Dynamic Learning for Post-Training Models in MQL5 — Beyond Frozen ONNX Models?

[Farhad Soltani]

Hello MQL5 community,

I’m working with post-training models (e.g., ONNX format) integrated into my trading system, but I am facing a limitation: the models are essentially frozen after training and deployment, and I am unable to update or adapt them dynamically based on new market data.

My question is:

How can we implement dynamic learning or continuous model adaptation in an MQL5 environment, especially for models that are traditionally static like ONNX?

More specifically:

• Are there recommended approaches or frameworks to support incremental or online learning within MQL5?
• Can we integrate external dynamic learning frameworks (e.g., Python-based) with MQL5 in a seamless way to update models on-the-fly?
• Are there best practices to combine static pre-trained models with adaptive modules to achieve dynamic behavior in trading?

Any code examples, links, or guidance to implement “dynamic learning” with post-training models in MQL5 would be very appreciated.

Thanks in advance!