BTX Neuro EA
- Experts
-
Wahyu Bintoro
joint t.me/Boytrafx_Official - 버전: 3.6
Product name:
BTX Neuro GPT
Short description:
Self-training GPT neural network EA for XAUUSD H1. Retrains itself every month on fresh data and checks its own out-of-sample performance before trading at full strength.
Description:
BTX Neuro GPT is a self-learning Expert Advisor built on a GPT-style attention neural network, written entirely in MQL5. It does not rely on a fixed set of rules. It trains its own model directly inside MetaTrader 5, retrains it automatically on new market data, and measures how well the model performs on recent data it has never seen before deciding how actively to trade.
How it works
- Neural network: Batch normalization → two convolution layers → GPT attention block (8 heads, 60-bar memory) → two dense layers. The network estimates the probability that a BUY or a SELL will reach its take profit before its stop loss.
- 9 inputs per bar: RSI, MACD (main, signal, histogram), candle body relative to ATR, volatility regime (ATR14/ATR96), distance to EMA200 relative to ATR, and time of day.
- Realistic training labels: every historical bar is labeled by simulating a real trade with the same TP, SL, maximum holding time, and the broker spread.
- Walk-forward retraining: the model retrains automatically about once a month, continuing from its previous knowledge.
- Out-of-sample quality gate: after each retrain, the EA simulates trading on the most recent data that was not used for training or model selection.
- If the model beats both zero and random entries, the EA trades in FULL mode (signals with probability ≥ 55%).
- Otherwise it switches to CONSERVATIVE mode and only takes high-confidence signals (≥ 65%).
Trade management
- One position per direction at most, with server-side SL and TP based on ATR
- Automatic time exit after the maximum holding period
- No new entries when a trade could stay open over the weekend close
- No grid, no martingale, no averaging, no hedging tricks
- Optional automatic position sizing by % risk of equity
Chart dashboard
Shows model status, training samples, gate statistics (hit rate, expectancy, comparison against random entries), current BUY/SELL probabilities, performance of this EA's magic number, and the retrain history.
Recommendations
- Symbol: XAUUSD. Timeframe: H1 (default; the EA reads H1 data on any chart timeframe).
- History: at least 4 months of H1 data; 16 months recommended.
- Leave all inputs at their defaults. They come from walk-forward research and do not need optimization.
- Risk: with fixed 0.01 lot, a typical stop loss on XAUUSD H1 is around 50 USD. For roughly 1% risk per trade, use about 5,000 USD per 0.01 lot, or set RiskPct = 1 on a cent account for small balances.
- On first start the EA trains its model, which takes about 1–2 minutes. This is normal.
- Test on a demo account before using real funds.
Input parameters
| Input | Default | Description |
|---|---|---|
| InpTF | H1 | Analysis timeframe |
| InpHorizon | 48 | Maximum holding time (bars) |
| InpTrainBars | 0 | Training history bars (0 = AUTO, 8000) |
| InpRetrainBars | 0 | Retrain interval (0 = AUTO, about 1 month) |
| InpMinProb | 0.55 | Minimum win probability for an entry |
| InpTP_ATR / InpSL_ATR | 2.0 / 2.0 | Take profit / stop loss in ATR multiples |
| InpLot | 0.01 | Fixed lot (used when RiskPct = 0) |
| InpMagic | 26101502 | Magic number |
| InpRiskPct | 0 | Risk per trade in % of equity (0 = fixed lot) |
Risk disclaimer
Trading leveraged products involves significant risk and may not be suitable for all investors. Past performance, including backtests, does not guarantee future results. Neural network models can lose effectiveness when market conditions change. Use only capital you can afford to lose.
Keywords:
neural network, GPT, AI, machine learning, XAUUSD, gold, self-learning, walk-forward, attention, transformer
What's new (v3.06):
Initial release. Hybrid quality gate (full and conservative modes), automatic walk-forward settings, risk-based position sizing, weekend protection, and an English chart dashboard.
