IPortfolioReward

Interfaces · AiDotNet.Finance.Trading.Rewards

A pluggable portfolio reward function for reinforcement-learning trading agents. Making the objective swappable is what lets the same environment run the reward as an EXPERIMENT — total return vs a risk-adjusted (Sharpe) objective vs a drawdown-penalized one — and rank them on an untouched holdout. Stateful rewards (e.g. the online Sharpe) must reset their running statistics in Reset.

Known implementations: DifferentialSharpeReward, TotalReturnReward

Methods

Method Summary
double Reward(PortfolioRewardContext context) Reward for one environment step given its context.
void Reset() Clears any running statistics at the start of an episode.