I'm trying to wrap my head around of how machine learning stuff in Zorro is supposed to work.

Using WFO, bunch of ml models gets generated.

When testing - all models are clawled through and report looks reasonable.

Zorro help file advises to set `SelectWFO = -1` (<--- last machine learning model) for live trading.

When I do that and plot advised long/short signals (either test or live) - it shows a single impulse and then (!) seems to flatline.




I'm not entirely sure plotting is accurate as I've seen test reports plotting wrongly when compared to actual entries/exits (trying to test such hypothesis right now).

But if it is - what am I doing wrong? What's the supposed workflow when dealing with machine learning models?

nevermind. I think it's faulty custom tensor extraction function

Why don't you share your code, so we could all learn...