guides
Training Parameters
Step-by-Step Guide
This guide will walk you through the process of training your algorithm in Spectral Alpha, optimizing its parameters for robust performance.
Preparing Your Model for Training
Before initiating the training process, ensure your model is properly configured.
Setting Parameters
- Sensible Ranges: Make sure your parameters have meaningful minimum and maximum values. These values define the search space that the optimization engine will explore.
- Parameter Impact: Choose parameters that have a significant impact on your algorithm's trading behavior. This ensures that the optimization process leads to meaningful improvements in performance.
Selecting the Objective
- Default Objective: The default objective of finding plateaus in the profit ratio is an excellent starting point. This objective prioritizes stable and robust performance over maximizing a single metric.
- Custom Objectives (Optional): You can explore other objective metrics or configure the engine to search for maximum/minimum values if desired.
Training Your Algorithm
Spectral Alpha simplifies parameter training, making it as easy as clicking a button.
- Click "Optimize": Initiate the optimization process. The engine will now systematically explore the parameter space, searching for optimal values based on your chosen objective.
- Monitor Progress: You can track the progress of the optimization run and view intermediate results.
Analyzing Results
- View Performance: Once the optimization is complete, examine the performance chart. This shows how the algorithm performed on the historical data with the selected parameters. Remember that the exact performance during training is less important than the overall profitability and stability of the strategy.
- Explore Parameter Space: Analyze the training space for each parameter. The chosen plateau, representing a region of robust performance, is highlighted in green.
- Load Optimized Parameters: Hold the "Load" button to update the parameter values in the Model Builder with the optimized values.
Testing Your Trained Algorithm
- Run a Backtest: After loading the optimized parameters, run another backtest to observe how the algorithm trades with the new settings. Pay attention to the trading behavior, the distribution of trades, and the overall performance metrics.
Understanding the Limitations of Training
While training is a crucial step in algorithm development, it's important to remember that it's not a guarantee of future performance.
- Historical Data: Training is based on historical data, which may not perfectly reflect future market conditions.
- Overfitting Risk: Even with robust optimization techniques, there's always a risk of overfitting to the training data.
Walk Forward Analysis:
To gain a more realistic assessment of your algorithm's potential, conduct a walk forward analysis. This simulates real-world trading by testing the algorithm on unseen data, providing a more reliable indication of its likely performance in live trading.
