search

matinaghaei / Portfolio-Management-ActorCriticRL

Portfolio management using Actor-Critic Deep Reinforcement Learning algorithms including A2C, DDPG, and PPO
MIT License
35 stars 10 forks source link

DDPG - Noise added during test phase #4

Closed julienlambert42 closed 4 months ago

julienlambert42 commented 1 year ago

Hello again,

I see that noise is added to predicted action (mu) even during test/eval phase. As noise is used for exploration, shouldn't it be used only during training phase ?

Thanks

matinaghaei commented 1 year ago

It’s because exploration is needed for training RL agents. I don't remember the exact implementation, but I probably should have used a smaller noise for the testing phase.