File size: 9,409 Bytes
13ee483
 
 
 
 
 
 
 
b759b87
 
 
 
13ee483
b759b87
13ee483
 
b759b87
abb320a
13ee483
 
 
 
 
 
 
 
b759b87
 
13ee483
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b759b87
13ee483
 
 
 
 
b759b87
 
 
 
13ee483
 
 
 
 
 
 
 
 
 
036a85e
13ee483
036a85e
13ee483
036a85e
13ee483
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b759b87
 
13ee483
 
b759b87
13ee483
b759b87
 
13ee483
 
 
 
 
 
 
 
 
b759b87
13ee483
b759b87
13ee483
b759b87
13ee483
 
b759b87
036a85e
b759b87
13ee483
b759b87
13ee483
 
 
b759b87
 
13ee483
 
 
 
 
 
 
 
 
 
b759b87
13ee483
 
b759b87
13ee483
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
250
251
252
253
254
255
# Attribution: Original code by Ruoxin Wang
# Repository: <your-repo-url>

"""
Module: refactored_chatbot
This module provides utilities for loading database schemas, extracting DDL,
indexing content, and a ChatBot class to generate SQL queries from natural language.
"""
import os
import json
import re
import sqlite3
import copy
from tqdm import tqdm

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from whoosh import index

from utils.db_utils import (
    get_db_schema,
    check_sql_executability,
    get_matched_contents,
    get_db_schema_sequence,
    get_matched_content_sequence
)
from schema_item_filter import SchemaItemClassifierInference, filter_schema


class DatabaseUtils:
    """
    Utilities for loading database comments, schemas, and DDL statements.
    """

    @staticmethod
    def _remove_similar_comments(names, comments):
        """
        Remove comments identical to table/column names (ignoring underscores and spaces).
        """
        filtered = []
        for name, comment in zip(names, comments):
            normalized_name = name.replace("_", "").replace(" ", "").lower()
            normalized_comment = comment.replace("_", "").replace(" ", "").lower()
            filtered.append("") if normalized_name == normalized_comment else filtered.append(comment)
        return filtered

    @staticmethod
    def load_db_comments(table_json_path):
        """
        Load additional comments for tables and columns from a JSON file.

        Args:
            table_json_path (str): Path to JSON file containing table and column comments.

        Returns:
            dict: Mapping from database ID to comments structure.
        """
        additional_info = json.load(open(table_json_path))
        db_comments = {}

        for db_info in additional_info:
            db_id = db_info["db_id"]
            comment_dict = {}

            # Process column comments
            original_cols = db_info["column_names_original"]
            col_names = [col.lower() for _, col in original_cols]
            col_comments = [c.lower() for _, c in db_info["column_names"]]
            col_comments = DatabaseUtils._remove_similar_comments(col_names, col_comments)
            col_table_idxs = [t_idx for t_idx, _ in original_cols]

            # Process table comments
            original_tables = db_info["table_names_original"]
            tbl_names = [tbl.lower() for tbl in original_tables]
            tbl_comments = [c.lower() for c in db_info["table_names"]]
            tbl_comments = DatabaseUtils._remove_similar_comments(tbl_names, tbl_comments)

            for idx, name in enumerate(tbl_names):
                comment_dict[name] = {
                    "table_comment": tbl_comments[idx],
                    "column_comments": {}
                }
                # Associate columns
                for t_idx, col_name, col_comment in zip(col_table_idxs, col_names, col_comments):
                    if t_idx == idx:
                        comment_dict[name]["column_comments"][col_name] = col_comment

            db_comments[db_id] = comment_dict

        return db_comments

    @staticmethod
    def get_db_schemas(db_path, tables_json):
        """
        Build a mapping from database ID to its schema representation.

        Args:
            db_path (str): Directory containing database subdirectories.
            tables_json (str): Path to JSON with table comments.

        Returns:
            dict: Mapping from db_id to schema object.
        """
        comments = DatabaseUtils.load_db_comments(tables_json)
        schemas = {}
        for db_id in tqdm(os.listdir(db_path), desc="Loading schemas"):
            sqlite_path = os.path.join(db_path, db_id, f"{db_id}.sqlite")
            schemas[db_id] = get_db_schema(sqlite_path, comments, db_id)
        return schemas

    @staticmethod
    def get_db_ddls(db_path):
        """
        Extract formatted DDL statements for all tables in each database.

        Args:
            db_path (str): Directory containing database subdirectories.

        Returns:
            dict: Mapping from db_id to its DDL string.
        """
        ddls = {}
        for db_id in os.listdir(db_path):
            conn = sqlite3.connect(os.path.join(db_path, db_id, f"{db_id}.sqlite"))
            cursor = conn.cursor()
            cursor.execute("SELECT name, sql FROM sqlite_master WHERE type='table';")
            ddl_statements = []

            for name, raw_sql in cursor.fetchall():
                sql = raw_sql or ""
                sql = re.sub(r'--.*', '', sql).replace("\t", " ")
                sql = re.sub(r" +", " ", sql)
                formatted = sqlparse.format(
                    sql,
                    keyword_case="upper",
                    identifier_case="lower",
                    reindent_aligned=True
                )
                # Adjust spacing for readability
                formatted = formatted.replace(", ", ",\n    ")
                if formatted.rstrip().endswith(";"):
                    formatted = formatted.rstrip()[:-1] + "\n);"
                formatted = re.sub(r"(CREATE TABLE.*?)\(", r"\1(\n    ", formatted)
                ddl_statements.append(formatted)

            ddls[db_id] = "\n\n".join(ddl_statements)
        return ddls


class ChatBot:
    """
    ChatBot for generating and executing SQL queries using a causal language model.
    """

    def __init__(self, model_name: str = "seeklhy/codes-1b", device: str = "cuda:0") -> None:
        """
        Initialize the ChatBot with model and tokenizer.

        Args:
            model_name (str): HuggingFace model identifier.
            device (str): CUDA device string or 'cpu'.
        """
        os.environ["CUDA_VISIBLE_DEVICES"] = device
        self.tokenizer = AutoTokenizer.from_pretrained(model_name)
        self.model = AutoModelForCausalLM.from_pretrained(
            model_name,
            device_map="auto",
            torch_dtype=torch.float16
        )
        self.max_length = 4096
        self.max_new_tokens = 256
        self.max_prefix_length = self.max_length - self.max_new_tokens

        # Schema item classifier
        self.schema_classifier = SchemaItemClassifierInference("Roxanne-WANG/LangSQL")

        # Initialize content searchers
        self.content_searchers = {}
        index_dir = "db_contents_index"
        for db_id in os.listdir(index_dir):
            path = os.path.join(index_dir, db_id)
            if index.exists_in(path):
                self.content_searchers[db_id] = index.open_dir(path).searcher()
            else:
                raise FileNotFoundError(f"Whoosh index not found for '{db_id}' at '{path}'")

        # Load schemas and DDLs
        self.db_ids = sorted(os.listdir("databases"))
        self.schemas = DatabaseUtils.get_db_schemas("databases", "data/tables.json")
        self.ddls = DatabaseUtils.get_db_ddls("databases")

    def get_response(self, question: str, db_id: str) -> str:
        """
        Generate an executable SQL query for a natural language question.

        Args:
            question (str): User question in natural language.
            db_id (str): Identifier of the target database.

        Returns:
            str: Executable SQL query or an error message.
        """
        # Prepare data
        schema = copy.deepcopy(self.schemas[db_id])
        contents = get_matched_contents(question, self.content_searchers[db_id])
        data = {
            "text": question,
            "schema": schema,
            "matched_contents": contents
        }
        data = filter_schema(data, self.schema_classifier, top_k=6, top_m=10)
        data["schema_sequence"] = get_db_schema_sequence(data["schema"])
        data["content_sequence"] = get_matched_content_sequence(data["matched_contents"])

        prefix = (
            f"{data['schema_sequence']}\n"
            f"{data['content_sequence']}\n"
            f"{question}\n"
        )

        # Tokenize and ensure length limits
        input_ids = [self.tokenizer.bos_token_id] + self.tokenizer(prefix)["input_ids"]
        if len(input_ids) > self.max_prefix_length:
            input_ids = [self.tokenizer.bos_token_id] + input_ids[-(self.max_prefix_length - 1):]
        attention_mask = [1] * len(input_ids)

        inputs = {
            "input_ids": torch.tensor([input_ids], dtype=torch.int64).to(self.model.device),
            "attention_mask": torch.tensor([attention_mask], dtype=torch.int64).to(self.model.device)
        }

        with torch.no_grad():
            outputs = self.model.generate(
                **inputs,
                max_new_tokens=self.max_new_tokens,
                num_beams=4,
                num_return_sequences=4
            )

        # Decode and choose executable SQL
        decoded = self.tokenizer.batch_decode(
            outputs[:, inputs['input_ids'].shape[1]:],
            skip_special_tokens=True,
            clean_up_tokenization_spaces=False
        )
        final_sql = None
        for sql in decoded:
            if check_sql_executability(sql, os.path.join("databases", db_id, f"{db_id}.sqlite")) is None:
                final_sql = sql.strip()
                break
        if not final_sql:
            final_sql = decoded[0].strip() or "Sorry, I cannot generate a suitable SQL query."

        return final_sql