How majdzidan-dev/reinforcement-learning-proj works
RLProject trains and compares four SARSA agents on Gymnasium Taxi-v4, using default, sparse, and custom rewards across five seeds. Its principal flow is experiment orchestration, agent training, metric collection, and comparison plotting; separate workflows sweep hyperparameters and record gameplay videos. The graph preserves those documented capabilities, but internals of the metrics and sweep runners were not sampled, so their internal relationships are not expanded.
Main entry points: create_animation.py Gameplay videos · main.py Run experiments · helpers.py Training helpers · PSSARSA.py Prioritized SARSA
Components 16
Experiment orchestration
- Run experiments
main.py - Seed metrics
metrics_test.py - Parameter sweeps
sweep_test.py - Training helpers
helpers.py - Episode metrics
metrics_wrapper.py
SARSA agents
- SARSA
algorithms/SARSA.py - Generalized SARSA
algorithms/SARSAGen.py - Prioritized SARSA
algorithms/PSSARSA.py - Combined SARSA
algorithms/PSSARSAGen.py
Taxi rewards
- Gymnasium Taxi
- Sparse rewards
rewards/Sparse.py - Custom rewards 1
rewards/Custom1.py - Custom rewards 2
rewards/Custom2.py
Analysis and media
- Comparison plots
draw_metrics.py - Gameplay videos
create_animation.py
Other components
- Engineer
Connections 17
- Engineer → Run experiments: runs
main.py - Run experiments → Seed metrics: runs
main.py - Run experiments → Comparison plots: draws comparisons
main.py - Run experiments → Parameter sweeps: runs
main.py - Run experiments → Gameplay videos: records gameplay
main.py - Training helpers → Episode metrics: wraps environment
helpers.py - Generalized SARSA → SARSA: extends
algorithms/SARSAGen.py - Prioritized SARSA → SARSA: extends
algorithms/PSSARSA.py - Combined SARSA → Prioritized SARSA: extends
algorithms/PSSARSAGen.py - Gameplay videos → Training helpers: trains with
create_animation.py - Gameplay videos → Combined SARSA: instantiates
create_animation.py - Gameplay videos → Gymnasium Taxi: creates environment
create_animation.py - Gameplay videos → Sparse rewards: wraps with
create_animation.py - Gameplay videos → Custom rewards 1: wraps with
create_animation.py - Gameplay videos → Custom rewards 2: wraps with
create_animation.py - Custom rewards 1 → Gymnasium Taxi: wraps
rewards/Custom1.py - Custom rewards 2 → Gymnasium Taxi: wraps
rewards/Custom2.py