Transforming Q-learning to SARSA Algorithm

Is it possible to transform this Q-learning approach to SARSA approach? I am not sure how to modify the act and relay from the DQNAgent class to change the approach.

import random
import gym
import numpy as np
from collections import deque
from keras.models import Sequential
from keras.layers import Dense
from keras.activations import softmax
from keras.optimizers import Adam
import os
import time
import numpy as np

def softmax(x, temperature=0.025): 
    """Compute softmax values for each sets of scores in x."""
    x = (x - np.expand_dims(np.max(x, 1), 1))
    x = x/temperature    
    e_x = np.exp(x)
    return e_x / (np.expand_dims(e_x.sum(1), -1) + 1e-5)

## DQNAgent
class DQNAgent:
  def __init__(self, state_size, action_size):
    self.state_size = state_size
    self.action_size = action_size
    self.memory = deque(maxlen=20000)
    self.gamma = 0.95       # discount rate
    self.epsilon = 1.0      # exploration rate
    self.epsilon_min = 0.01
    self.epsilon_decay = 0.995
    self.learning_rate = 0.001
    self.model = self._build_model()

  def _build_model(self):
    # Neural Net for Deep-Q learning Model
    model = Sequential()
    model.add(Dense(24, input_dim=self.state_size, activation='relu'))
    model.add(Dense(48, activation='relu'))
    model.add(Dense(self.action_size, activation='linear'))
    return model

  def remember(self, state, action, reward, next_state, done):
    self.memory.append((state, action, reward, next_state, done))

  def act(self, state):
    ## We implement the epsilon-greedy policy
    if np.random.rand() <= self.epsilon:
        return random.randrange(self.action_size)
    act_values = self.model.predict(state)
    return np.argmax(act_values[0]) # returns action

  def exploit(self, state): # When we test the agent we dont want it to explore anymore, but to exploit what it has learnt
    act_values = self.model.predict(state)
    return np.argmax(act_values[0]) 

  def replay(self, batch_size):
    minibatch = random.sample(self.memory, batch_size)
    state_b = np.squeeze(np.array(list(map(lambda x: x[0], minibatch))))
    action_b = np.squeeze(np.array(list(map(lambda x: x[1], minibatch))))
    reward_b = np.squeeze(np.array(list(map(lambda x: x[2], minibatch))))
    next_state_b = np.squeeze(np.array(list(map(lambda x: x[3], minibatch))))
    done_b = np.squeeze(np.array(list(map(lambda x: x[4], minibatch))))

    ### Q-learning (Modify into SARSA)
    target = (reward_b + self.gamma * np.amax(self.model.predict(next_state_b), 1))  # for not SARSA
    target[done_b==1] = reward_b[done_b==1]
    target_f = self.model.predict(state_b)

    for k in range(target_f.shape[0]):
      target_f[k][action_b[k]] = target[k]
    self.model.train_on_batch(state_b, target_f)
    if self.epsilon > self.epsilon_min:
        self.epsilon *= self.epsilon_decay

  def load(self, name):
  def save(self, name):

This topic was automatically closed 182 days after the last reply. New replies are no longer allowed.