Spaces:
Sleeping
Sleeping
File size: 11,245 Bytes
55ed80a | 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 | """
Agentic Workflow for Music Recommendations.
This module implements an AI agent that can:
1. Analyze user preferences
2. Generate recommendations
3. Evaluate recommendation quality
4. Refine recommendations based on feedback
5. Self-correct and improve through iteration
The agent follows a plan-act-evaluate loop for autonomous refinement.
"""
from typing import List, Dict, Tuple, Optional
from enum import Enum
import json
class AgentState(Enum):
"""States the recommendation agent can be in."""
ANALYZING = "analyzing"
GENERATING = "generating"
EVALUATING = "evaluating"
REFINING = "refining"
COMPLETE = "complete"
class RecommendationAgent:
"""
Agentic AI system that autonomously improves recommendations through iteration.
Workflow:
1. ANALYZE: Understand user preferences deeply
2. GENERATE: Create initial recommendations
3. EVALUATE: Assess recommendation quality
4. REFINE: Improve based on quality metrics
5. COMPLETE: Return final recommendations with confidence scores
"""
def __init__(self, recommender, rag_system=None):
"""
Initialize the agent.
Args:
recommender: The base recommender system
rag_system: Optional RAG system for LLM-enhanced explanations
"""
self.recommender = recommender
self.rag_system = rag_system
self.state = AgentState.ANALYZING
self.iteration_count = 0
self.max_iterations = 3
self.action_log: List[Dict] = []
self.quality_scores: List[float] = []
def plan(self, user_prefs: Dict) -> Dict:
"""
PLAN: Analyze user preferences and create a recommendation strategy.
Args:
user_prefs: User preference dictionary
Returns:
Strategy dictionary with analysis and approach
"""
strategy = {
"step": "plan",
"user_profile": user_prefs,
"constraints": [],
"scoring_mode": "default",
"diversity_focus": False,
}
# Analyze preference constraints
if user_prefs.get("energy", 0.5) > 0.8:
strategy["constraints"].append("high_energy_required")
elif user_prefs.get("energy", 0.5) < 0.4:
strategy["constraints"].append("low_energy_preferred")
if user_prefs.get("likes_acoustic"):
strategy["constraints"].append("prefer_acoustic")
else:
strategy["constraints"].append("prefer_electronic")
# Determine scoring strategy
mood = user_prefs.get("mood", "").lower()
if "intense" in mood or "aggressive" in mood:
strategy["scoring_mode"] = "energy_focused"
elif "chill" in mood or "relaxed" in mood:
strategy["scoring_mode"] = "mood_first"
strategy["num_recommendations"] = 5
strategy["diversity_penalty"] = len(strategy["constraints"]) > 2
self.action_log.append(strategy)
return strategy
def act(
self,
user_prefs: Dict,
strategy: Dict,
songs: List[Dict],
) -> List[Tuple[Dict, float, List[str]]]:
"""
ACT: Generate recommendations based on the strategy.
Args:
user_prefs: User preferences
strategy: Strategy from planning phase
songs: Available songs
Returns:
List of (song, score, reasons) tuples
"""
from .recommender import recommend_songs
self.state = AgentState.GENERATING
recommendations = recommend_songs(
user_prefs=user_prefs,
songs=songs,
k=strategy.get("num_recommendations", 5),
mode=strategy.get("scoring_mode", "default"),
diversity_penalty=strategy.get("diversity_penalty", False),
)
action = {
"step": "act",
"strategy_applied": strategy,
"recommendations_generated": len(recommendations),
"top_score": recommendations[0][1] if recommendations else 0,
}
self.action_log.append(action)
return recommendations
def evaluate(
self,
recommendations: List[Tuple[Dict, float, List[str]]],
user_prefs: Dict,
) -> Tuple[float, Dict]:
"""
EVALUATE: Assess the quality of recommendations.
Args:
recommendations: Generated recommendations
user_prefs: User preferences
Returns:
(quality_score, evaluation_dict) tuple
"""
self.state = AgentState.EVALUATING
if not recommendations:
return 0.0, {"error": "No recommendations generated"}
quality_metrics = {
"genre_match_rate": 0.0,
"mood_match_rate": 0.0,
"energy_alignment": 0.0,
"diversity_score": 0.0,
"avg_score": 0.0,
}
# Genre and mood match rates
genre_matches = sum(
1 for song, _, _ in recommendations
if song.get("genre") == user_prefs.get("genre")
)
quality_metrics["genre_match_rate"] = genre_matches / len(recommendations)
mood_matches = sum(
1 for song, _, _ in recommendations
if song.get("mood") == user_prefs.get("mood")
)
quality_metrics["mood_match_rate"] = mood_matches / len(recommendations)
# Energy alignment
target_energy = user_prefs.get("energy", 0.5)
energy_diffs = [
abs(song["energy"] - target_energy)
for song, _, _ in recommendations
]
quality_metrics["energy_alignment"] = 1.0 - (
sum(energy_diffs) / len(energy_diffs)
)
# Diversity (artist and genre diversity)
artists = set(song["artist"] for song, _, _ in recommendations)
genres = set(song["genre"] for song, _, _ in recommendations)
quality_metrics["diversity_score"] = (
len(artists) / len(recommendations) * 0.5 +
len(genres) / len(recommendations) * 0.5
)
# Average recommendation score
avg_score = sum(score for _, score, _ in recommendations) / len(
recommendations
)
quality_metrics["avg_score"] = avg_score
# Overall quality score (weighted)
quality_score = (
quality_metrics["genre_match_rate"] * 0.25 +
quality_metrics["mood_match_rate"] * 0.25 +
quality_metrics["energy_alignment"] * 0.20 +
quality_metrics["diversity_score"] * 0.15 +
min(avg_score / 4.5, 1.0) * 0.15 # Normalize avg score
)
self.quality_scores.append(quality_score)
evaluation = {
"step": "evaluate",
"iteration": self.iteration_count,
"quality_score": round(quality_score, 2),
"metrics": {k: round(v, 2) for k, v in quality_metrics.items()},
"pass_threshold": quality_score > 0.65,
}
self.action_log.append(evaluation)
return quality_score, evaluation
def refine(
self,
user_prefs: Dict,
quality_score: float,
strategy: Dict,
) -> Optional[Dict]:
"""
REFINE: Adjust strategy based on evaluation.
Args:
user_prefs: User preferences
quality_score: Quality score from evaluation
strategy: Current strategy
Returns:
Refined strategy or None if no refinement needed
"""
if quality_score > 0.65 or self.iteration_count >= self.max_iterations:
return None
self.state = AgentState.REFINING
refined_strategy = strategy.copy()
if quality_score < 0.4:
# Low quality: switch to different scoring mode
modes = ["default", "mood_first", "energy_focused"]
current_idx = modes.index(strategy.get("scoring_mode", "default"))
refined_strategy["scoring_mode"] = modes[(current_idx + 1) % len(modes)]
refined_strategy["diversity_penalty"] = True
elif quality_score < 0.6:
# Medium quality: increase diversity
refined_strategy["diversity_penalty"] = True
refinement = {
"step": "refine",
"iteration": self.iteration_count,
"quality_score": quality_score,
"changes": {
"old_mode": strategy.get("scoring_mode"),
"new_mode": refined_strategy.get("scoring_mode"),
"old_diversity": strategy.get("diversity_penalty"),
"new_diversity": refined_strategy.get("diversity_penalty"),
},
}
self.action_log.append(refinement)
return refined_strategy
def run(
self,
user_prefs: Dict,
songs: List[Dict],
) -> Tuple[List[Tuple[Dict, float, List[str]]], Dict]:
"""
Execute the full agentic workflow: plan → act → evaluate → refine (loop).
Args:
user_prefs: User preferences
songs: Available songs
Returns:
(final_recommendations, workflow_summary) tuple
"""
self.iteration_count = 0
best_recommendations = None
best_quality = 0.0
while self.iteration_count < self.max_iterations:
# PLAN
strategy = self.plan(user_prefs)
# ACT
recommendations = self.act(user_prefs, strategy, songs)
# EVALUATE
quality_score, evaluation = self.evaluate(recommendations, user_prefs)
# Store best result
if quality_score > best_quality:
best_quality = quality_score
best_recommendations = recommendations
# Check if quality is acceptable
if evaluation.get("pass_threshold"):
self.state = AgentState.COMPLETE
break
# REFINE
refined_strategy = self.refine(user_prefs, quality_score, strategy)
if refined_strategy is None:
self.state = AgentState.COMPLETE
break
self.iteration_count += 1
self.state = AgentState.COMPLETE
summary = {
"total_iterations": self.iteration_count,
"final_quality_score": round(best_quality, 2),
"quality_improvement": (
round(self.quality_scores[-1] - self.quality_scores[0], 2)
if len(self.quality_scores) > 1
else 0.0
),
"workflow_log": self.action_log,
}
return best_recommendations or [], summary
def export_workflow(self) -> str:
"""Export the agent's workflow and decisions as JSON."""
return json.dumps(
{
"final_state": self.state.value,
"iterations": self.iteration_count,
"quality_scores": self.quality_scores,
"action_log": self.action_log,
},
indent=2,
)
|