| Algorithm 2 Dynamic Learning Rate Adjustment Algorithm. |
|
Require: Initial value, learning rate decay coefficient for the moving average of squared gradients , small constant Ensure: Optimized value 1: Initialize randomly or with prior knowledge 2: Initialize to a small positive value 3: for each episode do 4: for each step in the episode do 5: Choose an action using the combined and Boltzmann strategy 6: Take action observe the next state and the reward 7: Calculate the TD error 8: Update using the decay factor 9: Update using the RMSprop update rule 10: Move to the next state 11: end for 12: end for 13: return |