dqn.py

#Snake: Deep Convolutional Q-Learning - Experience Replay Memory file

#Initializing the Experience Replay Memory
import numpy as np

class Dqn():
     
     def __init__(self, maxMemory, discount):
          self.maxMemory = maxMemory
          self.discount = discount
          self.memory = list()
          
     #Remembering new experience
     def remember(self, transition, gameOver):
          self.memory.append([transition, gameOver])
          if len(self.memory) > self.maxMemory:
               del self.memory[0]
     
     #Getting batches of inputs and targets
     def getBatch(self, model, batchSize):
          lenMemory = len(self.memory)
          numOutputs = model.output_shape[-1]
          
          #Initializing the inputs and targets
          inputs = np.zeros((min(batchSize, lenMemory), self.memory[0][0][0].shape[1], self.memory[0][0][0].shape[2], self.memory[0][0][0].shape[3]))
          targets = np.zeros((min(batchSize, lenMemory), numOutputs))
          
          #Extracting transitions from random experiences 
          for i, inx in enumerate(np.random.randint(0, lenMemory, size = min(batchSize, lenMemory))):
               currentState, action, reward, nextState = self.memory[inx][0]
               gameOver = self.memory[inx][1]
               
               #Updating inputs and targets
               inputs[i] = currentState
               targets[i] = model.predict(currentState)[0]
               if gameOver:
                    targets[i][action] = reward
               else:
                    targets[i][action] = reward + self.discount * np.max(model.predict(nextState)[0])
          
          return inputs, targets