def drawForLambdaZero():
montecarlo = MonteCarlo(100)
print('Training Monte Carlo')
montecarlo.train(500000)
print('Training of Monte Carlo Completed')
lambdaValue = 0
learningRate = []
learningRateIndex = []
sarsa = SARSA(100, lambdaValue)
print('Training SARSA and plotting graph')
for i in range(1000):
learningRateIndex.append(i)
sarsa.train(1)
squareMean = np.sum(np.square(sarsa.Q - montecarlo.Q)) / float(1000)
learningRate.append(squareMean)
fig = plt.figure("SARSAZERO")
surf = plt.plot(learningRateIndex, learningRate)
fig.savefig('lambdaZero.png')
plt.show()
def drawForAllLambdas():
montecarlo = MonteCarlo(100)
print('Training Monte Carlo')
montecarlo.train(500000)
print('Training of Monte Carlo Completed')
lambdas = [0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0]
squareMean = []
numberElements = montecarlo.Q.shape[0] * montecarlo.Q.shape[1] * 2
for lambdaValue in lambdas:
sarsa = SARSA(100, lambdaValue)
print('Training SARSA', lambdaValue)
sarsa.train(1000)
print('Training of SARSA Completed')
squareMeanCalc = np.sum(
np.square(sarsa.Q - montecarlo.Q)) / float(numberElements)
squareMean.append(squareMeanCalc)
fig = plt.figure("SARSA")
surf = plt.plot(lambdas[1:10], squareMean[1:10])
fig.savefig('lambdaALL.png')
plt.show()
"--- optimal policy for Q Learning agent using eligibility trace ---")
q_agent.get_optimal_policy()
q_agent.plot_policy()
print(
"--- plotting training for Q Learning agent using eligibility trace ---"
)
q_agent.plot(episode_steps)
# Performing SARSA with standard environment
sarsa_agent = SARSA(episodes=episodes,
lr=learning_rate,
discount=discount,
epsilon=epsilon,
king=False)
print("--- starting training for SARSA agent --- ")
episode_steps = sarsa_agent.train()
print("--- optimal policy for SARSA agent ---")
sarsa_agent.get_optimal_policy()
sarsa_agent.plot_policy()
print("--- plotting training for SARSA agent ---")
sarsa_agent.plot(episode_steps)
# Performing SARSA with standard environment using eligibility trace
sarsa_agent = SARSA(episodes=episodes,
lr=learning_rate,
discount=discount,
epsilon=epsilon,
king=False,
_lambda=0.9)
print("--- starting training for SARSA agent using eligibility trace --- ")
episode_steps = sarsa_agent.train()
