Larry Williams XGBoost Onnx - expert for MetaTrader 5

Backtest XAUUSD Time Frame H4 Date Range 24/01/2020 - 03/12/2025

📋 User Manual: Larry Williams AI-Filtered EA

This Expert Advisor (EA) combines the classical Larry Williams Outside Bar strategy with an Artificial Intelligence (ONNX) filter. It uses mechanical price action to find setups and AI to predict the probability of a successful trade.

1. File Preparation (Crucial)

For the EA to initialize correctly, you must place your pre-trained machine learning model in the correct directory:

• Filename: larry_model.onnx (or the name specified in the inputs).

• Path: MQL5 > Files > larry_model.onnx

• Requirement: The EA will fail to start ( INIT_FAILED ) if the file is missing from this folder.

2. Input Parameters

3. Trading Logic & Strategy

Phase 1: Mechanical Detection

At the opening of every new bar, the EA checks for an Outside Bar (the current candle's High is higher than the previous, and the Low is lower than the previous).

• Bullish Signal: Price closes above the previous bar's High.

• Bearish Signal: Price closes below the previous bar's Low.

Phase 2: AI Validation

If an Outside Bar is detected, the EA extracts 10 Data Features (Body size, Relative Range, ATR, Volume change, Day of the week, Hour, etc.) and sends them to the larry_model.onnx model.

• The EA executes a BUY if the AI probability for Class 1 (Buy) > InpThreshold .

• The EA executes a SELL if the AI probability for Class 2 (Sell) > InpThreshold .

Phase 3: Trade Management

• Stop Loss (SL): Placed at the Low (for Buy) or High (for Sell) of the signal candle.

• Take Profit (TP): Calculated automatically based on the InpRR ratio.

• Frequency: The EA only allows one open position at a time.

4. Technical Requirements for the ONNX Model

If you are training the model in Python (Scikit-Learn, PyTorch, etc.), ensure the output matches the EA requirements:

1. Input Shape: {1, 10} (10 features).

2. Output Node 0: Predicted Label (Long).

3. Output Node 1: Probabilities (Float array of 3 classes: [Neutral, Buy, Sell]).

4. Feature Order: The data must be fed in the exact order defined in the CalculateFeatures function (Body Size, Rel Range, Bull/Bear flag, ATR, Rel ATR, Day, Hour, Vol Change, Prev Direction).

5. How to Deploy & Training self

1. unzip larry_william.zip

2. run command pip install -r requirements.txt

3. open metatrader 5 first

4. run python download_csv_metatrader5.py

5. run python train_larry_williams.py

6. run python convert_onnx_larry.py