btc-Event-Contracts

KO | EN

gitlite — search

#hacktoberfest #kubernetes #tensorflow #privacy #cpp #agent #monitoring #go #swift #machine-learning

btc-Event-Contracts

(★ 9)

btc事件合约

파일 탐색기

ZIP 다운로드

- - __init__.py
  - server.py
- - __init__.py
  - simulator.py
- - __init__.py
  - app.py
- - __init__.py
  - fetcher.py
  - integrity.py
- - __init__.py
  - builder.py
  - CLAUDE.md
  - cross_tf.py
  - indicators.py
- - __init__.py
  - CLAUDE.md
  - ensemble.py
  - predictor.py
  - scheduler.py
- - __init__.py
  - ablation.py
  - evaluate.py
  - labels.py
  - registry.py
  - split.py
  - train.py
  - tune.py
- - __init__.py
  - ev.py
  - kelly.py
- __init__.py
- config.py
- - ralph.sh
- backfill_history.py
- cleanup_models.py
- manual_tune.py
- migrate_predictions_db.py
- predict_signal.py
- retrain.py
- rollback_model.py
- run_scheduler.py
- run_service.py
- train_all.py
- dual_confirm_instruction.md
- prd-00-common.md
- prd-01-phase1-fixes.md
- prd-100-training-pipeline.md
- prd-99-optional-enhancements.md
- prd-archive-original.md
- prd-btc-event-contract-predictor.md
- __init__.py
- test_ablation.py
- test_api.py
- test_backfill.py
- test_backtest.py
- test_backtest_split.py
- test_cleanup_models.py
- test_config.py
- test_cross_tf.py
- test_dashboard.py
- test_ensemble.py
- test_evaluate.py
- test_features.py
- test_fetcher.py
- test_labels_split.py
- test_manual_tune.py
- test_migrate_predictions_db.py
- test_model_versioning.py
- test_plan_doc.py
- test_predictor.py
- test_retrain.py
- test_retrain_pnl.py
- test_rollback_model.py
- test_scheduler.py
- test_strategy.py
- test_train.py
- test_train_all_cli.py
- test_tune.py
.gitignore
btc-event-contract-predictor-plan.md
check.py
CLAUDE.md
prd.json
progress.txt
ralph.sh
README.md
README_dual_predict.md
run_checks.py
run_mypy.bat
run_tests.bat

# CDN으로 사용하기

jsDelivr

jsDelivr는 공개 GitHub 리포지토리를 별도 설정 없이 CDN으로 즉시 서빙합니다. 버전과 파일을 고르면 웹페이지에 바로 붙일 수 있는 링크와 예시 코드가 만들어집니다.

버전

파일

링크

예시

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

Predict BTC price direction for Binance event contracts using dual-interval (1h + 5m) XGBoost models with calibrated probabilities, Kelly position sizing, and automated scheduling.

## Installation

```bash
pip install -e .[dev]
```

## Entry Points

### Training

```bash
# Train both intervals
python scripts/train_all.py

# With Optuna tuning
python scripts/train_all.py --tune

# With feature ablation
python scripts/train_all.py --ablate

# Retrain (only saves model if val_logloss improves)
python scripts/retrain.py
```

### API Server

```bash
uvicorn btc_predictor.api.server:app --host 0.0.0.0 --port 8000
```

### Scheduler (separate process)

```bash
python scripts/run_scheduler.py
```

### Dashboard

```bash
streamlit run btc_predictor/dashboard/app.py
```

### Backfill Historical Data

```bash
python scripts/backfill_history.py
```

### Evaluation

```bash
# Generate HTML evaluation reports
python -c "from btc_predictor.models.evaluate import evaluate_all; evaluate_all()"
```

## Project Structure

```
btc_predictor/
  config.py          # Central configuration
  data/fetcher.py    # Binance OHLCV data fetcher
  features/
    builder.py       # Feature engineering
    cross_tf.py      # Cross-timeframe features
  models/
    train.py         # XGBoost training with calibration
    evaluate.py      # Final test-set evaluation
    registry.py      # Model save/load
    split.py         # Time-based train/val/test split
  strategy/
    ev.py            # Expected value computation
    kelly.py         # Kelly criterion position sizing
  inference/
    predictor.py     # Single-interval prediction
    ensemble.py      # Dual-interval consistency
    scheduler.py     # Automated scheduling
  api/server.py      # FastAPI endpoints
  dashboard/app.py   # Streamlit dashboard
scripts/
  train_all.py       # Multi-interval training CLI
  retrain.py         # Retrain with improvement check
  backfill_history.py # Historical data download
  run_service.py     # API server entrypoint
  run_scheduler.py   # Scheduler entrypoint
```

## Known Limitations

1. Event-contract payout rate is 80% (odds 1.80x), with a break-even probability of 55.56%. **Consistent long-term profit is extremely difficult**; this system does not constitute investment advice.
2. 5m interval has high noise, AUC ceiling ~0.53; 1h ~0.56.
3. The model generalizes poorly to structural events (black swans, policy changes).
4. Probability calibration depends on training distribution stability; retraining is needed after market regime changes.
5. Backtesting does not include slippage, order-book depth, or exchange latency; live performance may be significantly worse than backtest results.

btc-Event-Contracts

파일 탐색기

# CDN으로 사용하기

Was this content helpful?

Recent Feedback

Download README