File size: 10,497 Bytes
bfcc872
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cf9b3dc
bfcc872
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cf9b3dc
bfcc872
 
 
 
 
 
 
 
 
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
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
"""
query_logger.py — Logs user queries and AI responses for RLHF analysis

Automatically logs every interaction to CSV and JSONL formats.
Export to RLHF training format with export_logs.py script.
"""
from __future__ import annotations

import csv
import json
import logging
from datetime import datetime
from pathlib import Path
from typing import Any, Dict, List, Optional

logger = logging.getLogger(__name__)


class QueryLogger:
    """Logs user queries and AI responses to JSON and CSV for RLHF analysis."""

    def __init__(self, log_dir: str = "logs"):
        """
        Initialize query logger.

        Args:
            log_dir: Directory to store log files (default: "logs/")
        """
        self.log_dir = Path(log_dir)
        self.log_dir.mkdir(parents=True, exist_ok=True)

        # Get today's date for log file naming
        today = datetime.now().strftime("%Y%m%d")
        self.jsonl_path = self.log_dir / f"queries_{today}.jsonl"
        self.csv_path = self.log_dir / f"queries_{today}.csv"

        # Initialize CSV file with headers if it doesn't exist
        if not self.csv_path.exists():
            self._init_csv()

    def _init_csv(self):
        """Initialize CSV file with headers."""
        headers = [
            'timestamp',
            'user_query',
            'ai_response',
            'tools_called',
            'response_time_ms',
            'success',
            'rating',
            'feedback',
            'model',
            'tokens_used'
        ]
        with open(self.csv_path, 'w', newline='', encoding='utf-8') as f:
            writer = csv.DictWriter(f, fieldnames=headers)
            writer.writeheader()

    def log_interaction(
        self,
        user_query: str,
        ai_response: str,
        *,
        tools_called: Optional[List[str]] = None,
        response_time_ms: Optional[int] = None,
        success: bool = True,
        rating: Optional[int] = None,
        feedback: Optional[str] = None,
        model: Optional[str] = None,
        tokens_used: Optional[int] = None,
        metadata: Optional[Dict[str, Any]] = None
    ):
        """
        Log a user query and AI response.

        Args:
            user_query: The user's input query
            ai_response: The AI's response
            tools_called: List of tool names that were called
            response_time_ms: Response time in milliseconds
            success: Whether the interaction was successful
            rating: Optional 1-5 rating (for RLHF)
            feedback: Optional text feedback (for RLHF)
            model: Model name used (e.g., "gpt-4-mini")
            tokens_used: Number of tokens consumed
            metadata: Additional metadata to log
        """
        timestamp = datetime.utcnow().isoformat()

        # Prepare log entry
        log_entry = {
            'timestamp': timestamp,
            'user_query': user_query,
            'ai_response': ai_response,
            'tools_called': tools_called or [],
            'response_time_ms': response_time_ms,
            'success': success,
            'rating': rating,
            'feedback': feedback,
            'model': model,
            'tokens_used': tokens_used,
            'metadata': metadata or {}
        }

        # Write to JSONL (one JSON object per line)
        try:
            with open(self.jsonl_path, 'a', encoding='utf-8') as f:
                f.write(json.dumps(log_entry) + '\n')
        except Exception as e:
            logger.warning(f"Failed to write to JSONL log: {e}")

        # Write to CSV (Excel-compatible)
        try:
            with open(self.csv_path, 'a', newline='', encoding='utf-8') as f:
                writer = csv.DictWriter(f, fieldnames=[
                    'timestamp', 'user_query', 'ai_response', 'tools_called',
                    'response_time_ms', 'success', 'rating', 'feedback',
                    'model', 'tokens_used'
                ])
                writer.writerow({
                    'timestamp': timestamp,
                    'user_query': user_query,
                    'ai_response': ai_response,
                    'tools_called': ','.join(tools_called or []),
                    'response_time_ms': response_time_ms,
                    'success': success,
                    'rating': rating or '',
                    'feedback': feedback or '',
                    'model': model or '',
                    'tokens_used': tokens_used or ''
                })
        except Exception as e:
            logger.warning(f"Failed to write to CSV log: {e}")

        logger.debug(f"Logged interaction: {user_query[:50]}...")

    def get_stats(self) -> Dict[str, Any]:
        """
        Get statistics about logged queries.

        Returns:
            Dictionary with statistics (total queries, success rate, avg response time, etc.)
        """
        try:
            with open(self.csv_path, 'r', encoding='utf-8') as f:
                reader = csv.DictReader(f)
                rows = list(reader)

            if not rows:
                return {
                    'total_queries': 0,
                    'successful_queries': 0,
                    'failed_queries': 0,
                    'avg_response_time_ms': 0,
                    'avg_rating': 0,
                    'rated_queries': 0,
                    'tool_usage': {}
                }

            total = len(rows)
            successful = sum(1 for r in rows if r.get('success') == 'True')
            failed = total - successful

            # Calculate average response time
            response_times = [
                int(r['response_time_ms'])
                for r in rows
                if r.get('response_time_ms') and r['response_time_ms'].isdigit()
            ]
            avg_response_time = sum(response_times) / len(response_times) if response_times else 0

            # Calculate average rating
            ratings = [
                int(r['rating'])
                for r in rows
                if r.get('rating') and r['rating'].isdigit()
            ]
            avg_rating = sum(ratings) / len(ratings) if ratings else 0
            rated_queries = len(ratings)

            # Count tool usage
            tool_usage = {}
            for row in rows:
                tools = row.get('tools_called', '').split(',')
                for tool in tools:
                    tool = tool.strip()
                    if tool:
                        tool_usage[tool] = tool_usage.get(tool, 0) + 1

            return {
                'total_queries': total,
                'successful_queries': successful,
                'failed_queries': failed,
                'avg_response_time_ms': avg_response_time,
                'avg_rating': avg_rating,
                'rated_queries': rated_queries,
                'tool_usage': tool_usage
            }
        except Exception as e:
            logger.warning(f"Failed to calculate stats: {e}")
            return {
                'total_queries': 0,
                'successful_queries': 0,
                'failed_queries': 0,
                'avg_response_time_ms': 0,
                'avg_rating': 0,
                'rated_queries': 0,
                'tool_usage': {}
            }

    def export_for_rlhf(self, output_path: Optional[Path] = None) -> Path:
        """
        Export logs in RLHF training format.

        Format:
        [
            {
                "prompt": "user query",
                "completion": "ai response",
                "rating": 5,
                "feedback": "Great!",
                "tools_used": ["list_grants"],
                "timestamp": "2025-10-23T10:15:30"
            },
            ...
        ]

        Args:
            output_path: Optional custom output path

        Returns:
            Path to exported file
        """
        if output_path is None:
            today = datetime.now().strftime("%Y%m%d")
            output_path = self.log_dir / f"rlhf_data_{today}.json"

        rlhf_data = []

        try:
            # Read from JSONL
            with open(self.jsonl_path, 'r', encoding='utf-8') as f:
                for line in f:
                    entry = json.loads(line)

                    # Only include successful interactions
                    if not entry.get('success', False):
                        continue

                    rlhf_entry = {
                        'prompt': entry['user_query'],
                        'completion': entry['ai_response'],
                        'rating': entry.get('rating'),
                        'feedback': entry.get('feedback'),
                        'tools_used': entry.get('tools_called', []),
                        'timestamp': entry['timestamp'],
                        'response_time_ms': entry.get('response_time_ms'),
                        'model': entry.get('model')
                    }

                    rlhf_data.append(rlhf_entry)

            # Write RLHF format
            with open(output_path, 'w', encoding='utf-8') as f:
                json.dump(rlhf_data, f, indent=2, ensure_ascii=False)

            logger.info(f"Exported {len(rlhf_data)} interactions to {output_path}")
            return output_path

        except Exception as e:
            logger.error(f"Failed to export RLHF data: {e}")
            raise


# Global singleton instance
_query_logger: Optional[QueryLogger] = None


def get_query_logger(log_dir: str = "logs") -> QueryLogger:
    """
    Get or create the global query logger instance.

    Args:
        log_dir: Directory to store log files

    Returns:
        QueryLogger instance
    """
    global _query_logger
    if _query_logger is None:
        _query_logger = QueryLogger(log_dir=log_dir)
    return _query_logger


# Quick self-test
if __name__ == "__main__":
    logger = get_query_logger(log_dir="_out/test_logs")

    # Log a test interaction
    logger.log_interaction(
        user_query="What grants are available for batteries?",
        ai_response="Here are the battery-related grants: 1. Battery Innovation Grant...",
        tools_called=["list_grants"],
        response_time_ms=1500,
        success=True,
        rating=5,
        feedback="Very helpful!",
        model="gpt-4-mini"
    )

    # Get stats
    stats = logger.get_stats()
    print("Statistics:", json.dumps(stats, indent=2))

    # Export for RLHF
    output = logger.export_for_rlhf()
    print(f"Exported to: {output}")