ML Models in StrategyQuant

Hi everyone,

I’m currently exploring different ways to integrate Machine Learning models (Python-based) directly into the StrategyQuant X workflow, specifically as Indicators to be used during the Building/Generation phase. I’m curious if anyone here has successfully implemented ML models that work bar-by-bar during the backtest for strategy generation.

I’m particularly interested in your experience regarding two main challenges:

1. Backtest Speed: Calling an external Python process (<code class=”whitespace-pre-wrap”>ProcessBuilder) on every bar seems to be a major bottleneck for the Builder. If you are doing this, are you using a local API/Server (like FastAPI/Flask) to reduce latency, or have you moved to native Java libraries (like Smile or Weka)?
2. MQL4/MQL5 Export: For those using complex ML indicators (like Hidden Markov Models or Neural Nets), how are you handling the export to MetaTrader? Are you manually replicating the logic in MQL, using DLL bridges, or perhaps exporting models to ONNX?

I’m currently considering a workflow involving Java Snippets + Custom MQL4 Indicators + TPL translation templates, but I’d love to hear if there’s a more efficient or “standard” way the community is handling this.

Thank you