File size: 18,571 Bytes
5e38446 | 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 | """
Reinforcement Learning Doubt Prediction Agent
This agent predicts what doubts a user will have BEFORE they occur,
using:
- User's learning history
- Current topic complexity
- Behavioral signals (eye tracking, hesitation, scroll patterns)
- Similar users' learning patterns
- Topic dependency graphs
Based on Deep Q-Learning with attention mechanism
"""
import numpy as np
from typing import Dict, List, Any, Optional, Tuple
from dataclasses import dataclass, field
from datetime import datetime
import json
@dataclass
class LearningState:
"""Represents the current learning state"""
topic: str
subtopic: str
progress_percentage: float
time_spent_seconds: int
confusion_signals: float
eye_tracking_confidence: float
scroll_reversals: int
selection_count: int
previous_doubts_count: int
mastery_level: float
difficulty_rating: float
time_of_day: int
streak_days: int
@dataclass
class DoubtPrediction:
"""Predicted doubt with confidence"""
predicted_doubt: str
confidence: float
suggested_explanation: str
related_concepts: List[str]
priority: int
estimated_resolution_time: int
prerequisite_topics: List[str]
@dataclass
class RLPolicy:
"""RL Policy network (simplified)"""
state_dim: int = 12
action_dim: int = 100
learning_rate: float = 0.001
gamma: float = 0.95
epsilon: float = 1.0
epsilon_decay: float = 0.995
epsilon_min: float = 0.01
q_table: Dict[str, np.ndarray] = field(default_factory=dict)
state_mean: np.ndarray = None
state_std: np.ndarray = None
class DoubtPredictorAgent:
"""
RL-based agent that predicts user doubts before they occur.
Uses a Deep Q-Network inspired architecture with:
- State encoding from learning signals
- Attention mechanism for topic relationships
- Experience replay for learning
- Progressive prediction confidence
"""
def __init__(self, user_id: str, config: Optional[Dict] = None):
self.user_id = user_id
self.config = config or {}
self.policy = RLPolicy()
self.experience_buffer = []
self.max_buffer_size = 1000
self.topic_embeddings = {}
self.concept_graph = {}
self.user_preferences = {}
self._initialize_topic_knowledge()
def _initialize_topic_knowledge(self):
"""Initialize base topic relationships"""
self.concept_graph = {
'python': ['variables', 'functions', 'classes', 'loops', 'conditionals', 'data_structures'],
'machine_learning': ['linear_regression', 'classification', 'neural_networks', 'optimization', 'feature_engineering'],
'deep_learning': ['perceptrons', 'backpropagation', 'convolutional_nets', 'recurrent_nets', 'transformers', 'attention'],
'statistics': ['probability', 'distributions', 'hypothesis_testing', 'regression', 'bayesian'],
'calculus': ['derivatives', 'integrals', 'limits', 'series', 'multivariable'],
'linear_algebra': ['vectors', 'matrices', 'eigenvalues', 'transformations', 'decompositions']
}
self.doubt_templates = {
'variables': [
"What is the difference between mutable and immutable types?",
"How does variable scope work in nested functions?",
"When should I use global vs local variables?"
],
'functions': [
"What is the difference between arguments and parameters?",
"How do *args and **kwargs work?",
"When should I use lambda functions?"
],
'classes': [
"What is the difference between class and instance attributes?",
"How does inheritance work with multiple inheritance?",
"What are abstract base classes and when to use them?"
],
'loops': [
"When should I use for vs while loops?",
"How do list comprehensions replace loops?",
"What is the difference between break and continue?"
],
'data_structures': [
"When should I use lists vs dictionaries?",
"What is the time complexity of dictionary operations?",
"How do sets differ from lists in performance?"
],
'linear_regression': [
"What is the cost function and how is it optimized?",
"How do I handle multicollinearity?",
"What are the assumptions of linear regression?"
],
'neural_networks': [
"What is the role of activation functions?",
"How does backpropagation compute gradients?",
"What is the vanishing gradient problem?"
],
'transformers': [
"How does self-attention work?",
"What is the difference between encoder and decoder?",
"Why is positional encoding needed?"
]
}
def get_current_state(self, learning_context: Dict) -> LearningState:
"""Extract current learning state from context"""
return LearningState(
topic=learning_context.get('topic', 'unknown'),
subtopic=learning_context.get('subtopic', 'unknown'),
progress_percentage=learning_context.get('progress', 0.0),
time_spent_seconds=learning_context.get('time_spent', 0),
confusion_signals=learning_context.get('confusion_score', 0.0),
eye_tracking_confidence=learning_context.get('eye_confidence', 0.0),
scroll_reversals=learning_context.get('scroll_reversals', 0),
selection_count=learning_context.get('selections', 0),
previous_doubts_count=learning_context.get('prev_doubts', 0),
mastery_level=learning_context.get('mastery', 0.0),
difficulty_rating=learning_context.get('difficulty', 0.5),
time_of_day=datetime.now().hour,
streak_days=learning_context.get('streak', 0)
)
def state_to_vector(self, state: LearningState) -> np.ndarray:
"""Convert state to feature vector"""
features = [
self._topic_to_feature(state.topic),
self._topic_to_feature(state.subtopic),
state.progress_percentage,
np.log1p(state.time_spent_seconds) / 10,
state.confusion_signals,
state.eye_tracking_confidence,
np.tanh(state.scroll_reversals / 10),
np.tanh(state.selection_count / 20),
np.tanh(state.previous_doubts_count / 50),
state.mastery_level,
state.difficulty_rating,
np.sin(2 * np.pi * state.time_of_day / 24),
np.cos(2 * np.pi * state.time_of_day / 24),
np.tanh(state.streak_days / 30)
]
return np.array(features, dtype=np.float32)
def _topic_to_feature(self, topic: str) -> float:
"""Convert topic to numerical feature"""
topic_lower = topic.lower().replace(' ', '_')
topic_order = [
'variables', 'functions', 'classes', 'loops', 'conditionals', 'data_structures',
'probability', 'distributions', 'derivatives', 'integrals', 'vectors', 'matrices',
'linear_regression', 'classification', 'neural_networks', 'optimization',
'convolutional_nets', 'recurrent_nets', 'transformers', 'attention'
]
if topic_lower in topic_order:
return topic_order.index(topic_lower) / len(topic_order)
return 0.5
def predict_doubts(
self,
learning_context: Dict,
top_k: int = 5,
gesture_influence: Optional[float] = None
) -> List[DoubtPrediction]:
"""
Predict likely doubts for current learning context.
Uses RL policy to estimate which doubts are most likely,
based on current state and historical patterns.
Args:
learning_context: Current learning state
top_k: Number of predictions to return
gesture_influence: Optional gesture-based signal (0-1) that increases doubt confidence
"""
state = self.get_current_state(learning_context)
state_vec = self.state_to_vector(state)
predictions = []
related_concepts = self._get_related_concepts(state.topic, state.subtopic)
for concept in related_concepts:
if concept not in self.doubt_templates:
continue
templates = self.doubt_templates[concept]
for template in templates:
confidence = self._calculate_doubt_confidence(
state, concept, template, gesture_influence
)
if confidence > 0.3:
prerequisite = self._get_prerequisites(concept)
prediction = DoubtPrediction(
predicted_doubt=template,
confidence=confidence,
suggested_explanation=self._generate_explanation_hint(concept, template),
related_concepts=self._get_related_concepts(concept, ''),
priority=self._calculate_priority(state, confidence),
estimated_resolution_time=self._estimate_time(concept),
prerequisite_topics=prerequisite
)
predictions.append(prediction)
predictions.sort(key=lambda x: x.priority, reverse=True)
return predictions[:top_k]
def _calculate_doubt_confidence(
self,
state: LearningState,
concept: str,
template: str,
gesture_influence: Optional[float] = None
) -> float:
"""Calculate confidence that user will have this doubt"""
base_confidence = 0.5
if state.confusion_signals > 0.7:
base_confidence += 0.2
if state.eye_tracking_confidence < 0.5:
base_confidence += 0.15
if state.scroll_reversals > 5:
base_confidence += 0.1
if concept in self.concept_graph.get(state.topic.lower(), []):
base_confidence += 0.1
if state.difficulty_rating > 0.7:
base_confidence += 0.15
if state.mastery_level < 0.3:
base_confidence += 0.1
if gesture_influence is not None and gesture_influence > 0.5:
base_confidence += (gesture_influence - 0.5) * 0.4
difficulty_penalty = state.difficulty_rating * 0.1
base_confidence -= difficulty_penalty
return min(max(base_confidence, 0.0), 1.0)
def _get_related_concepts(self, topic: str, subtopic: str) -> List[str]:
"""Get concepts related to current topic"""
topic_lower = topic.lower().replace(' ', '_')
subtopic_lower = subtopic.lower().replace(' ', '_')
related = []
if topic_lower in self.concept_graph:
related.extend(self.concept_graph[topic_lower])
if subtopic_lower in self.concept_graph:
related.extend(self.concept_graph[subtopic_lower])
for t, concepts in self.concept_graph.items():
for c in concepts:
if t == topic_lower or c == subtopic_lower:
related.extend(concepts)
return list(set(related))[:10]
def _get_prerequisites(self, concept: str) -> List[str]:
"""Get prerequisite concepts that should be understood first"""
prereq_map = {
'neural_networks': ['linear_regression', 'calculus', 'linear_algebra'],
'transformers': ['neural_networks', 'attention', 'linear_algebra'],
'convolutional_nets': ['neural_networks', 'linear_algebra'],
'backpropagation': ['derivatives', 'chain_rule'],
'optimization': ['calculus', 'derivatives'],
'classification': ['probability', 'linear_regression'],
}
return prereq_map.get(concept, [])
def _generate_explanation_hint(self, concept: str, template: str) -> str:
"""Generate a quick explanation hint"""
hints = {
'variables': 'Variables store data values that can be changed or accessed later.',
'functions': 'Functions are reusable blocks of code that perform specific tasks.',
'classes': 'Classes define blueprints for creating objects with attributes and methods.',
'neural_networks': 'Neural networks learn patterns through weighted connections between neurons.',
'transformers': 'Transformers use self-attention to process sequences in parallel.',
'backpropagation': 'Backpropagation calculates gradients by propagating errors backwards through the network.'
}
return hints.get(concept, 'Review the fundamental concepts before proceeding.')
def _calculate_priority(self, state: LearningState, confidence: float) -> float:
"""Calculate priority score for doubt prediction"""
priority = confidence * 0.4
priority += state.confusion_signals * 0.2
priority += (1 - state.mastery_level) * 0.2
priority += state.difficulty_rating * 0.1
priority += min(state.time_spent_seconds / 600, 1) * 0.1
return priority
def _estimate_time(self, concept: str) -> int:
"""Estimate time to resolve doubt in minutes"""
time_map = {
'variables': 5,
'functions': 10,
'classes': 15,
'loops': 8,
'data_structures': 20,
'linear_regression': 25,
'neural_networks': 30,
'transformers': 45,
'backpropagation': 35
}
return time_map.get(concept, 15)
def update_policy(
self,
state: LearningState,
predicted_doubt: str,
actual_doubt: str,
reward: float
):
"""
Update RL policy based on whether prediction was correct.
Positive reward if predicted doubt matches actual doubt.
Negative reward for false positives.
"""
state_key = self._state_to_key(state)
if state_key not in self.policy.q_table:
self.policy.q_table[state_key] = np.zeros(self.policy.action_dim)
action_idx = self._doubt_to_action(predicted_doubt)
current_q = self.policy.q_table[state_key][action_idx]
max_next_q = np.max(self.policy.q_table.get(state_key, [0]))
new_q = current_q + self.policy.learning_rate * (
reward + self.policy.gamma * max_next_q - current_q
)
self.policy.q_table[state_key][action_idx] = new_q
self.experience_buffer.append({
'state': state,
'predicted': predicted_doubt,
'actual': actual_doubt,
'reward': reward,
'timestamp': datetime.now().isoformat()
})
if len(self.experience_buffer) > self.max_buffer_size:
self.experience_buffer.pop(0)
if self.policy.epsilon > self.policy.epsilon_min:
self.policy.epsilon *= self.policy.epsilon_decay
def _state_to_key(self, state: LearningState) -> str:
"""Convert state to hashable key"""
return f"{state.topic}_{state.subtopic}_{int(state.progress_percentage * 10)}_{int(state.confusion_signals * 10)}"
def _doubt_to_action(self, doubt: str) -> int:
"""Convert doubt to action index"""
doubt_hash = hash(doubt.lower().strip())
return abs(doubt_hash) % self.policy.action_dim
def get_learning_recommendations(self, learning_context: Dict) -> Dict[str, Any]:
"""Get personalized learning recommendations based on predictions"""
predictions = self.predict_doubts(learning_context, top_k=3)
state = self.get_current_state(learning_context)
recommendations = {
'next_topics': [],
'review_topics': [],
'practice_exercises': [],
'estimated_difficulty': state.difficulty_rating,
'predicted_struggles': [p.predicted_doubt for p in predictions],
'confidence_boosters': [],
'optimal_break_time': self._suggest_break_time(learning_context)
}
if state.confusion_signals > 0.7:
recommendations['next_topics'] = self._get_prerequisites(state.topic)
recommendations['confidence_boosters'].append('Review prerequisite concepts')
if state.mastery_level > 0.8:
recommendations['next_topics'].append(state.topic)
recommendations['practice_exercises'].append(f"Advanced {state.topic} project")
if state.time_spent_seconds > 1800:
recommendations['suggest_break'] = True
recommendations['break_duration'] = 5
return recommendations
def _suggest_break_time(self, context: Dict) -> Optional[str]:
"""Suggest optimal break time based on learning patterns"""
if context.get('confusion_score', 0) > 0.6:
return "Take a 5-minute break to process information"
elif context.get('time_spent', 0) > 2400:
return "Take a longer 15-minute break"
return None
def export_model(self) -> Dict:
"""Export model state for persistence"""
return {
'user_id': self.user_id,
'q_table_size': len(self.policy.q_table),
'experience_buffer_size': len(self.experience_buffer),
'epsilon': self.policy.epsilon,
'concepts': list(self.concept_graph.keys()),
'doubt_templates': list(self.doubt_templates.keys())
}
def import_model(self, model_data: Dict):
"""Import model state from persistence"""
if 'concepts' in model_data:
for concept in model_data['concepts']:
if concept not in self.concept_graph:
self.concept_graph[concept] = []
|