FeedforwardPolicyAgent<T>

Models & Types · AiDotNet.Finance.Trading.Agents

A MEMORYLESS (feed-forward MLP) continuous policy for the portfolio-manager harness — the controlled counterpart to RecurrentPolicyAgent. It uses the SAME on-policy REINFORCE-with-baseline training, exploration σ, and objective, differing ONLY in that the policy is a plain MLP over the current observation with no hidden state carried across steps. That makes recurrent-vs-MLP a clean controlled experiment: run both through PortfolioExperimentRunner / study and rank on the untouched holdout — any gap is attributable to memory, not the algorithm.

Implements: IPortfolioAgent<T>

Constructors

Constructor Summary
FeedforwardPolicyAgent(int stateDim, int actionDim, int hidden, double explorationSigma, double learningRate, double gamma, int seed)

Methods

Method Summary
T Train()
Vector<T> SelectAction(Vector<T> state, bool explore)
void ResetEpisode()
void StoreExperience(Vector<T> state, Vector<T> action, T reward, Vector<T> nextState, bool done)