I am trying to write an reinforcement based based trading system and while trying to do that, the only way that I can do and I found is rewarding the model for each action but it actually performs bad results and couldn’t add the all parameters that I want. F.E I want to add these parameters but for a “acting one by one” model, I couldn’t add these parameters for reward because all of these parameters could be returned after all the long short hold classification finishes

if roi_percent < 30:
    f = 11.8
elif roi_percent < 70:
    f = 9.5
elif roi_percent < 140:
    f = 7.6
elif roi_percent < 250:
    f = 6.4
elif roi_percent < 340:
    f = 5
elif roi_percent < 600:
    f = 3.6
elif roi_percent < 1000:
    f = 2.46
elif roi_percent < 1600:
    f = 1.34
else:
    f = 0.7

sayi = entrynumber

if sayi == 20:
    sayi = 70
elif sayi < 10:
    sayi = 30 - (10*(10-sayi))
elif sayi < 20:
    sayi = 70 - (4*(20-sayi))
else:
    sayi = 70 - (1.6*(sayi - 20))

reward = (roi_percent * f) + (win_loss_ratio * 45) - ((self.max_drawdown * 444) / (93 - self.max_drawdown)) + (sharpe_ratio * 64) + sayi

And also we can create reinforcement learning models via DQN, DDQN, PPO, A2C models but I wonder how we can do this with such models like GRU, LSTM, CNN etc.