Reward function

[Nicolas99-9]

I would like to know how is computed the reward function ? Is it based on the open/low price ? Also, what is the meaning of the action "close" ?

[Kismuz]

@Nicolas99-9 ,

what is the meaning of the action "close" ?

= issue a [set of] orders necessary to close entire position opened at time present. Refer to https://www.backtrader.com/docu/order.html for details.

how is computed the reward function ?

In it's base form it is log utility of current to initial portfolio value ratio. See: https://kismuz.github.io/btgym/btgym.strategy.html#btgym.strategy.base.BTgymBaseStrategy.get_reward

In it's more advanced and real-applicable form it is a weighted sum of:

• normalized single-trade realized profit/loss;

... and potential-based reward shaping functions:

• potential FI_1 is current normalized unrealized profit/loss;
• potential FI_2 is current normalized broker value.

See: https://kismuz.github.io/btgym/btgym.research.html#btgym.research.strategy_4.DevStrat_4_6.get_reward and source code for details.

[Nicolas99-9]

Thank you very much