How aaqibzahid/pokemon-battlestate-analytics works
Pokemon BattleState Analytics benchmarks real-time win-probability estimates from Pokémon Showdown replays. Its main workflow collects replay URLs, downloads logs, parses events, reconstructs turn-by-turn battle states, derives observable features, splits by battle, trains baselines, and evaluates predictions. The Streamlit app and replay-prediction module are an additional user-facing branch. Edges follow README pipeline documentation and SOURCE INDEX references; unsampled implementation details and UI call paths are not asserted.
Main entry points: evaluate_models.py Evaluation pipeline · build_dataset.py Dataset builder · state_machine.py State machine · parser.py Log parser
Components 27
Replay ingestion
- Replay collector
src/battletrace/ingest/collect_replays.py - Replay URL list
data/raw/replay_urls.txt - Log downloader
src/battletrace/ingest/download_logs.py - Raw battle logs
Battle reconstruction
- Dataset builder
src/battletrace/pipeline/build_dataset.py - Log parser
src/battletrace/parsing/parser.py - Protocol definitions
src/battletrace/parsing/protocol.py - Battle events
src/battletrace/parsing/events.py - State machine
src/battletrace/state/state_machine.py - Battle state
src/battletrace/state/battle_state.py - Turn snapshots
src/battletrace/state/snapshot.py - Reconstructed states
data/processed
Features and models
- Feature extraction
src/battletrace/features/extract_features.py - Feature dataset
data/processed - Battle-level splits
data/processed/splits - Training pipeline
src/battletrace/pipeline/train_baselines.py - Classical baselines
src/battletrace/models/baselines.py - Heuristic baseline
src/battletrace/models/heuristics.py
Evaluation
- Evaluation pipeline
src/battletrace/pipeline/evaluate_models.py - Probability metrics
src/battletrace/evaluation/metrics.py - Calibration analysis
src/battletrace/evaluation/calibration.py - Turn-wise analysis
src/battletrace/evaluation/turnwise_eval.py - Evaluation reports
reports/metrics
User experience
- Streamlit app
app/streamlit_app.py - Replay prediction
src/battletrace/demo/predict_replay.py
Other components
- Showdown API
- Analyst
Connections 28
- Showdown API → Replay collector: provides replays
README.md - Replay collector → Replay URL list: writes URLs
README.md - Replay URL list → Log downloader: supplies URLs
README.md - Log downloader → Raw battle logs: downloads logs
README.md - Raw battle logs → Dataset builder: provides logs
README.md - Dataset builder → Log parser: parses logs
src/battletrace/pipeline/build_dataset.py - Log parser → Protocol definitions: uses definitions
src/battletrace/parsing/parser.py - Log parser → Battle events: creates events
src/battletrace/parsing/parser.py - Dataset builder → State machine: reconstructs states
src/battletrace/pipeline/build_dataset.py - State machine → Battle events: dispatches events
src/battletrace/state/state_machine.py - State machine → Battle state: updates state
src/battletrace/state/state_machine.py - State machine → Turn snapshots: emits snapshots
src/battletrace/state/state_machine.py - Dataset builder → Reconstructed states: writes states
README.md - Reconstructed states → Feature extraction: supplies snapshots
README.md - Feature extraction → Feature dataset: derives features
README.md - Feature dataset → Battle-level splits: splits by battle
README.md - Battle-level splits → Training pipeline: supplies partitions
README.md - Training pipeline → Classical baselines: trains models
src/battletrace/pipeline/train_baselines.py - Training pipeline → Heuristic baseline: trains baseline
src/battletrace/pipeline/train_baselines.py - Feature dataset → Evaluation pipeline: evaluates predictions
README.md - Evaluation pipeline → Classical baselines: evaluates models
src/battletrace/pipeline/evaluate_models.py - Evaluation pipeline → Heuristic baseline: evaluates model
src/battletrace/pipeline/evaluate_models.py - Evaluation pipeline → Probability metrics: computes metrics
src/battletrace/pipeline/evaluate_models.py - Evaluation pipeline → Calibration analysis: computes calibration
src/battletrace/pipeline/evaluate_models.py - Evaluation pipeline → Turn-wise analysis: computes turnwise
src/battletrace/pipeline/evaluate_models.py - Evaluation pipeline → Evaluation reports: writes results
README.md - Analyst → Streamlit app: uses interface
README.md - Streamlit app → Replay prediction: uses predictor
app/streamlit_app.py