw_output_path.mkdir(exist_ok=True)
out_suffix = "-preds.jsonl"
out_file_name = (
raw_output_path
/ f"{out_prefix}{query_prefix_name}{collection_postfix_name}{baseline_str}{out_suffix}"
)
logger.info(f"Writing out predictions to `{out_file_name}`")
with open(out_file_name, "w") as file:
for target_word, target_id, candidates, cur_syn_ids in zip(
target_words, target_ids, matched_ids, synonym_ids
):
cur_data = dict(
target_word=target_word,
target_id=target_id,
matched_word_ids=candidates,
synonym_ids=cur_syn_ids,
)
file.write(json.dumps(cur_data, ensure_ascii=False) + "\n")