triple-barrier

KO | EN

gitlite — search

#python #ai #typescript #llm #golang #android #deep-learning #macos #java #kubernetes #vue #pytorch #security #mcp

(★ 12)

No description available.

triple-barrier 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Triple-Barrier ML Strategy

Source code for the **Triple-Barrier Machine Learning** YouTube video by [Jesse](https://jesse.trade).

This repository contains a complete end-to-end example of how to build, train, and deploy a machine-learning-powered trading strategy using Jesse built-in ML pipeline, verified against deterministic synthetic data before touching real market data.

---

## Prerequisites

You must have **Jesse already installed** before using this repository. Follow the official installation guide here:

**[https://docs.jesse.trade/docs/getting-started/](https://docs.jesse.trade/docs/getting-started/)**

---

## What is the Triple-Barrier Method?

The triple-barrier method is a labeling technique for financial time series introduced by Marcos Lopez de Prado in *Advances in Financial Machine Learning*. Instead of labeling a bar based on a fixed future horizon, it labels each observation based on which of three barriers is hit first:

| Barrier | Condition | Label |
|---|---|---|
| Upper barrier | Price rises by distance | +1 (profit target) |
| Lower barrier | Price falls by distance | -1 (stop-loss) |
| Vertical barrier | N bars elapse | 0 (time expiry) |

The barrier distance is max(ATR * 2, price * 1%), which scales naturally with volatility - wide in volatile markets, tight in quiet ones.

---

## Repository Structure

```
triple-barrier-ml/
triple_barrier.py              # TDD / validation script (gather -> train -> deploy)
strategies/
  TripleBarrierTest/
    __init__.py                # The strategy itself (gather + deploy modes)
README.md
```

### triple_barrier.py - the TDD script

Generates 6 months of synthetic 1-minute candles across 16 deterministic price regimes (4 repetitions of up -> sideways -> down -> sideways) and runs the full pipeline in three phases:

- **Phase 1 - Gather:** Runs the strategy in gather mode to collect triple-barrier-labeled feature samples.
- **Phase 2 - Train:** Trains a Random Forest classifier and saves model artefacts to strategies/TripleBarrierTest/.
- **Phase 3 - Deploy:** Runs a full backtest in deploy mode using the saved model to confirm the pipeline produces actual trades.

### strategies/TripleBarrierTest/ - the strategy

A Jesse strategy class that operates in two modes controlled by the ml_mode class variable:

- **gather** - runs the triple-barrier labeling loop in before(), records features and labels, opens no real trades.
- **deploy** - loads the trained model and enters trades based on ML confidence thresholds, with stop-loss and take-profit mirroring the labeling barriers.

**10 features (all dimensionless):**

| Feature | Description |
|---|---|
| adx_centered | (ADX - 25) / 25 - trend strength, centred at the trending threshold |
| atr_pct | ATR / price - volatility as a fraction of price |
| ema21_50_ratio | (EMA21 - EMA50) / EMA50 - medium-term trend alignment |
| ema9_21_ratio | (EMA9 - EMA21) / EMA21 - short-term trend alignment |
| ema9_dist | (price - EMA9) / EMA9 - stretch from the fast EMA |
| keltner_pos | Position within the Keltner channel [0..1] |
| log_return_1 | log(close[-1] / close[-2]) - single-bar momentum |
| log_return_5 | log(close[-1] / close[-6]) - five-bar momentum |
| rsi_centered | (RSI - 50) / 50 - RSI scaled to [-1, +1] |
| supertrend_dist | (price - supertrend) / ATR - ATR-normalised distance from the trend line |

---

## Usage

Place the strategies/TripleBarrierTest/ folder inside your Jesse project strategies/ directory, then run the TDD script from your Jesse project root:

```bash
python triple_barrier.py
```

This will:
1. Generate synthetic candles and collect labeled samples (gather phase).
2. Train a Random Forest and save model.pkl, scaler.pkl, and feature_importance.pkl into strategies/TripleBarrierTest/.
3. Run a deploy-mode backtest to confirm the pipeline produces actual trades.

Once you have verified the pipeline on synthetic data, swap out the synthetic candles for real ones and adapt the strategy to your own setup.

---

## License

MIT

triple-barrier

triple-barrier 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

triple-barrier

triple-barrier 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README