How to improve take-profit and stop-loss settings and order management ?

[eromoe]

I trained a minute level model to predict 15 minutes pct return with accuracy 0.7 (cv result from out of time testing)
By using sample strategy MLWalkForwardStrategy , though with some tweaking, backtest always got negative final return.

I thought the problem came form

• take-profit and stop-loss settings
• How long to hold the asset

Do you have any guidelines? very appreciate :)