File size: 27,701 Bytes
23804b3 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 |
"""
Episodic Memory System for Experience Replay and Learning
Captures temporal sequences of agent experiences for learning
"""
import sqlite3
import json
import uuid
from datetime import datetime, timedelta
from typing import Dict, List, Any, Optional, Tuple
from dataclasses import dataclass, asdict
import logging
from pathlib import Path
import pickle
logger = logging.getLogger(__name__)
@dataclass
class Episode:
"""Individual episode with temporal sequence"""
id: str
agent_id: str
session_id: str
start_time: datetime
end_time: Optional[datetime]
episode_type: str # operation, training, evaluation, etc.
context: Dict[str, Any]
actions: List[Dict[str, Any]]
observations: List[Dict[str, Any]]
rewards: List[float]
outcome: Optional[str]
success: bool
metadata: Dict[str, Any]
@dataclass
class ExperienceReplay:
"""Experience replay record for learning"""
episode_id: str
replay_count: int
last_replayed: datetime
replay_effectiveness: float
learning_insights: List[str]
class EpisodicMemorySystem:
"""Advanced episodic memory with experience replay capabilities"""
def __init__(self, db_path: str = "data/cognitive/episodic_memory.db"):
"""Initialize episodic memory system"""
self.db_path = Path(db_path)
self.db_path.parent.mkdir(parents=True, exist_ok=True)
self._init_database()
self._active_episodes = {}
def _init_database(self):
"""Initialize database schemas"""
with sqlite3.connect(self.db_path) as conn:
conn.execute("""
CREATE TABLE IF NOT EXISTS episodes (
id TEXT PRIMARY KEY,
agent_id TEXT NOT NULL,
session_id TEXT NOT NULL,
start_time TEXT NOT NULL,
end_time TEXT,
episode_type TEXT NOT NULL,
context TEXT,
actions TEXT,
observations TEXT,
rewards TEXT,
outcome TEXT,
success BOOLEAN,
metadata TEXT,
created_at TEXT DEFAULT CURRENT_TIMESTAMP
)
""")
conn.execute("""
CREATE TABLE IF NOT EXISTS experience_replay (
id TEXT PRIMARY KEY,
episode_id TEXT,
replay_count INTEGER DEFAULT 0,
last_replayed TEXT,
replay_effectiveness REAL DEFAULT 0.0,
learning_insights TEXT,
created_at TEXT DEFAULT CURRENT_TIMESTAMP,
FOREIGN KEY (episode_id) REFERENCES episodes(id)
)
""")
conn.execute("""
CREATE TABLE IF NOT EXISTS episode_patterns (
id TEXT PRIMARY KEY,
pattern_type TEXT,
pattern_description TEXT,
episodes TEXT,
frequency INTEGER,
success_rate REAL,
discovered_at TEXT DEFAULT CURRENT_TIMESTAMP
)
""")
# Create indices
conn.execute("CREATE INDEX IF NOT EXISTS idx_agent_episodes ON episodes(agent_id)")
conn.execute("CREATE INDEX IF NOT EXISTS idx_episode_type ON episodes(episode_type)")
conn.execute("CREATE INDEX IF NOT EXISTS idx_episode_success ON episodes(success)")
conn.execute("CREATE INDEX IF NOT EXISTS idx_episode_start_time ON episodes(start_time)")
def start_episode(self, agent_id: str, session_id: str,
episode_type: str, context: Dict[str, Any] = None) -> str:
"""Start a new episode recording"""
try:
episode_id = str(uuid.uuid4())
episode = Episode(
id=episode_id,
agent_id=agent_id,
session_id=session_id,
start_time=datetime.now(),
end_time=None,
episode_type=episode_type,
context=context or {},
actions=[],
observations=[],
rewards=[],
outcome=None,
success=False,
metadata={}
)
self._active_episodes[episode_id] = episode
# Store initial episode data
with sqlite3.connect(self.db_path) as conn:
conn.execute("""
INSERT INTO episodes (
id, agent_id, session_id, start_time, episode_type,
context, actions, observations, rewards, success, metadata
) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
""", (
episode.id, episode.agent_id, episode.session_id,
episode.start_time.isoformat(), episode.episode_type,
json.dumps(episode.context), json.dumps(episode.actions),
json.dumps(episode.observations), json.dumps(episode.rewards),
episode.success, json.dumps(episode.metadata)
))
logger.info(f"Started episode {episode_id} for agent {agent_id}")
return episode_id
except Exception as e:
logger.error(f"Error starting episode: {e}")
return ""
def record_action(self, episode_id: str, action: Dict[str, Any]):
"""Record an action in the current episode"""
try:
if episode_id in self._active_episodes:
episode = self._active_episodes[episode_id]
action['timestamp'] = datetime.now().isoformat()
episode.actions.append(action)
logger.debug(f"Recorded action in episode {episode_id}: {action.get('type', 'unknown')}")
else:
logger.warning(f"Episode {episode_id} not active")
except Exception as e:
logger.error(f"Error recording action: {e}")
def record_observation(self, episode_id: str, observation: Dict[str, Any]):
"""Record an observation in the current episode"""
try:
if episode_id in self._active_episodes:
episode = self._active_episodes[episode_id]
observation['timestamp'] = datetime.now().isoformat()
episode.observations.append(observation)
logger.debug(f"Recorded observation in episode {episode_id}")
else:
logger.warning(f"Episode {episode_id} not active")
except Exception as e:
logger.error(f"Error recording observation: {e}")
def record_reward(self, episode_id: str, reward: float):
"""Record a reward signal in the current episode"""
try:
if episode_id in self._active_episodes:
episode = self._active_episodes[episode_id]
episode.rewards.append(reward)
logger.debug(f"Recorded reward in episode {episode_id}: {reward}")
else:
logger.warning(f"Episode {episode_id} not active")
except Exception as e:
logger.error(f"Error recording reward: {e}")
def end_episode(self, episode_id: str, success: bool = False,
outcome: str = "", metadata: Dict[str, Any] = None):
"""End an episode and store final results"""
try:
if episode_id in self._active_episodes:
episode = self._active_episodes[episode_id]
episode.end_time = datetime.now()
episode.success = success
episode.outcome = outcome
if metadata:
episode.metadata.update(metadata)
# Update database with final episode data
with sqlite3.connect(self.db_path) as conn:
conn.execute("""
UPDATE episodes SET
end_time = ?, actions = ?, observations = ?,
rewards = ?, outcome = ?, success = ?, metadata = ?
WHERE id = ?
""", (
episode.end_time.isoformat(),
json.dumps(episode.actions),
json.dumps(episode.observations),
json.dumps(episode.rewards),
episode.outcome,
episode.success,
json.dumps(episode.metadata),
episode_id
))
# Create experience replay record
self._create_replay_record(episode_id)
# Remove from active episodes
del self._active_episodes[episode_id]
logger.info(f"Ended episode {episode_id}: success={success}, outcome={outcome}")
else:
logger.warning(f"Episode {episode_id} not active")
except Exception as e:
logger.error(f"Error ending episode: {e}")
def get_episodes_for_replay(self, agent_id: str = "", episode_type: str = "",
success_only: bool = False, limit: int = 10) -> List[Episode]:
"""Get episodes suitable for experience replay"""
try:
with sqlite3.connect(self.db_path) as conn:
conditions = []
params = []
if agent_id:
conditions.append("agent_id = ?")
params.append(agent_id)
if episode_type:
conditions.append("episode_type = ?")
params.append(episode_type)
if success_only:
conditions.append("success = 1")
where_clause = " AND ".join(conditions) if conditions else "1=1"
cursor = conn.execute(f"""
SELECT * FROM episodes
WHERE {where_clause} AND end_time IS NOT NULL
ORDER BY start_time DESC
LIMIT ?
""", params + [limit])
episodes = []
for row in cursor.fetchall():
episode = Episode(
id=row[0],
agent_id=row[1],
session_id=row[2],
start_time=datetime.fromisoformat(row[3]),
end_time=datetime.fromisoformat(row[4]) if row[4] else None,
episode_type=row[5],
context=json.loads(row[6]) if row[6] else {},
actions=json.loads(row[7]) if row[7] else [],
observations=json.loads(row[8]) if row[8] else [],
rewards=json.loads(row[9]) if row[9] else [],
outcome=row[10],
success=bool(row[11]),
metadata=json.loads(row[12]) if row[12] else {}
)
episodes.append(episode)
logger.info(f"Retrieved {len(episodes)} episodes for replay")
return episodes
except Exception as e:
logger.error(f"Error getting episodes for replay: {e}")
return []
def replay_experience(self, episode_id: str) -> Dict[str, Any]:
"""Replay an episode and extract learning insights"""
try:
with sqlite3.connect(self.db_path) as conn:
cursor = conn.execute("SELECT * FROM episodes WHERE id = ?", (episode_id,))
row = cursor.fetchone()
if not row:
return {'error': 'Episode not found'}
episode = Episode(
id=row[0],
agent_id=row[1],
session_id=row[2],
start_time=datetime.fromisoformat(row[3]),
end_time=datetime.fromisoformat(row[4]) if row[4] else None,
episode_type=row[5],
context=json.loads(row[6]) if row[6] else {},
actions=json.loads(row[7]) if row[7] else [],
observations=json.loads(row[8]) if row[8] else [],
rewards=json.loads(row[9]) if row[9] else [],
outcome=row[10],
success=bool(row[11]),
metadata=json.loads(row[12]) if row[12] else {}
)
# Analyze episode for learning insights
insights = self._analyze_episode_for_insights(episode)
# Update replay statistics
self._update_replay_stats(episode_id, insights)
logger.info(f"Replayed episode {episode_id} with {len(insights)} insights")
return {
'episode': episode,
'insights': insights,
'replay_time': datetime.now().isoformat()
}
except Exception as e:
logger.error(f"Error replaying experience: {e}")
return {'error': str(e)}
def discover_patterns(self) -> Dict[str, Any]:
"""Discover patterns across episodes"""
try:
with sqlite3.connect(self.db_path) as conn:
# Get all completed episodes
cursor = conn.execute("""
SELECT * FROM episodes
WHERE end_time IS NOT NULL
ORDER BY start_time
""")
episodes = cursor.fetchall()
patterns = {
'action_patterns': self._discover_action_patterns(episodes),
'success_patterns': self._discover_success_patterns(episodes),
'temporal_patterns': self._discover_temporal_patterns(episodes),
'agent_patterns': self._discover_agent_patterns(episodes)
}
# Store discovered patterns
for pattern_type, pattern_list in patterns.items():
for pattern in pattern_list:
self._store_pattern(pattern_type, pattern)
logger.info(f"Discovered patterns: {sum(len(p) for p in patterns.values())} total")
return patterns
except Exception as e:
logger.error(f"Error discovering patterns: {e}")
return {'error': str(e)}
def _create_replay_record(self, episode_id: str):
"""Create experience replay record"""
try:
with sqlite3.connect(self.db_path) as conn:
conn.execute("""
INSERT INTO experience_replay (id, episode_id, last_replayed)
VALUES (?, ?, ?)
""", (str(uuid.uuid4()), episode_id, datetime.now().isoformat()))
except Exception as e:
logger.error(f"Error creating replay record: {e}")
def _analyze_episode_for_insights(self, episode: Episode) -> List[str]:
"""Analyze episode and extract learning insights"""
insights = []
try:
# Action sequence analysis
if len(episode.actions) > 1:
action_types = [a.get('type', 'unknown') for a in episode.actions]
unique_actions = len(set(action_types))
insights.append(f"Used {unique_actions} different action types in sequence")
# Reward trajectory analysis
if episode.rewards:
total_reward = sum(episode.rewards)
avg_reward = total_reward / len(episode.rewards)
insights.append(f"Average reward per step: {avg_reward:.3f}")
# Reward trend
if len(episode.rewards) > 2:
if episode.rewards[-1] > episode.rewards[0]:
insights.append("Improving performance throughout episode")
else:
insights.append("Declining performance throughout episode")
# Success factor analysis
if episode.success:
insights.append(f"Success achieved with {len(episode.actions)} actions")
if episode.outcome:
insights.append(f"Success outcome: {episode.outcome}")
else:
insights.append(f"Failed after {len(episode.actions)} actions")
if episode.outcome:
insights.append(f"Failure reason: {episode.outcome}")
# Context relevance
if episode.context:
context_keys = list(episode.context.keys())
insights.append(f"Context factors: {', '.join(context_keys[:3])}")
except Exception as e:
logger.error(f"Error analyzing episode insights: {e}")
insights.append(f"Analysis error: {str(e)}")
return insights
def _update_replay_stats(self, episode_id: str, insights: List[str]):
"""Update replay statistics"""
try:
with sqlite3.connect(self.db_path) as conn:
# Calculate effectiveness based on insights
effectiveness = min(len(insights) / 10.0, 1.0) # Scale to 0-1
conn.execute("""
UPDATE experience_replay SET
replay_count = replay_count + 1,
last_replayed = ?,
replay_effectiveness = ?,
learning_insights = ?
WHERE episode_id = ?
""", (
datetime.now().isoformat(),
effectiveness,
json.dumps(insights),
episode_id
))
except Exception as e:
logger.error(f"Error updating replay stats: {e}")
def _discover_action_patterns(self, episodes: List[Tuple]) -> List[Dict[str, Any]]:
"""Discover common action patterns"""
patterns = []
action_sequences = {}
for episode in episodes:
if episode[7]: # actions column
actions = json.loads(episode[7])
action_types = [a.get('type', 'unknown') for a in actions]
# Look for sequences of length 2-4
for seq_len in range(2, min(5, len(action_types) + 1)):
for i in range(len(action_types) - seq_len + 1):
sequence = tuple(action_types[i:i + seq_len])
if sequence not in action_sequences:
action_sequences[sequence] = {'count': 0, 'success_count': 0}
action_sequences[sequence]['count'] += 1
if episode[11]: # success column
action_sequences[sequence]['success_count'] += 1
# Convert to patterns
for sequence, stats in action_sequences.items():
if stats['count'] >= 3: # Minimum frequency
success_rate = stats['success_count'] / stats['count']
patterns.append({
'pattern': ' -> '.join(sequence),
'frequency': stats['count'],
'success_rate': success_rate
})
return sorted(patterns, key=lambda x: x['frequency'], reverse=True)
def _discover_success_patterns(self, episodes: List[Tuple]) -> List[Dict[str, Any]]:
"""Discover patterns that lead to success"""
patterns = []
success_factors = {}
for episode in episodes:
# Analyze context factors for all episodes
if episode[6]: # context column
context = json.loads(episode[6])
for key, value in context.items():
factor_key = f"{key}={value}"
if factor_key not in success_factors:
success_factors[factor_key] = {'success': 0, 'total': 0}
success_factors[factor_key]['total'] += 1
if episode[11]: # success column
success_factors[factor_key]['success'] += 1
# Convert to patterns
for factor, stats in success_factors.items():
if stats['total'] >= 3: # Minimum frequency
success_rate = stats['success'] / stats['total'] if stats['total'] > 0 else 0
if success_rate > 0.7: # High success rate threshold
patterns.append({
'pattern': f"Context factor: {factor}",
'frequency': stats['total'],
'success_rate': success_rate
})
return sorted(patterns, key=lambda x: x['success_rate'], reverse=True)
def _discover_temporal_patterns(self, episodes: List[Tuple]) -> List[Dict[str, Any]]:
"""Discover temporal patterns in episodes"""
patterns = []
# Group episodes by hour of day
hour_stats = {}
for episode in episodes:
start_time = datetime.fromisoformat(episode[3])
hour = start_time.hour
if hour not in hour_stats:
hour_stats[hour] = {'total': 0, 'success': 0}
hour_stats[hour]['total'] += 1
if episode[11]: # success column
hour_stats[hour]['success'] += 1
# Find patterns
for hour, stats in hour_stats.items():
if stats['total'] >= 2: # Minimum episodes
success_rate = stats['success'] / stats['total']
patterns.append({
'pattern': f"Episodes at hour {hour}",
'frequency': stats['total'],
'success_rate': success_rate
})
return sorted(patterns, key=lambda x: x['frequency'], reverse=True)
def _discover_agent_patterns(self, episodes: List[Tuple]) -> List[Dict[str, Any]]:
"""Discover agent-specific patterns"""
patterns = []
agent_stats = {}
for episode in episodes:
agent_id = episode[1] # agent_id column
episode_type = episode[5] # episode_type column
key = f"{agent_id}:{episode_type}"
if key not in agent_stats:
agent_stats[key] = {'total': 0, 'success': 0}
agent_stats[key]['total'] += 1
if episode[11]: # success column
agent_stats[key]['success'] += 1
# Convert to patterns
for key, stats in agent_stats.items():
if stats['total'] >= 3: # Minimum episodes
success_rate = stats['success'] / stats['total']
patterns.append({
'pattern': f"Agent pattern: {key}",
'frequency': stats['total'],
'success_rate': success_rate
})
return sorted(patterns, key=lambda x: x['success_rate'], reverse=True)
def _store_pattern(self, pattern_type: str, pattern: Dict[str, Any]):
"""Store discovered pattern"""
try:
pattern_id = str(uuid.uuid4())
with sqlite3.connect(self.db_path) as conn:
conn.execute("""
INSERT OR REPLACE INTO episode_patterns (
id, pattern_type, pattern_description,
frequency, success_rate
) VALUES (?, ?, ?, ?, ?)
""", (
pattern_id, pattern_type, pattern['pattern'],
pattern['frequency'], pattern['success_rate']
))
except Exception as e:
logger.error(f"Error storing pattern: {e}")
def get_episodic_statistics(self) -> Dict[str, Any]:
"""Get comprehensive episodic memory statistics"""
try:
with sqlite3.connect(self.db_path) as conn:
stats = {}
# Basic episode counts
cursor = conn.execute("SELECT COUNT(*) FROM episodes")
stats['total_episodes'] = cursor.fetchone()[0]
cursor = conn.execute("SELECT COUNT(*) FROM episodes WHERE success = 1")
stats['successful_episodes'] = cursor.fetchone()[0]
# Episode type distribution
cursor = conn.execute("""
SELECT episode_type, COUNT(*),
SUM(CASE WHEN success = 1 THEN 1 ELSE 0 END) as successes
FROM episodes
GROUP BY episode_type
""")
episode_types = {}
for row in cursor.fetchall():
episode_types[row[0]] = {
'total': row[1],
'successes': row[2],
'success_rate': row[2] / row[1] if row[1] > 0 else 0
}
stats['episode_types'] = episode_types
# Agent performance
cursor = conn.execute("""
SELECT agent_id, COUNT(*),
SUM(CASE WHEN success = 1 THEN 1 ELSE 0 END) as successes
FROM episodes
GROUP BY agent_id
""")
agent_performance = {}
for row in cursor.fetchall():
agent_performance[row[0]] = {
'total_episodes': row[1],
'successes': row[2],
'success_rate': row[2] / row[1] if row[1] > 0 else 0
}
stats['agent_performance'] = agent_performance
# Replay statistics
cursor = conn.execute("SELECT COUNT(*) FROM experience_replay")
stats['total_replays'] = cursor.fetchone()[0]
cursor = conn.execute("SELECT AVG(replay_effectiveness) FROM experience_replay")
avg_effectiveness = cursor.fetchone()[0]
stats['average_replay_effectiveness'] = avg_effectiveness or 0.0
# Pattern discovery
cursor = conn.execute("SELECT COUNT(*) FROM episode_patterns")
stats['discovered_patterns'] = cursor.fetchone()[0]
return stats
except Exception as e:
logger.error(f"Error getting episodic statistics: {e}")
return {'error': str(e)}
# Export the main classes
__all__ = ['EpisodicMemorySystem', 'Episode', 'ExperienceReplay']
|