How kartikeybishnoi/grow21-ml works
Grow21 is a course-project ML research layer for an existing adaptive learning app, not a replacement app or service. Its central experiment workflow generates learner events through a read-only bridge to the external simulator, builds causal features, and compares classical models; separate experiments investigate touch-slip trust weighting, Q-learning controllers, and cold-start clustering. The collection page supplies touch trials for the slip investigation. The README documents these capability branches, while the sampled sources establish only some internal calls; uncertain wiring is omitted.
Main entry points: simulator_bridge.py Simulator bridge · run_hyperparameter_tuning.py Model comparison · run_q_learning.py Controller comparison · slip_and_trust.py Slip and trust
Components 14
Simulation data
- Simulator bridge
src/grow21_ml/simulator_bridge.py
Skill prediction
- Causal features
src/grow21_ml/feature_engineering.py - BKT baseline
src/grow21_ml/bayesian_knowledge_tracing.py - Classical models
src/grow21_ml/classification_models.py
Motor reliability
- Touch collection
collect/index.html - Touch trials
- Slip and trust
src/grow21_ml/slip_and_trust.py
Adaptive policies
- Q-learning
src/grow21_ml/q_learning_controller.py - Cold-start clustering
src/grow21_ml/clustering_coldstart.py
Experiments
- Model comparison
run_hyperparameter_tuning.py - Controller comparison
run_q_learning.py - Experiment results
results
Other components
- Researcher
- Grow21 simulator
Connections 17
- Researcher → Simulator bridge: runs experiments
README.md - Simulator bridge → Grow21 simulator: imports simulator
src/grow21_ml/simulator_bridge.py - Model comparison → Simulator bridge: generates splits
run_hyperparameter_tuning.py - Model comparison → Causal features: builds features
run_hyperparameter_tuning.py - Causal features → Classical models: provides features
- Researcher → Touch collection: collects trials
README.md - Touch collection → Touch trials: downloads JSON
README.md - Touch trials → Slip and trust: loads trials
README.md - Simulator bridge → Slip and trust: supplies simulated events
src/grow21_ml/slip_and_trust.py - Slip and trust → Q-learning: uses simulator helpers
src/grow21_ml/slip_and_trust.py - Cold-start clustering → Q-learning: uses simulator helpers
src/grow21_ml/clustering_coldstart.py - Cold-start clustering → Simulator bridge: uses simulator bridge
src/grow21_ml/clustering_coldstart.py - Researcher → Model comparison: runs comparison
README.md - Researcher → Controller comparison: runs comparison
README.md - Controller comparison → Q-learning: trains and evaluates
run_q_learning.py - Model comparison → Experiment results: writes metrics
run_hyperparameter_tuning.py - Controller comparison → Experiment results: writes metrics
run_q_learning.py