File size: 4,627 Bytes
778d47d | 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 | """
Utilities for loading BIRD dataset column/value descriptions from
database_description/*.csv files (one CSV per table).
Adapted from CHESS (https://github.com/ShayanTalaei/CHESS)
chess/src/database_utils/db_catalog/csv_utils.py.
Schema returned:
{
table_name_lower: {
column_name_lower: {
"original_column_name": str,
"column_name": str, # expanded/human-readable name
"column_description": str,
"data_format": str,
"value_description": str,
}
}
}
"""
import logging
from pathlib import Path
from typing import Dict
import pandas as pd
def load_db_descriptions(
db_directory_path: str,
use_value_description: bool = True,
) -> Dict[str, Dict[str, Dict[str, str]]]:
"""Load table/column descriptions from BIRD database_description/*.csv.
Args:
db_directory_path: Path to the database directory (containing
database_description/ sub-folder).
use_value_description: Whether to include value_description field.
Returns:
Nested dict table → column → field → value.
Returns {} when the description folder does not exist.
"""
encoding_types = ["utf-8-sig", "cp1252", "latin-1"]
description_path = Path(db_directory_path) / "database_description"
if not description_path.exists():
return {}
table_description: Dict[str, Dict[str, Dict[str, str]]] = {}
for csv_file in sorted(description_path.glob("*.csv")):
table_name = csv_file.stem.lower().strip()
table_description[table_name] = {}
loaded = False
for encoding in encoding_types:
try:
df = pd.read_csv(csv_file, index_col=False, encoding=encoding)
for _, row in df.iterrows():
col_key = str(row.get("original_column_name", "")).lower().strip()
if not col_key:
continue
def _clean(val, remove_prefix: str = "") -> str:
if not pd.notna(val):
return ""
s = str(val).replace("\n", " ").strip()
if remove_prefix and s.lower().startswith(remove_prefix):
s = s[len(remove_prefix):].strip()
return s
col_description = _clean(
row.get("column_description", ""),
remove_prefix="commonsense evidence:",
)
value_desc = ""
if use_value_description:
value_desc = _clean(
row.get("value_description", ""),
remove_prefix="commonsense evidence:",
)
if value_desc.lower().startswith("not useful"):
value_desc = value_desc[10:].strip()
table_description[table_name][col_key] = {
"original_column_name": str(row.get("original_column_name", col_key)),
"column_name": _clean(row.get("column_name", "")),
"column_description": col_description,
"data_format": _clean(row.get("data_format", "")),
"value_description": value_desc,
}
logging.debug("Loaded descriptions from %s (%s)", csv_file, encoding)
loaded = True
break
except Exception:
continue
if not loaded:
logging.warning("Could not read descriptions from %s", csv_file)
return table_description
def load_all_db_descriptions(
bird_dataset_path: str,
use_value_description: bool = True,
) -> Dict[str, Dict[str, Dict[str, Dict[str, str]]]]:
"""Load descriptions for every database under a BIRD split directory.
Args:
bird_dataset_path: Path to a BIRD split, e.g.
/data/bird/train/train_databases
Each sub-directory is one db_id.
use_value_description: Whether to include value_description.
Returns:
{db_id: load_db_descriptions(db_dir)}
"""
root = Path(bird_dataset_path)
all_descriptions: Dict[str, Dict] = {}
for db_dir in sorted(root.iterdir()):
if db_dir.is_dir():
db_id = db_dir.name
all_descriptions[db_id] = load_db_descriptions(
str(db_dir), use_value_description=use_value_description
)
return all_descriptions
|