How charan9665/advml2 works
MovieMind is a full-stack personalized movie recommender. A profile selects a MovieLens user; the API serves catalog recommendations, accepts ratings, converts them to rewards, updates per-profile Q-learning state, and persists feedback for later recommendations and history. Separate offline workflows compare epsilon strategies and evaluate tabular and fitted Q-learning against popularity using chronological MovieLens holdouts. The graph preserves those capabilities and poster lookup; unsampled route details are represented only where documented, so some internal links are omitted.
Main entry points: main.py FastAPI service · evaluate.py Offline evaluation · q_learning.py Tabular Q-learner · data_loader.py MovieLens loader
Components 19
Web interface
- Web app
frontend/src/App.jsx
API and catalog
- FastAPI service
backend/main.py - MovieLens loader
backend/data_loader.py - Movie catalog
backend/main.py - SQLite persistence
backend/database.py - Poster lookup
backend/poster_loader.py - Popularity ranking
backend/recommender/popularity.py
Interactive learning
- Profile preferences
backend/main.py - Preference state
backend/rl/state_manager.py - Rating reward
backend/rl/reward.py - Tabular Q-learner
backend/rl/q_learning.py - Persisted Q-table
backend/main.py - Rating feedback
backend/main.py
Offline models and evaluation
- Fitted Q model
backend/rl/fitted_q_learning.py - Offline evaluation
experiments/evaluate.py - Ranking metrics
backend/recommender/evaluation.py - Epsilon replay
experiments/epsilon_experiments.py
Other components
- MovieMind user
- Wikipedia
Connections 26
- MovieMind user → Web app: uses
README.md - Web app → FastAPI service: requests API
frontend/src/App.jsx - FastAPI service → MovieLens loader: loads dataset
backend/main.py - FastAPI service → SQLite persistence: persists profile data
README.md - FastAPI service → Movie catalog: serves catalog
backend/main.py - FastAPI service → Poster lookup: loads poster
backend/main.py - Poster lookup → Wikipedia: looks up artwork
backend/poster_loader.py - FastAPI service → Popularity ranking: ranks popular titles
README.md - FastAPI service → Profile preferences: loads preferences
README.md - Profile preferences → Preference state: derives state
backend/rl/state_manager.py - MovieMind user → Rating feedback: rates recommendation
README.md - FastAPI service → Rating feedback: accepts rating
README.md - Rating feedback → Rating reward: maps rating
README.md - Rating feedback → Preference state: updates preferences
README.md - Rating feedback → Tabular Q-learner: updates Q-values
README.md - Tabular Q-learner → Persisted Q-table: saves policy
README.md - Persisted Q-table → Tabular Q-learner: restores policy
README.md - FastAPI service → SQLite persistence: stores feedback history
README.md - Offline evaluation → MovieLens loader: loads ratings
experiments/evaluate.py - Offline evaluation → Tabular Q-learner: replays training
experiments/evaluate.py - Offline evaluation → Popularity ranking: builds baseline
experiments/evaluate.py - Offline evaluation → Fitted Q model: trains model
experiments/evaluate.py - Offline evaluation → Ranking metrics: measures rankings
experiments/evaluate.py - Epsilon replay → MovieLens loader: loads ratings
experiments/epsilon_experiments.py - Epsilon replay → Tabular Q-learner: compares policies
experiments/epsilon_experiments.py - FastAPI service → Ranking metrics: runs training evaluation
backend/main.py